LIVARSKI VESTNIK Izdajatelj / Publisher: Društvo livarjev Slovenije Lepi pot 6, P.P. 424, SI-1001 Ljubljana Tel.: + 386 1 252 24 88 E-mail: drustvo.livarjev@siol.net Spletna stran: www.drustvo-livarjev.si Glavni in odgovorni urednik / Chief and responsible editor: prof. dr. Alojz Križman E-mail: probatus@triera.net Tehnicno urejanje / Technical editoring: mag. Mirjam Jan-Blažic Uredniški odbor / Editorial board: prof. dr. Alojz Križman, Univerza v Mariboru prof. dr. Primož Mrvar, Univerza v Ljubljani prof. dr. Jožef Medved, Univerza v Ljubljani prof. dr. Rebeka Rudolf, Univerza v Mariboru prof. dr. Andreas Buhrig-Polaczek, Giesserei Institut RWTH Aachen prof. dr. Peter Schumacher, Montanuniversität Leoben prof. dr. Rüdiger Bähr, Otto-von Güricke-Universität Magdeburg prof. dr. Reinhard Döpp, TU Clausthal prof. dr. Jerzy Józef Sobczak, Foundry Research Institute, Krakow prof. dr. Jaromir Roucka, Institut Brno prof. dr. Branko Bauer, Univerza v Zagrebu Prevod v angleški jezik / Translation into English: Marvelingua, Aljaž Senicar s.p. Lektorji / Lectors: Angleški jezik / English: Yvonne Rosteck, Düsseldorf Slovenski jezik / Slovene: Marvelingua, Aljaž Senicar s.p. Tisk / Print: Fleks d.o.o. Naklada / Circulation: 4 številke na leto / issues per year 800 izvodov / copies Letna narocnina: 35 EUR z DDV Year subscription: 35 EUR (included PP) Dano v tisk: marec 2022 INSTRO d.o.o. Sedež: Preglov trg 11, 1000 Ljubljana Pisarna: Stegne 7, 1000 Ljubljana Direktor: Jovan JOVANOVIC Tel.: +386 (4)0 243 755 e-mail: info@instro.si spletna stran: www.instro.si VSEBINA / CONTENTS Stran / Page: Baitiang, Chinnadit: Napovedovanje mehanskih lastnosti litoželeznega dela in analiza napak med litjem z uporabo proizvodnih podatkov iz avtomatske livarne s pešcenimi formami / Prediction of Mechanical Properties of Cast Iron Part and Analysis of Casting Defects Using Production Data from an Automatic Sand Moulding Foundry 2 Brodarac, Zovko: Numericne simulacije za optimizacijo tankostenskega ulitka EN-GJL-200 / Numerical Simulation In Optimization Of Thin-Walled EN-GJL-200 Casting 29 Fercec, J.: Karakterizacija napak v ulitkih iz aluminija, proizvedenih z nagibnim gravitacijskim litjem, v povezavi s procesnimi parametri / Characterization of Defects in Aluminum Castings Produced by Tilt Gravity Casting in Relation to Process Parameters 43 Kirbiš, Peter: Kontinuirno litje visokoogljicnega nanostrukturnega bainitnega jekla / Continuous Casting of High Carbon Nanostructured Bainitic Steel 53 AKTUALNO / CURRENT Pregled svetovne livarske proizvodnje v letu 2020 60 Seje organov Društva livarjev Slovenije 62 STEM D.O.O. bogatejši za nove proizvodne prostore 65 62. IFC Portorož 2022 66 Livarski vestnik je vpisan v razvid medijev Ministrstva za kulturo pod zaporedno številko 588 Baitiang, Chinnadit1; Krüger, Mathias2; Piesker, Sven2; H. Williams-Boock, Bernd2; Weiß, Konrad1; Volk, Wolfram3 1 RWP. Gesellschaft beratender Ingenieure für Berechnung und rechnergestützte Simulation m.b.H., Nemcija / Germany 2 Ortrander Eisenhütte GmbH, Nemcija / Germany 3 Katedra za preoblikovanje in litje kovin, Tehnicna univerza v Münchnu, Nemcija / Chair of Metal Forming and Casting, Technical University of Munich, Germany Napovedovanje mehanskih lastnosti litoželeznega dela in analiza napak med litjem z uporabo proizvodnih podatkov iz avtomatske livarne s pešcenimi formami Prediction of Mechanical Properties of Cast Iron Part and Analysis of Casting Defects Using Production Data from an Automatic Sand Moulding Foundry Povzetek Zmanjšanje kolicine odpadkov je eden najpomembnejših ciljev livarn, ki vodi do višjega dobicka. Vendar pa je pri vecini livarn izmet mogoce zaznati šele na koncu proizvodne linije, kar je nekaj ur prepozno, da bi popravili proces. Za livarne bi bilo zaželeno omogociti zgodnejše odkrivanje izmeta in izvajati popravke procesa, ce bi bilo mogoce napovedati lastnosti in kakovost litih delov na podlagi vdelanih procesnih podatkov. V ta namen smo uvedli inovativne metode za napovedovanje kakovosti in lastnosti litoželeznega dela ter ugotavljanje vzrokov za izmet na podlagi proizvodnih podatkov (brez specificne sledljivosti dela) za livarne z avtomatskim postopkom izdelave pešcenih form (Disamatic). Najprej so bili izdelani modeli specificnih geometrij za napovedovanje mehanskih lastnosti (trdota in natezna trdnost) na podlagi kemicne sestave taline. Uporabili smo dve razlicni metodi modeliranja, tj. veckratno linearno regresijo (MLR) in umetno nevronsko mrežo (ANN). Modele smo ovrednotili s primerjavo izmerjenih in napovedanih podatkov s pomocjo grafikonov in povprecne absolutne napake. S primerjalno študijo vdelanih parametrov med dobrimi proizvodnimi serijami in izmetom smo dolocili relevantne parametre in njihove ustrezne delovne meje. Posledicno tako model MLR kot ANN kažeta dobro natancnost napovedovanja trdote (ki je pri modelu MLR malce boljša). Ugotovljeno je bilo, da so vzroki za vecino izmeta nestabilna vlaga in temperatura recikliranega peska, vsebnost gline in vode v materialu forme. Ustrezne delovne meje za te parametre smo dolocili z opazovanjem vrednosti parametrov v dobih proizvodnih serijah. Z našimi modeli in mejami lahko livarna z opazovanjem vdelanih parametrov predvidi lastnosti delov in zazna morebiten izmet vsaj eno uro vnaprej. Tako lahko zmanjša izmet, kar vodi v vecji prihranek stroškov v livarni. Kljucne besede: lito železo, Disamatic, mehanske lastnosti, veckratna linearna regresija, umetna nevronska mreža, napake med litjem Abstract Scrap reduction is one of the most important goals for foundries, leading to higher profit. However, for most foundries scrap parts can only be detected at the end of the production line, which is after hours too late to correct the process. It would be desirable for foundries, if cast part properties and quality can be forecasted based on inline process data, to allow earlier scrap detection and process correction. For this, we introduced innovative methods to predict the quality and properties of cast iron parts and find out the scrap causes based on production data (without part-specific traceability) from a foundry with an automatic sand moulding process (Disamatic). Firstly, geometry-specific models for the prediction of mechanical properties (hardness and tensile strength) based on the chemical composition of the melt were created. Two different modeling methods, multiple linear regression (MLR) and artificial neural network (ANN) were used. The models were evaluated by comparing measured and predicted data through charts and mean absolute error. Secondly, scrap relevant parameters and their proper working boundaries were determined using a comparative study of inline parameters between good and scrap production batches. As a result, both MLR- and ANN models show good prediction performance in predicting hardness (slightly better for the MLR). Unstable recycled sand moisture and temperature, clay- and water contents of moulding material were found to be the reasons for most scrap. Proper working boundaries for these parameters were determined by observing parameter values from good production batches. With our models and boundaries, the foundry can predict part properties and detect possible scrap at least one hour in advance by observing inline parameters. This contributes to scrap reduction and thus more cost savings for the foundry. Keywords: cast iron; disamatic; mechanical properties; multiple linear regression; artificial neural network; casting defects Uvod V dobi Industrije 4.0 so stroji in senzorji ustvarili veliko kolicino procesnih podatkov, ki proizvodnim podjetjem pomagajo spremljati in nadzorovati procese. Kljub razpoložljivim podatkom in tehnicnim izkušnjam pa se vecina današnjih livarn še vedno sooca z veliko kolicino zavrženih delov. Za livarne litega železa s postopkom avtomatskega formanja pešcenih form (Disamatic), na katerega se osredotoca postopek v tem prispevku, lahko pricakujemo stopnjo izmeta 5–8 % [1]. Pomembno je omeniti, da je zmanjšanje stopnje izmeta na primer z 8 % na 4 % enakovredno prihranku vec kot 2 milijona evrov v primeru livarne litega železa, ki proizvede 30 tisoc ton izdelkov [1]. Zato obstajajo možnosti za izboljšave, s katerimi bi dosegli višji dobicek zgolj z zmanjšanjem stopnje izmeta. Zmanjševanje izmeta ne pomeni samo velike gospodarske koristi, temvec tudi zmanjšanje vpliva na okolje. Težava je, da je mogoce lastnosti in kakovost ulitka prvic preveriti šele na koncu proizvodne linije, kar je nekaj ur prepozno za popravek postopka. To lahko privede do velike kolicine odpadnih delov. Livarne, 1 Introduction In the age of Industry 4.0 large amount of processes, data have been generated by machines and sensors, helping production companies to monitor and control their processes. Despite the available data and technical experiences, most of today’s foundries still face the undesired amount of rejected parts. For cast iron foundries with an automatic sand moulding process (Disamatic), which is the focused process in this article, a scrap rate of 5-8% can be expected [1]. It is important to note that reducing scrap rate, for instance, from 8% to 4% is equivalent to savings of over 2 million euros for a 30k-ton cast iron foundry [1]. Hence, there exists room for improvement to gain higher profits just by reducing the scrap rate. Reducing scrap means not only huge economic gain but also reduction of environmental impact. The problem is that the properties and quality of cast parts can only be inspected first at the end of the production line, which is after hours too late to correct the process. This could lead to a huge amount of scrap parts. Foundries, especially those with automatic processes, zlasti tiste z avtomatskim procesom, potrebujejo možnost napovedovanja kakovosti ulitka (ne da bi morali cakati in si ogledati proizvedene dele). To lahko dosežemo z izdelavo procesnih modelov, ki nam omogocajo oceno lastnosti in kakovosti ulitih delov. Trdota (HB) in natezna trdnost (Rm) sta dve tipicni zahtevi v povezavi z mehanskimi lastnostmi litoželeznih delov, ki jih je treba preveriti. Ti lastnosti sta v osnovi povezani z vec dejavniki, kot so kemicna sestava taline, hitrost strjevanja in debelina stene ulitka. Raziskava M.A. Essam in sod. (2021) je pokazala, da se trdota litega železa GJV zmanjša zaradi vecje debeline preseka in daljšega trajanja litja [2]. V prispevku so R. Kumar in sod. (2014) navedli, da dodatek bakra (mocan pospeševalec perlita) poveca tako trdoto kot natezno trdnost sive litine [3]. V raziskavi je G. Gumienny (2015) porocal, da se trdota litega železa GJV poveca ob višji vsebnosti kroma [4]. PA Heller in H. Jungbluth (1955) sta predstavila enacbe za oceno natezne trdnosti (Rm) za sivo litino na podlagi stopnje nasicenosti Sc in preskusnih palic razlicnih premerov, litih v pesek (za palico s premerom 30 mm: Rm = (102 - 82,5*Sc)*9,81 MPa) [5]. W. Patterson in sod. (1965) so predlagali ocena trdote sive litine na podlagi preskusne palice s premerom 30 mm (HB = 530 - 344*Sc za HB <186 in HB = 930 - 740*Sc za HB > 186) [6]. Deike in sod. (2000) so predlagali izracun teoreticne trdote po Brinellu na podlagi kemicne sestave taljenega železa (HB = 444 - 71,2*C - 13,9*Si + 21*Mn + 170*S) [7]. Te ugotovitve in formule temeljijo na podatkih iz preskusov vzorcnih ulitkov v litem stanju, vendar jih zaradi skrbi glede natancnosti ne bi smeli neposredno uporabiti v konkretnem proizvodnem procesu (kjer obstaja veliko razlicnih geometrij delov). Po našem mnenju potrebujejo livarne modele, specificne za geometrije, ki temeljijo na need a possibility to prognose casting quality (without having to wait and see the resulting parts). This can be achieved by creating process models, which allow us to estimate the properties and quality of cast parts. Hardness (HB) and tensile strength (Rm) are two typical mechanical property requirements of cast iron parts, to be inspected. These properties are basically associated with several factors such as the chemical composition of the melt, solidification rate, and wall thickness of the cast part. Astudy by M.A. Essam et al. (2021) showed that the hardness of GJV cast iron decreased by larger section thickness and longer pouring duration [2]. An article by R. Kumar et al. (2014) mentioned that the addition of Copper content (a strong pearlite promoter) increases both hardness and tensile strength of grey iron [3]. A research by G. Gumienny (2015) reported that the hardness of GJV cast iron increases with higher Chromium content [4]. P. A. Heller and H. Jungbluth (1955) presented the equations for estimation of the tensile strength (Rm) for grey iron based on the degree of saturation Sc and test rods with different diameters cast in the sand (for a rod with 30-mm diameter: Rm = (102 - 82.5*Sc)*9.81 MPa) [5]. W. Patterson et al. (1965) proposed estimation of hardness for grey iron based on test rod with 30-mm diameter (HB = 530 - 344*Sc for HB < 186 and HB = 930 - 740*Sc for HB > 186) [6]. Deike et al. (2000) proposed calculation of theoretical Brinell Hardness based on the chemical composition of the molten iron (HB = 444 - 71.2*C - 13.9*Si + 21*Mn + 170*S) [7]. These findings and formulas are based on experimental data from as-cast sample parts, however, should not be directly deployed in a specific production process (where there are many part geometries) due to accuracy concerns. In our opinion, podatkih industrijske proizvodnje in jih je mogoce uporabiti za natancno oceno lastnosti dela, ki se proizvaja. Poleg napovedovanja mehanskih lastnosti je pomembno tudi, da so livarne sposobne predvideti morebitne napake pri litju z opazovanjem vdelanih podatkov. V ta namen je treba vnaprej dolociti procesne parametre, ki so obcutljivi na napake pri litju, in njihove meje izmeta. Na podlagi prejšnjih prispevkov obstaja vec pristopov za analizo vzroka napak pri litju, kot so vzrocno-posledicni diagram (ustvarjen na podlagi izkušenj v livarstvu), nacrtovanje eksperimenta (DoE), metoda Taguchi in analiza variance (ANOVA) [8, 9]. Ocitno je, da so napake pri litju odvisne od vec dejavnikov, kot so slaba zasnova napajalnika in dovajalnega sistema, nestabilni vdelani parametri (npr. lastnosti oblikovanega materiala in taline, parametri litja), okvara stroja, cloveška napaka itd. Livarne potrebujejo tudi, na podlagi obstojecih meritev vdelanih parametrov, metodo za iskanje težav v trenutni proizvodnji kot tudi vidikov, ki jih je treba dodelati. Sledljivi procesni podatki (povezava med ulitkom in njegovimi vdelanimi procesnimi parametri) vsekakor pripo­morejo k pridobivanju zelo natancnih modelov procesov in iskanju vzrokov za izmet v procesu. Žal to ne velja vedno za vecino livarn, ki lijejo v pesek. Na podrocju sledljivosti livarskih procesov obstajajo prispevki, katerih cilj je uskladiti vdelane podatke s podatki o kakovosti dela [10, 11] skozi oznacbe na ulitku. NK Vedel-Smith in TA Lenau (2012) sta uvedla neposredno oznacevanje delov med procesom oblikovanja z uporabo rekonfigurabilnega orodja z zatici, ki je integrirano v modelno plošco [12]. Podobno je podjetje DISA leta 2019 predstavilo rešitev DISA TAG (Trace and Guidance), v model integriran sistem avtomatskega oznacevanja z vrtljivimi foundries need geometry-specific models which are based on industrial production data and can be used to accurately estimate the properties of the part being produced. Apart from the prediction of the mechanical properties, it is also important that foundries should be able to predict possible casting defects via observation of inline data. For this, process parameters, which are sensitive to casting defects, and their scrap boundaries should be determined in advance. Based on previous works, there exist several approaches to analyzing the cause of casting defects such as the Cause-Effect Diagram (created based on experience in a foundry), Design of Experiment (DoE), Taguchi method, and Analysis of variance (ANOVA) [8,9]. It is clear, that casting defects depend on several factors, such as the poor design of the riser and gating system, unstable inline parameters (e.g. moulding material and melt properties, pouring parameters), machine failure, human failure, etc. What foundries need in addition is, based on existing measurements of inline parameters, a method to locate where exactly the problems are in the current production and what to correct. A traceable process data (the connection between a cast part and its inline process parameters) definitely helps obtain highly accurate process models and find scrap causes in the process. Unfortunately, this is not always the case for most sand foundries. There have been articles in the area of foundry process traceability, aiming to match inline data with part quality data [10,11] via marking on the cast part. N. K. Vedel-Smith and T. A. Lenau (2012) introduced a direct part marking during the moulding process using reconfigurable pin-type tooling, which is integrated into the pattern plate [12]. Similarly, DISA TAG (Trace and Guidance), a pattern-integrated številcnicami [13]. Vendar pa se zaradi potrebe po dodatnih prizadevanjih in dodatnih stroškov številne livarne še vedno odlocajo, ali je uporaba koncepta smiselna. Menimo, da bi bilo treba uvesti tudi rešitev za livarne za modeliranje procesa brez sledljivosti posameznih delov. Da bi izpolnili vse zgoraj omenjene potrebe, smo najprej izdelali geometrijsko specificne modele za napovedovanje mehanskih lastnosti (trdota in natezna trdnost) litoželeznega dela, ki jih lahko livarna uporablja za napovedovanje in nadzor lastnosti delov, ki se proizvajajo na podlagi izmerjenih vdelanih podatkov. Vse podatke je zbrala in posredovala livarna litega železa z avtomatskim postopkom oblikovanja pešcenih form »Ortrander Eisenhütte GmbH« v Ortrandu v Nemciji. Ker ti podatki ne zagotavljajo sledljivosti posameznih delov, smo predlagali izvedbo povezave med vdelanimi podatki in podatki o kakovosti dela na podlagi datuma (edina skupna spremenljivka), kar pomeni, da so bili za modeliranjeuporabljeni dnevni povprecni podatki. Pri izdelavi modelov smo uporabili dva razlicna pristopa strojnega ucenja, ki temeljita na regresiji, in sicer veckratno linearno regresijo (MLR) in umetno nevronsko mrežo (ANN). Vse modele iz obeh metod smo ovrednotili s preskusnimi podatki in primerjali njihovo ucinkovitost napovedovanja. Prav tako smo predlagali primerjalno študijo vdelanih podatkov med proizvodno serijo z malo in veliko kolicino izmeta, da bi opredelili parametre, ki so pomembni za napake pri litju. Ustrezne delovne meje za te parametre smo dolocili z opazovanjem vrednosti parametrov v dobrih proizvodnih serijah. Te modele in delovne meje, specificne za geometrijo, je mogoce uporabiti neposredno v proizvodnji livarne in zagotoviti dosledno kakovost in lastnosti delov. automatic marking system with rotating dials, was introduced by DISAin 2019 [13]. However, due to additional effort and cost, it is still a decision to be made by many foundries whether to apply the concept. Our thought is that a solution for foundries to model a process without part-specific traceability should be introduced as well. To fulfil all the above-mentioned needs, firstly, we created the geometry-specific models for the prediction of mechanical properties (hardness and tensile strength) of cast iron parts, so foundry can use them to forecast and control the being-produced part properties based on measured inline data. All the data were collected and provided by the cast iron foundry with automatic sand moulding process ”Ortrander Eisenhütte GmbH” in Ortrand, Germany. As these data are without part-specific traceability, we proposed to realize the connection between inlinedata and part quality data via date (the only variable in common), which means day-average data were used for modelling. In the creation of the models, two different regression-based machine learning approaches, which are multiple linear regression (MLR) and artificial neural network (ANN), were used. All models from the two methods were evaluated with test data and their prediction performances were compared. Furthermore, we proposed a comparative study of inline data between low and high-scrap production batches to identify parameters that are relevant to casting defects. Propper working boundaries for these parameters were determined by observing parameter values from good production batches. These geometry-specific models and working boundaries can be applied directly to the production of the foundry to ensure consistent part quality and properties. Napoved mehanskih lastnosti Prva glavna naloga tega prispevka je bila ustvariti geometrijsko specificne modele za napovedovanje dveh pomembnih mehanskih lastnosti, tj. trdote (HB) in natezne trdnosti (Rm), na podlagi vdelanih parametrov. Kot primer je bil izbran rotacijsko simetricni ulitek (premer 250 mm, debelina stene: 15 mm, liti material: vermikularni grafit ali GJV, teža 3 kg), eden najpogosteje izdelanih delov v livarni. Seveda na HB in Rm vpliva vec dejavnikov, kot so kemicna sestava, hitrost strjevanja in debelina stene dela. Vendar pa je ob pogledu na podatke posameznega dela samo kemicna sestava taline izkazala variacijo vrednosti, medtem ko je bila hitrost litja konstantna za vsa litja. Konstantna je bila tudi debelina stene, saj je bilo treba meritve HB in Rm opraviti na istem mestu ulitka. Zato so bili v tem prispevku kot neodvisne spremenljivke modela obravnavani samo kemicni elementi. 2.1 Zbiranje podatkov in njihova priprava Pregled procesne postavite livarne je prikazan na Sliki 1. Taljeno železo smo pripravili v talilni peci in ga nato obdelali z magnezijem. Pešcene forme smo pripravili z avtomatsko napravo za formanje s casom cikla 15 sekund na formo in jih po liniji prepeljali do livne postaje. Za analizirani del GJV smo forme z votlino za dva dela lili s ciljno temperaturo litja 1380 °C. Polne forme smo nato transportirali naprej po hladilni liniji in približno 70 minut kasneje je potekel proces izmetavanja. Uliti deli so potovali naprej po transporterju še 10–15 minut in bili ocišceni z avtomatsko napravo za peskanje. Nazadnje so dokoncani deli prestali potrebne preglede, kot so vizualni pregled, kontrola dimenzij, mehanski 2 Prediction of Mechanical Properties The first main task of this work was to create geometry-specific models for predicting two important mechanical properties which are hardness (HB) and tensile strength (Rm) based on inline parameters. As an example, a rotationally symmetric cast part (diameter 250 mm, wall thickness: 15 mm, casted material: vermicular graphite iron or GJV, weight 3 kg), one of the most produced parts at the foundry, was chosen. Naturally, HB and Rm are influenced by several factors, such as chemical composition, solidification speed and wall thickness of the part. However, when looking at the data of a specific part, only the chemical composition of the melt showed variation in values, while the pouring speed was set as constant for all pours. The wall thickness was also considered as a constant, because HB and Rm measurements were to be carried out at the same spot on the cast part. Therefore, in this work, only the chemical elements were considered as the independent variables of the model. 2.1 Data collection and preparation As an overview, the process layout of the foundry is shown in Figure 1. The molten iron was prepared at a melting furnace and then treated with Magnesium. The sand moulds were prepared with an automatic sand moulding machine with a cycle time of 15 seconds per mould and transported along the line to the pouring station. For the analyzed GJV part, the moulds with a cavity for two parts were poured with a target pouring temperature of 1380 °C. The poured moulds were then transported further along the cooling lineand around 70 minutes later the shakeout process took place. The cast parts travelled further along the conveyer Slika 1. Pregled postopka litja v pesek z avtomatsko linijo za izdelavo pešcenih form Figure 1. The layout of the sand casting process with automatic sand moulding line preskus, analiza mikrostrukture in akusticni preskus. Zbrani so bili podatki o proizvodnji, kot so podatki o taljenju, formanju in kakovosti delov. Za izdelavo modelov za napovedovanje HB in Rm analiziranega dela so potrebni podatki o kemicni sestavi, HB in Rm. Kemicno sestavo taline smo izmerili dvakrat, najprej v talilni peci in drugic pri livni postaji (po obdelavi z magnezijem), in sicer s spektralno analizo s spektrometrom. Doloceni so bili masni odstotki (masni %) naslednjih kemicnih elementov: C, Si, Mn, S, P, Cr, Mo, Cu, Mg, Sn, Zn, Ni, Al, Co, Nb, Ti, V, W, Pb, As, Ce in N (skupaj 22 elementov). Za modeliranje je bila izbrana kemicna sestava, izmerjena pri livni postaji, saj je zagotavljala najbolj posodobljene podatke o stanju sestave z obdelavo z magnezijem pred litjem. Kemicna sestava na livni postaji je bila oznacena s sklicem na številko vzorca formanja in indeksa forme, kar je pripomoglo k sledenju (katera talina pripada kateremu vzorcu). V Preglednici 1 je prikazana izmerjena kemicna sestava for another 10-15 minutes and were cleaned with an automatic sandblasting machine. Finally, the parts came out and went through necessary inspections such as visual inspection, dimension control, mechanical test, microstructure analysis and acoustic test. The production data such as melting, moulding and part quality data, were collected. To create the models for predicting HB and Rm of the analyzed part, the chemical composition, HB and Rm data of the part are required. The chemical composition of the melt was measured twice, firstly at the melting furnace and secondly at the pouring station (after magnesium treatment), via spectral analysis using a spectrometer. The per cent weight contents (wt%) of the following chemical elements were determined: C, Si, Mn, S, P, Cr, Mo, Cu, Mg, Sn, Zn, Ni, Al, Co, Nb, Ti, V, W, Pb, As, Ce and N (in total 22 elements). The chemical composition measured at the pouring station was chosen for modelling because it provided the latest Preglednica 1. Kemicne sestave taline (v masnem odstotku) analiziranega dela GJV, izmerjene na livni postaji Table 1. Chemical Compositions of the Melt (in wt%) of the Analyzed GJV Part, Measured at Pouring Station Številka / Number Datum in cas / Date and time C Si Mn P S Cr Mo .. .. N 1 7. 1. 2020 08:17:02 3,53 2,56 0,24 0,054 0,004 0,02 0,005 .. .. 0,025 2 7. 1. 2020 09:14:39 3,54 2,55 0,20 0,048 0,004 0,01 0,005 .. .. 0,001 3 7. 1. 2020 10:01:05 3,59 2,64 0,19 0,045 0,006 0,01 0,005 .. .. 0,001 4 7. 1. 2020 10:13:07 3,59 2,62 0,18 0,047 0,007 0,02 0,005 .. .. 0,001 5 7. 1. 2020 10:27:23 3,60 2,66 0,18 0,047 0,007 0,02 0,005 .. .. 0,001 6 7. 1. 2020 10:42:02 3,60 2,64 0,18 0,043 0,006 0,02 0,005 .. .. 0,001 7 7. 1. 2020 10:52:04 3,60 2,61 0,18 0,049 0,007 0,02 0,005 .. .. 0,027 8 7. 1. 2020 11:10:14 3,59 2,69 0,17 0,043 0,007 0,01 0,005 .. .. 0,001 9 7. 1. 2020 11:29:49 3,59 2,66 0,17 0,047 0,006 0,02 0,005 .. .. 0,001 .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 466 8. 12. 2020 21:30:44 3,56 2,64 0,18 0,045 0,005 0,02 0,005 .. .. 0,001 467 8. 12. 2020 21:38:51 3,60 2,63 0,18 0,045 0,006 0,02 0,005 .. .. 0,002 analiziranega dela GJV, proizvedenega v letu 2020 (skupaj 467 meritev). Meritve so bile izvedene na pribl. 10–15 minut. Po postopku peskanja smo približno vsakih 30 minut med proizvodnjo vzeli kos ulitega dela za porušni preskus trdote po Brinellu. Vrednost HB je bila izmerjena z merilnikom trdote Hegewald & Peschke Universal 750 na tocki v delu, ki ga je dolocil narocnik livarne. Merjenje natezne trdnosti, ki je prav tako porušni preskus, je bilo izvedeno z merilnikom natezne trdnosti Hegewald & Peschke Inspekt 250, in sicer na vzorcu, izrezanim iz litega dela. Vrednosti Rm pa so bile pridobljene na približno 1–2 uri, in sicer zaradi zahtevnejše priprave rezanega vzorca. V Preglednici 2 so prikazane izmerjene mehanske lastnosti dela (skupaj 361 meritev HB in 114 meritev Rm). state of composition with magnesium treatment before pouring. Moreover, the chemical composition at the pouring station was recorded with reference to moulding pattern number and mould index, which helps trace (which melt belongs to which pattern). Table 1 shows the measured chemical composition of the analyzed GJV part produced in the year 2020 (in total 467 measurements). Each measurement was carried out approximately every 10-15 minutes. After sandblasting process, a piece of the cast part was picked up for the destructive Brinell hardness test approximately every 30 minutes during the production. HB value was carried out by a hardness tester, Hegewald & Peschke Universal 750, at a point in the part that was specified by the foundry´s customer. The measurement of the tensile strength, also a destructive test, was carried out by tensile tester, Hegewald & Peschke Inspekt 250, with a cut specimen from the cast part. Rm values were obtained Preglednica 2. Mehanske lastnosti analiziranega dela GJV Table 2. Mechanical Properties of the Analyzed GJV Part Številka / Number Podatki in cas / Data and time Trdota (HB) / Hardness (HB) Natezna trdnost (Rm) / Tensile strength (Rm) 1 7. 1. 2020 09:57:44 190 491 2 7. 1. 2020 10:20:34 190 - 3 7. 1. 2020 10:54:33 188 489 4 7. 1. 2020 11:26:36 191 - 5 7. 1. 2020 11:56:31 171 - 6 7. 1. 2020 12:21:15 175 - 7 7. 1. 2020 12:56:08 175 448 8 7. 1. 2020 13:29:06 179 - 9 7. 1. 2020 13:42:47 176 - .. .. .. .. .. .. .. .. 360 9. 12. 2020 06:48:44 181 - 361 9. 12. 2020 06:49:31 178 421 Meritve kemicne sestave na livni postaji so bile izvedene približno 80–100 minut pred meritvami HB in Rm. Te meritve so bile izvedene neodvisno in brez vseh metod sledljivosti. Posledicno je bilo zaradi negotovosti v casovnem okviru oteženo sklepanje, katera vrednost HB in Rm v Preglednici 2 pripada kateri vrednosti kemicne sestave v Preglednici 1. Potrebna je bila rešitev za pridobitev podatkovne matrike iz teh podatkov. Ce si ogledamo podatke, je edini podatek, ki je skupen tako v Preglednici 1 kot Preglednici 2, datum meritve. Zato smo se odlocili za rešitev z uskladitvijo obeh podatkov na podlagi datuma. Tako smo pridobili novo matriko dnevne povprecne vrednosti (Preglednica 3). Ta nova podatkovna matrika je bila nato uporabljena pri modeliranju. Kot je bilo predstavljeno, sta bili za izdelavo modelov uporabljeni dve metodi modeliranja, tj. veckratna linearna regresija (MLR) in umetna nevronska mreža (ANN). Ucinkovitost napovedi modelov na podlagi however approximately every 1-2 hours due to more effort in the preparation of the cut specimen. Table 2 presents the resulting measured mechanical properties of the part (in total 361 measurements of HB and 114 measurements of Rm). Measurements of the chemical composition at the pouring station were carried out approximately 80-100 minutes before those of HB and Rm. These measurements were conducted independently without any traceability methods. Consequently, due to uncertainty in the time frame, it was difficult to infer which value of HB and Rm in Table 2 belongs to which value of chemical composition in Table 1. Asolution to obtain the data matrix from these data was needed. When looking at the data, the only information that both Table 1 and Table 2 have in common is the date of measurement. Therefore, our solution was to synchronize both data via date. As a result, a new matrix of day-average value was obtained (Table 3). This new data matrix was then used for the modeling work. As introduced, two Preglednica 3. Dnevne povprecne vrednosti kemicne sestave in mehanskih lastnosti analiziranega dela GJV (matrika dnevnih povprecnih podatkov) Table 3. Day-Average Values of Chemical Composition and Mechanical Properties of the Analyzed GJV Part (Day-Average Data Matrix) Številka / Number Datum / Date C Si Mn P S Cr Mo .. .. HB Rm 1 7. 1. 2020 3,57 2,65 0,177 0,046 0,006 0,016 0,005 .. .. 176,33 447,23 2 22. 1. 2020 3,60 2,65 0,188 0,048 0,005 0,020 0,005 .. .. 178,95 436,33 3 29. 1. 2020 3,59 2,67 0,199 0,054 0,006 0,026 0,005 .. .. 182,45 419,43 4 4. 2. 2020 3,58 2,65 0,227 0,055 0,006 0,020 0,005 .. .. 184,15 433,14 5 5. 2. 2020 3,57 2,68 0,258 0,054 0,006 0,026 0,005 .. .. 183,85 434,50 6 12. 2. 2020 3,57 2,67 0,222 0,053 0,005 0,021 0,005 .. .. 182,17 419,83 7 18. 2. 2020 3,57 2,62 0,196 0,055 0,005 0,020 0,005 .. .. 184,27 436,33 8 4. 3. 2020 3,57 2,67 0,168 0,051 0,006 0,015 0,005 .. .. 183,25 430,00 9 11. 3. 2020 3,56 2,61 0,166 0,058 0,005 0,012 0,005 .. .. 181,80 451,00 .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. .. 21 11. 11. 2020 3,57 2,68 0,170 0,039 0,005 0,020 0,005 .. .. 178,00 409,40 22 8. 12. 2020 3,56 2,65 0,183 0,045 0,005 0,017 0,005 .. .. 175,38 415,10 obeh metod je bila ovrednotena in nato primerjana. 2.2 Veckratna linearna regresija (MLR) Modeli veckratne linearne regresije (MLR) temeljijo na linearnem razmerju med vec neodvisnimi spremenljivkami (X1, X2, X3,…, Xn) in odvisno spremenljivko (Y). Ko upoštevamo razmerje med spremenljivkami, se vzpostavi regresijska enacba: Y = ß + ß X + ß X + ß X +… + ß X (1), 0112233nn kjer so ß 0, ß 1, ß 2,…, ß n regresijski koeficienti, ki jih je treba oceniti. Pri uporabi metode MLR je treba upoštevati nekatera merila: • podatki spremenljivk morajo biti normalno porazdeljeni, • med neodvisnimi spremenljivkami ne sme biti mocnih korelacij, • število vzorcev podatkov mora biti vecje modeling methods which are multiple linear regression (MLR) and artificial neural network (ANN) were used to create the models. The prediction performance of the models from both methods was evaluated and compared afterwards. 2.2 Multiple linear regression (MLR) Multiple linear regression (MLR) models are based on the linear relationship between multiple independent variables (X1, X2, X3,…, Xn) and a dependent variable (Y). When considering the relationship between the variables, a regression equation is established: Y = ß + ß X + ß X + ß X +… + ß X (1), 0112233nn where ß 0, ß 1, ß 2,…, ß n are the regression coefficients to be estimated. There are some criteria to follow when using the MLR method, • The data of the variables should be od števila neodvisnih spremenljivk, • predpostavlja se linearna zveza med spremenljivkami. Model MLR se obicajno oceni z izracunom njegovega koeficienta determinacije R2, ki nakazuje odstotek variance odvisne spremenljivke, ki jo neodvisne spremenljivke pojasnjujejo skupaj (kako dobro model ustreza podatkom o usposabljanju). Vrednosti R2 se gibljejo med 0–100 % (višja kot je vrednost, boljše je prileganje), doloci pa se z naslednjo enacbo (enacba 2): (2), kjer je yi dejanska ali izmerjena vrednost Y pri indeksu i, y je napovedana vrednost Y, yŻ je povprecje Y in n je število podatkovnih tock. Vpliv kemicne sestave na HB in Rm lahko z metodo MLR opišemokvantitativno. Izbira neodvisnih spremenljivk za model je potekala skladno s prej omenjenimi merili. Preskus normalne porazdelitve na podatkih za vsak kemicni element je bil izveden s histogramom. Glede na rezultat preskusa vseh 22 kemicnih elementov ni bilo normalno porazdeljenih. Spremenljivke, ki niso bile normalno porazdeljene (Cr, Mo, Zn, Co, Nb, W, Pb, N), smo zavrgli. Na koncu so ostali elementi C, Si, Mn, P, S, Cu, Mg, Sn, Ni, Al, Ti, V, As, Ce (14 elementov). Zaradi poenostavitve regresijske analize smo še dodatno zmanjšali število neodvisnih spremenljivk. C, Si, Mn, P, Cu (5 glavnih elementov z najvišjo vsebnostjo teže), Mg (po obdelavi z magnezijem) in S zlitine (vsota ostalih 8 elementov S, Sn, Ni, Al, Ti, V, As, Ce) smo uporabili za neodvisne spremenljivke (skupaj 7 neodvisnih spremenljivk). Pridobili smo korelacijsko matriko teh 7 spremenljivk in med njimi ni bilo pomembne korelacije. normally distributed • There should be no strong correlation between independent variables • Number of data samples must be greater than the number of independent variables • Linear relationship between the variables is assumed An MLR model is usually evaluated by calculating its coefficient of determination R2, which indicates the percentage of the variance in the dependent variable that the independent variables explain collectively (how well the model fits the training data). The R2 values ranges between 0-100% (the higher the value, the better the goodness of fit), which is determined using the following equation (Eq. 2): (2), where yi is the actual or measured value of Y at index i, y is the predicted value of Y, yŻ is the average of Y and n is the number of data points. The impact of chemical composition on HB and Rm can be quantitatively described using the MLR method. The selection of independent variables for the model was in regard to the aforementioned criteria. Test of the normal distribution on the data for each chemical element was carried out by histogram. Based on the test result, not all 22 chemical elements were normally distributed. The variables which were not normally distributed (Cr, Mo, Zn, Co, Nb, W, Pb, N) were discarded. As a result, the remaining elements were C, Si, Mn, P, S, Cu, Mg, Sn, Ni, Al, Ti, V, As, Ce (14 elements). For simplicity of regression analysis, we further reduced the number of independent variables. C, Si, Mn, P, Cu (top 5 main elements with the highest percent weight content), Mg (after magnesium treatment) and S(the sum of the other 8 Alloys Ce povzamemo, smo za neodvisne spremenljivke izbrali C, Si, Mn, P, Cu, Mg in S, HB in Rm analiziranega dela zlitine GJV pa za odvisni spremenljivki. Po oceni regresijskih koeficientov smo pridobljene modele MLR zapisali v obliki regresijskih enacb, ki so prikazane v delu z rezultati in razpravo. Modele smo nato ovrednotili z drugim naborom podatkov (preskusnimi podatki), zbranimi v drugem proizvodnem obdobju. Obicajno se napovedana ucinkovitost modela oceni z izracunom kolicine odstopanja predvidenih vrednosti od dejanskih ali pa izmerjenih vrednosti. Eden preprostih in lahko razumljivih nacinov za merjenje napake ali izgube modela napovedi je izracun povprecne absolutne napake (MAE). V tem delu smo z MAE ocenili napake napovedovanja pridobljenih modelov. Formula za MAE je podana kot: (3) kjer je yi predvidena vrednost, y je dejanska ali izmerjena vrednost, n pa skupno število podatkovnih tock. Poleg metode MLR smo v tem prispevku predlagali tudi drugo metodo modeliranja, imenovano umetne nevronske mreže (ANN). Pregled in postopek modeliranja metode ANN sta predstavljena v naslednjem razdelku. 2.3 Umetna nevronska mreža (ANN) Umetna nevronska mreža, vrsta algoritmov za strojno ucenje, je izdelana po vzoru cloveških možganov. Podobno kot se nevroni v našem živcnem sistemu ucijo iz preteklih podatkov se ANN uci iz danih podatkov in zagotavlja odgovore v obliki klasifikacij ali napovedi. ANN se v raziskavah obširno uporablja za modeliranje kompleksnih odnosov. Izbirati je mogoce elements S, Sn, Ni, Al, Ti, V, As, Ce), were used as independent variables (in total seven independent variables). Acorrelation matrix of these 7 variables was obtained and there was no significant correlation among them. In summary, C, Si, Mn, P, Cu, Mg and Swere selected as independent Alloys variables and HB and Rm of the analyzed GJV part are dependent variables. After estimating the regression coefficients, the obtained MLR models are presented as regression equations, shown in the result and discussion part. The models were then evaluated with another set of data (test data), collected in another production period. Usually, the prediction performance of a model is evaluated by calculating the amount of deviation of the predicted values from the actual or measured values. One simple and easy-to-understand way to measure the error or loss of a prediction model is to calculate mean absolute error (MAE). In this work, MAE was used to assess the prediction errors of the obtained models. The formula for MAE is given as: (3) where yi is the predicted value, y is the actual or measured value and n is the total number of data points. Apart from the MLR method, this article also proposed another modeling method called artificial neural networks (ANN). The overview and modeling procedure of the ANN method is presented in the next section. 2.3 Artificial neural network (ANN) Artificial Neural Network is a type of machine learning algorithm, which is modeled after the human brain. Similar to how the med vec algoritmi ANN, in sicer glede na raziskovalne probleme, kot so klasifikacija, grozdenje in regresija. Predvidevanje trdote in natezne trdnosti litega dela, ki sta numericni spremenljivki, se tako obravnava kot regresijski problem, ki predvideva uporabo posebnega algoritma ANN za dolocitev numericnih vrednosti. Model nevronske mreže, uporabljen v tej študiji, je usmerjena mreža z algoritmom z vzvratnim postopkom ucenja. Vsak nevron priredi skalarni produkt med vhodnimi podatki (x) in utežmi (w), doda pristranskosti, uporabi aktivacijsko funkcijo in poda izhodne podatke. Slika 2 prikazuje, kako se izracunajo izhodni podatki, ko je podanih vec vhodnih podatkov za nevrone. Mreža je sestavljena iz vhodne plasti (tj. plasti, ki sprejemajo vhode na podlagi danih podatkov), enega ali vec skritih slojev (tj. slojev, ki delujejo na podlagi povratnega razširjanja za optimizacijo uteži vhodnih spremenljivk, da bi izboljšali natancnost napovedovanja) in izhodne plasti (tj. izhodne napovedi na podlagi podatkov iz vhodnih in skritih plasti). Uporabili smo aktivacijsko funkcijo rektificirane linearne enote (ReLU) za nevrone v skritih plasteh, saj jo je lažje uciti in pogosto zagotavlja boljšo natancnost [14]. Ker gre za regresijski problem, je aktivacijska funkcija za izhodno plast preprosto linearna, f(x) = x. neurons in our nervous system can learn from past data, the ANN can learn from the given data and provide responses in the form of classifications or predictions. ANN has been widely used in most researches to model a complex relationship. There are several ANN algorithms to choose from, depending on the research problems, such as classification, clustering, and regression. Predicting hardness and tensile strength of cast part, which are numerical variables, is thereby considered as a regression problem that needs a specific ANN algorithm for the determination of a numerical output. The neural network model used in this study is a feedforward network with the back-propagation learning algorithm. Each neuron performs a dot product between the inputs (x) and weights (w), adds biases, applies an activation function, and gives out the outputs. Figure 2 shows how an output is computed, given several inputs to the neurons. The network consists of an input layer (layers that take inputs based on given data), one or more hidden layers (layers that use backpropagation to optimize the weights of the input variables to improve the prediction performance), and an output layer (output of predictions based on the data from the input and the hidden layers). We used the rectified linear unit (ReLU) activation function for the neurons in the hidden layers, as it is easier to train and Slika 2. Algoritem usmerjene umetne nevronske mreže za problem regresije Figure 2. Algorithm of feedforward artificial neural network for regression problem Izbira števila nevronov za vkljucitev v vhodne in skrite plasti je pomemben dejavnik pri ucenju nevronske mreže. Podobno kot pri modelu MLR je bilo za vhodno plast privzeto upoštevanih 7 nevronov za teh 7 neodvisnih spremenljivk. Ker predstavlja vrednost HB ali Rm rezultat, ima izhodna plast privzeto 1 nevron. Število skritih plasti in nevronov je bilo doloceno po ucenju in preskušanju številnih modelov z razlicnimi konfiguracijami omrežja, zacenši od 1 skrite plasti z 1 nevronom. Model ANN je bil ucen z enakimi podatki o ucenju kot z metodo MLR. Podobno je bila za vrednotenje modelov ANN uporabljena tudi povprecna absolutna napaka (MAE). Rezultat je pokazal, da se vrednost MAE zmanjša, ko se poveca število skritih plasti ali nevronov. Model z 2 skritima slojema vedno privede do nižje MAE kot model z 1 skrito plastjo. Kljub temu se vrednost MAE ni dodatno zmanjšala pri uporabi 3 skritih plasti ali vec. Na podlagi modelov z 2 skritima slojema se vrednost MAE ni dodatno zmanjšala, saj je bilo število nevronov v skriti plasti enako 4. Isti koncept je bil uporabljen tudi za iskanje zadostnega števila ponovitev usposabljanja (niter). Ugotovili smo, da se vrednost MAE often achieves better performance [14]. As this is a regression problem, the activa­tion function for the output layer is simply linear, f(x) = x. Choosing the number of neurons to include in the input and hidden layers is an important consideration, when training a neural network. Similar to the MLR model, 7 neurons for those 7 independent variables were considered for the input layer by default. Given that a value of HB or Rm is an outcome, the output layer therefore has 1 neuron by default. The number of hidden layers and neurons was determined after training and testing many models with various network configurations, starting from 1 hidden layer with 1 neuron. The ANN model was trained with the same training data as done with the MLR method. Similarly, mean absolute error (MAE) was also used to evaluate the ANN models. The result showed that MAE decreases, as either number of hidden layers or neurons increases. A model with 2 hidden layers always results in lower MAE than a model with 1 hidden layer. Nevertheless, MAE did not further decrease when using 3 hidden layers or more. Based on the models Slika 3. Arhitektura umetne nevronske mreže z dvema skritima slojema za napovedovanje trdote na podlagi kemicne sestave taline Figure 3. Artificial neural network architecture with two hidden layers for prediction of hardness based on chemical composition of the melt zmanjšuje, ko se niter povecuje. Vrednost MAE se ni dodatno zmanjšala, ko je bila vrednost niter enaka 20000. Ce povzamemo: konfiguracija mreže z najnižjo vrednostjo MAE je bila vhodna plast s 7 nevroni, 2 skritima plastema s 4 nevroni na vsaki ter izhodna plast z enim nevronom (Slika 3). Ucinkovitost napovedovanja ustvarje­nega modela ANN smo nato v delu z rezultati in razpravo primerjali z rezultati modelov MLR. Analiza napak med litjem Dober ulitek pomeni ne samo doseganje ciljnih mehanskih lastnosti, temvec tudi njegovo stanje brez napak. Odkrivanje vzroka za izmet iz takšnega avtomatskega postopka je bil za livarno vedno izziv. V tem prispevku prikazujemo, kako je mogoce dolociti relevantne parametre za izmet in njihove ustrezne delovne meje z uporabo proizvodnih podatkov brez sledljivosti za posamezen del. Analizirali smo pet pogostih skupin izmeta v livarni (1. plinski with 2 hidden layers, MAE did not further decrease, as the number of neurons in a hidden layer reached 4. The same concept was also used to find a sufficient number of training iterations (niter). We found that MAE decreases, as niter increases. MAE did not further decrease, when niter reached 20000. In summary, the resulting network configuration with the lowest MAE value was an input layer with 7 neurons, 2 hidden layers with 4 neurons on each one, and an output layer with one neuron (Figure 3). The prediction performance of the created ANN model was then compared with that of the MLR models in the results and discussion part. 3 Analysis of Casting Defects A good cast part means not only achieving target mechanical properties but also defect-free conditions. Finding out the reason for scrap from such an automatic process has always been a challenge for the foundry. Here, we show how the Preglednica 4. Stopnja izmeta preiskovanih ulitkov iz razlicnih proizvodnih serij Table 4. Scrap Rate of The Investigated Cast Parts from Different Production Batches Številka skupine / Group number Skupina ostankov / Scrap group Preiskani ulitki / Investigated cast parts Proizvodna serija / Production batch Stopnja izmeta / Scrap rate Plinski mehurcek, 1A 16,6 % 1 vkljucek peska, poroznosti / Gas bubble, sand inclusion, porosities Infrastrukturna rešetka / Infrastructural grate 1B 0,45 % 2 Zvari v hladnem / Cold shuts Stranska plošca pecice / Oven side plate 2A 100 % 2B 5,1 % 3 Razpoke forme / Mould tears Vrata pecice / Oven door 3A 32 % 3B 0 % 4 Mehanska praska / Mechanical scratch Okvir pecice / Oven frame 4A 12,1 % 4B 2,25 % Pomarancasta 5A 23,46 % 5 površina / Orange-skin surface Infrastrukturna rešetka / Infrastructural grate 5B 0,82 % mehurcek, vkljucek peska, poroznosti, 2. zvari v hladnem, 3. razpoke forme, 4. mehanska praska, 5. pomarancasta površina). Analizna metoda je bila izvedena s primerjalno študijo vdelanih parametrov med proizvodnimi serijami z nizko in visoko stopnjo izmeta. Za analizo so bili izbrani ulitki z visoko in izrazito stopnjo izmeta (po skupinah) (Preglednica 4). Tukaj ugotavljamo, da prva skupina izmeta (plinski mehurcki, vkljucki peska, poroznosti) sodi med napake, ki jih lahko opazimo pri vizualnem pregledu v livarni in se vecinoma nanašajo na kakovost materiala forme. Za izvedbo primerjalne študije smo primerjali razpršenost vdelanih parametrov iz serije A(visoka stopnja izmeta) in serije B (nizka stopnja izmeta) s pomocjo grafikonov v delu z rezultati in razpravo. Tako smo lahko prepoznali razlike v vrednostih parametrov med serijami z dobro in neugodno stopnjo izmeta ter posledicno izpostavili parametre, ki jih povzrocajo. 4 Rezultati in razprava 4.1 Modeli za napovedovanje mehanskih lastnosti Pripravili smo modela MLR in ANN za napovedovanje mehanskih lastnosti (HB in Rm) analiziranega dela GJV na podlagi kemicne sestave (C, Si, Mn, P, Cu, Mg in S). Modeli MLR so prikazani kot zlitine naslednje linearne regresijske enacbe: HB = 249,77 - 34,03×C + 2,73×Si + 43,87×Mn + 278,12×P+ 116,26×Cu + 1176,35×Mg + 87,78×S, zlitine z R2 = 0,69 (4) Rm = 934,13 - 33,57×C -175,95×Si + 35,91×Mn - 78,85×P+ 648,56×Cu + 6554,34×Mg - 356,50×S, zlitine z R2 = 0,43 (5) scrap relevant parameters and their proper working boundaries can be determined using the production data without part-specific traceability. Five common groups of scrap at the foundry (1. Gas bubble, sand inclusion, porosities, 2. Cold shuts, 3. Mould tears, 4. Mechanical scratch, 5. Orange-skin surface) were analyzed. The analysis method was carried out using comparative study of inline parameters between low- and high-scrap production batches. Cast parts with a high and prominent scrap rate (by group) were chosen for the analysis (Table 4). Here we note that, the first group of scrap (gas bubble, sand inclusion, porosities) belongs to defects that can be seen by eyes during a visual inspection at the foundry and mostly relates to the quality of moulding material. To perform the comparative study, scatter of inline parameters from batch A (high scrap rate) and batch B (low scrap rate) were compared through charts in the results and discussion part. This allows us to recognize the differences of parameter values between good- and scrap batches and therefore to point out the causing parameters. 4 Results and Discussion 4.1 Mechanical property prediction models As a result, the MLR and ANN models for predicting mechanical properties (HB and Rm) of the analyzed GJV part based on the chemical composition (C, Si, Mn, P, Cu, Mg and S), were obtained. The MLR models Alloys are shown as the following linear regression equations: HB = 249,77 - 34,03×C + 2,73×Si + 43,87×Mn + 278,12×P+ 116,26×Cu + 1176,35×Mg + 87,78×S, Alloys with R2 = 0,69 (4) Model z R2 0,69 kaže, da je 69 % variacije HB mogoce pojasniti z variacijami C, Si, Mn, P, Cu, Mg in S. Vendar ima zlitine model MLR za napovedovanje Rm vrednost R2 0,43, kar pomeni nižjo primernost prileganja. Pridobljena sta bila tudi dva modela ANN za oceno HB in Rm. Ucinkovitost napovedovanja obeh modelov MLR in ANN je bila nato ovrednotena s preskusnimi podatki (podatki iz drugega proizvodnega obdobja) s pomocjo povprecne absolutne napake (MAE) in primerjalnih grafikonov (Slika 4). Za preskusne podatke je bilo na voljo 219 meritev kemicne sestave, 146 meritev HB in 44 meritev Rm. Tudi matrika povprecnih dnevnih podatkov teh preskusnih podatkov je bila pripravljena na enak nacin, kot smo pripravili podatke Rm = 934,13 - 33,57×C -175,95×Si + 35,91×Mn - 78,85×P+ 648,56×Cu + 6554,34×Mg - 356,50×S, Alloys with R2 = 0,43 (5) The model with R2 of 0.69 indicates that 69% of the variation in HB can be explained by variations in C, Si, Mn, P, Cu, Mg and S. However, the MLR model for Alloys predicting Rm has an R2 of 0.43, implying lower goodness of fit. Two ANN models for estimating HB and Rm were also obtained. The prediction performance of both MLR and ANN models was then evaluated with test data (data from another production period) via mean absolute error (MAE) and comparison charts (Figure 4). For the test data, there were 219 measurements of chemical composition, 146 measurements of HB, and 44 Slika 4. Ucinkovitost modelov veckratne linearne regresije (MLR) in modelov umetne nevronske mreže (ANN) pri napovedovanju trdote analiziranega dela GJV Figure 4. Performance of multiple linear regression (MLR)- and artificial neural network (ANN) models in predicting hardness of the analyzed GJV part o ucenju (8 dni ali 8 vrstic podatkov), zato je mogoce napovedane vrednosti primerjati z izmerjenimi vrednostmi med razlicnimi datumi. Na tej podlagi je mogoce pripraviti vrednosti modelov MAE in primerjalne grafikone (Slika 4 – levo). Ker so bili modeli pridobljeni z uporabo dnevnih povprecnih podatkov, je pomembno modele ovrednotiti z uporabo ne samo povprecnih dnevnih podatkov, temvec tudi izvirnih podatkov (Slika 4 – desno). Posledicno se je pri modelu MLR (MAE = 1,81) izkazala boljša ucinkovitost pri napovedovanju HB kot pri modelu ANN (MAE = 4,97). Ceprav je bil model pridobljen z uporabo dnevnih povprecnih podatkov, so vrednosti in trendi HB, ki ga predvideva model MLR, mocno primerljivi z izmerjenimi vrednostmi HB. Z vidika napovedovanja vrednosti Rm sta se oba modela izkazala za manj natancna. Razlog je lahko v tem, da je bila kolicina podatkov za Rm 3–4-krat manjša od kolicine podatkov za HB. Na podlagi številnih clankov model ANN obicajno prekaša model MLR [15, 16], kar je v nasprotju z rezultati v tem delu. Razlog za to bi lahko bila kolicina podatkov in njihova sledljivost. Na splošno je za model ANN potrebna veliko vecja kolicina podatkov, da bi dosegli dobro natancnost napovedi, kar v tem delu ne velja. Kolicina podatkov se znatno zmanjša, ko se prvotne vrednosti pretvorijo v dnevne povprecne vrednosti (samo 22 vrstic podatkov) zaradi priprave matrike dnevnih povprecnih podatkov. Dobra prednost metode MLR je možnost preverjanja, kateri napovedovalci (neodvisne spremenljivke) v modelu so statisticno pomembni, kar je zelo pomembno za korekcijo parametrov med samo izdelavo. To dosežemo z izracunom p-vrednosti za vsako spremenljivko (verjetnost preskusa hipoteze za korelacijo podatkov). Nizka p-vrednost vrednost nakazuje visoko stopnjo pomembnosti. P-vrednost je measurements of Rm. Day-average data matrix of this test data was also prepared in the same way as done with training data (8 days or 8 rows of data), so that the predicted values can be compared with the measured values date by date. With this MAE of the models and the comparison, charts can be carried out (Figure 4 - left). Because the models were obtained using day-average data, it is important to evaluate the models using not only day-average data but also the original data (Figure 4 - right). As a result, the MLR model (MAE = 1.81) showed better performance in predicting HB than the ANN model (MAE = 4.97). Although the model was obtained using day-average data, the values and trends of the MLR-predicted HB are very comparable with the measured HB. When it comes to predicting the Rm, both models showed lower performance. The reason could be that the amount of Rm data was 3-4 times less than the amount of HB data. Based on many articles, the ANN usually outperforms the MLR [15, 16], which is contradictory to the results in this work. The reason for this could be the amount and traceability of the data. Generally, ANN needs a large amount of data to achieve good prediction performance, which is not the case in this work. The amount of data is reduced significantly when original values are transformed into day-average values (only 22 rows of data), to realize the day-average data matrix. One good advantage of the MLR method is the possibility to check which predictors (independent variables) in a model are statistically significant, which is very important for parameter correction during production. This is achieved by calculating the p-value for each variable (probability of the hypothesis test for correlation in data). A low p-value indicates a high level of significance. The p-value is površina na repu krivulje porazdelitve Pr{|t| > t-Stat}, kjer je t t-razdeljena nakljucna spremenljivka z n-k prostostnih stopenj (k je skupno število koeficientov, vkljucno s presekom) in t-Stat izracunana vrednost t-statistike, ki je funkcija regresijskega the area in the tail of the t-distribution curve, Pr{|t| > t-Stat}, where t is a t-distributed random variable with n-k degrees of freedom (k is a total number of coefficient including the intercept) and t-Stat is the computed value of the t-statistic which is a Slika 5. Preskus pomembnosti parametrov s p-vrednostmi skupaj z razsevnimi grafikoni izmerjenih vrednosti P, Mn in HB Figure 5. Parameter significance test with p-values together with scatter charts of measured P, Mn and HB Slika 6. Primerjava natancnosti napovedovanja trdote med modelom veckratne linearne regresije (MLR) in Deikejevo formulo Figure 6. Comparison of hardness prediction performance between multiple linear regression (MLR) model and Deike’s formula koeficienta in njegove standardne napake. Na podlagi modela HB (enacba 4) so bile izracunane p-vrednosti za vsak kemicni element (Slika 5). S Slike 5 je razvidno, da ima vsebnost P najnižjo p-vrednost, kar pomeni najvecji vpliv na trdoto (HB). Na tej podlagi lahko livarna poveca vrednost HB dela GJV s povecanjem vsebnosti P. Druga možnost je, da bi bilo povecanje vrednosti HB mogoce tudi z zvišanjem vsebnosti Mn in Mg. Ucinkovitost napovedi modela MLR pri ocenjevanju trdote analiziranega dela GJV smo primerjali tudi s splošno formulo za oceno trdote litega železa avtorjev Deike in sod. (HB = 444 - 71,2×C - 13,9×Si + 21×Mn + 170×S), ki se uporablja v livarni [7]. Kot je razvidno s Slike 6 je model MLR (enacba 4), ki je bil razvit posebej za ta del GJV, povezan z vecjo natancnosti napovedovanja kot Deikejeva formula, ki se uporablja v livarni. Zato bi bilo bolje, da bi v livarni za napovedovanje trdote dela uporabili ta model MLR. Na splošno je model MLR najboljša metoda za napovedovanje mehanskih lastnosti. Z uporabo modelne livarne lahko dobro ocenimo trdoto dela GJV skozi opazovanje kemicne sestave na livni postaji, ki jo pridobimo približno 100 minut (400 form ali 800 delov) pred dejansko meritvijo trdote. 4.2 Pomembni parametri izmetanih delov in njihove ustrezne delovne meje V tem poglavju so predstavljeni in obravnavani rezultati primerjalne študije vdelanih parametrov, povezanih s temi petimi obicajnimi skupinami izmeta (Preglednica 4). function of the regression coefficient and its standard error. Based on the HB model (Eq. 4), the p-values for each chemical elements were calculated (Figure 5). According to Figure 5, it can be seen that the P content has the lowest p-value, which means the highest influence on hardness (HB). With this, foundry can increase HB of the GJV part by increasing the P content. Alternatively, increase of HB should also be possible by increasing the Mn and Mg content. Prediction performance of the MLR model in estimating hardness of the analyzed GJV part was also compared with general formula for cast iron hardness estimation by Deike et al. (HB = 444 - 71.2×C - 13.9×Si + 21×Mn + 170×S) which has been used by the foundry [7]. As can be seen in Figure 6, the MLR model (Eq. 4), which was developed especially for this GJV part, showed higher prediction accuracy than the formula by Deike, used at the foundry. Therefore, it is better for foundry to use this MLR model to predict the hardness of the part. Overall, the MLR is the preferred method for predicting the mechanical properties. By using the model foundry can well estimate the hardness of the GJV part through observation of chemical composition at pouring station, which is obtained around 100 minutes (400 moulds or 800 parts) before the real hardness measurement. 4.2 Scrap relevant parameters and their proper working boundaries The results of a comparative study of inline parameters, related to those common five groups of scrap (Table 4), are presented and discussed in this section. 4.2.1 Plinski mehurcek, vkljucek peska, poroznosti Glede na trenutno znanje so napake zaradi plinskih mehurckov in poroznosti vecinoma nastale z ujetjem zraka zaradi turbulentnega toka med litjem taljene kovine in ujetja plina, ki nastane iz uparjene vode v formi zaradi nezadostnega odzracevanja [17]. Napaka zaradi vkljucka peska nastane zaradi obrabe površine forme kot posledica delovanja toka kovine [18]. Po preiskavi vdelanih podatkov infrastrukturnega dela rešetke, kjer je bila zaznana tovrstna napaka, smo ugotovili, da sta povezana parametra previsoka vsebnost gline in vode Slika 7. Primerjava sestave materiala forme med proizvodno serijo 1A in 1B v povezavi z izmetom zaradi plinskih mehurckov, vkljuckov peska in poroznosti Figure 7. Comparison of moulding material composition between production batch 1A and 1B, associated with the gas bubble, sand inclusion and porosities scrap Slika 8. Primerjava parametrov litja med proizvodno serijo 2A in 2B v povezavi z izmetom zaradi zvarov v hladnem Figure 8. Comparison of pouring parameters between production batch 2A and 2B, associated with the cold shuts scrap 4.2.1 Gas bubble, sand inclusion, porosities Based on existing knowledge, gas bubble and porosity defects are mainly formed by entrapment of air by the turbulent flow of molten metal during pouring, and entrapment of gas produced from vaporized water from insufficient-venting mould [17]. Sand inclusion defect is formed due to abrasion of the mould surface by the impact of the metal flow [18]. After investigating the inline data of the infrastructural grate part, which had a problem with this kind of defect, too high clay and water contents of the moulding material were found to be the v materialu forme (Slika 7). Te ugotovitve ustrezajo tudi rezultatom dela W. Ali (2020) [19]. 4.2.2 Hladni zvari Napaka zaradi hladnih zvarov je posledica prezgodnjega strjevanja taline med litjem, zlasti na tankih površinah in pri majhnem dovodnem lijaku [20]. To napako je mogoce opaziti na površini litega dela. Na podlagi izmerjenih vdelanih parametrov dela stranske plošcepecice je iz Slike 8 razvidno, associated parameters (Figure 7). These findings also correspond with the work results from W. Ali (2020) [19]. 4.2.2 Cold shuts Cold shuts defect is due to premature solidification of the melt during casting, especially at the thin area and small ingate [20]. This defect can be seen on the cast part surface. Based on the measured inline parameters of the oven side plate part, it can be seen from Figure 8 that too low Slika 9. Primerjava sestave materiala forme med proizvodno serijo 3A in 3B v povezavi z ostankom razpokane forme Figure 9. Comparison of moulding material composition between production batch 3A and 3B, associated with the mould tears scrap pomarancaste površine Figure 10. Comparison of moulding material composition and pouring temperature between production batch 5A and 5B, associated with the orange-skin surface scrap da sta bila vzroka za hladne zvare prenizka temperatura litja in predolg cas litja. 4.2.3 Razpoke forme Napaka zaradi razpoke forme, ki je vidna na površini litega dela, nastane, ko je locitvena odpornost materiala forme vecja od njegove trdnosti (previsoka zbitost in/ali prenizka trdnost peska). Na podlagi rezultatov preiskave vdelanih parametrov za liti del vrat pecice, ki je povezan s težavami zaradi razpok, je bilo ugotovljeno, da sta povezana parametra prenizka vsebnost gline in vode v materialu forme (Slika 9). 4.2.4 Mehanska praska Mehanska praska je poškodba na površini litega dela zaradi mehanskega udarca. Po preiskavi vdelanih parametrov dela okvirja pecice, ki je povezan s težavami zaradi mehanskih prask, ni bil prepoznan noben ociten dokaz. Kljub temu smo ugotovili, da se mehanske praske pri velikih in težkih litih delih pojavljajo pogosto. Vzrok takega izmeta je lahkov tem, da se deli medsebojno drgnejo vzdolž transportne linije med procesom izmetavanja in cišcenja. Veliki in težki deli torej morda niso primerni za tovrstne postopke. 4.2.5 Pomarancasta površina Za pomarancasto površino je znacilen širok razpon vdolbin v strjevalnem sloju ulitka zaradi plinov, ki nastanejo na meji pri reakcijah med formo in staljeno kovino [21]. Po raziskavi vdelanih parametrov infrastrukturnega dela rešetke, povezanega s težavami zaradi pomarancaste površine, smo ugotovili, da so povezani parametri pouring temperature and too long pouring time were the causes of cold shuts. 4.2.3 Mould tears Mould tears defect, which is visible on the cast part surface, occurs when the separation resistance of the moulding material is greater than its strength (too high compaction and/or too low sand strength). Based on the investigation result of inline parameters for the oven door cast part, which had mould tears problem, too low clay and water contents of the moulding material were found to be associated parameters (Figure 9). 4.2.4 Mechanical scratch The mechanical scratch defect is a damage on the cast part surface due to mechanical impact. After investigating the inline parameters of the oven frame part which had the mechanical scratch issue, no clear evidence was recognized. Nevertheless, we found that mechanical scratch often occurred with big and heavy cast part. The reason of such scrap could be, parts scratched with each other along the transport line during shakeout and cleaning process. So, big and heavy parts may not suite this kind of process. 4.2.5 Orange-skin surface Orange-skin surface is characterized by depressions in the solidification film of the casting over a broad range because of gasses generated by reactions between the mould and molten metal at the boundary [21]. After investigating inline parameters of the infrastructural grate part, which had a problem with orange-skin surface, too low med pripravo materiala forme Figure 11. Scatter charts of the recycled sand moisture and temperature and water content added during moulding material preparation prenizka vsebnost gline in vode v materialu forme ter previsoka temperatura litja (Slika 10). Ustrezne delovne meje teh vzrocnih parametrov so bile dolocene z zajemom vrednosti iz serije z malo ostankov (serija 1B). Na splošno je bilo ugotovljeno, da je nestabilna vsebina materiala forme glavna težava za vecino izmeta. Nadaljnja preiskava je bila opravljena s preiskavo recikliranega peska. Ugotovljeno je bilo, da je vsebnost vode v materialu forme mocno povezana z vlago in temperaturo uporabljenega peska (Slika 11). Ocitno je bil mešalnik peska za formanje programiran za dodajanje vode glede na vlago in temperaturo recikliranega peska. Zato je pomembno stabilizirati vlago in temperaturo recikliranega peska ter zmanjšati korekcijo vlage v mešalniku peska naprave za formanje, da bi zagotovili stabilnejše lastnosti materiala forme in s tem zmanjšali izmet ulitkov. clay and water contents of the moulding material and too high pouring temperature were found to be associated parameters (Figure 10). The proper working boundaries of these causing parameters were determined by taking values from the low-scrap batch (Batch 1B). Overall, unstable moulding material contents were found to be the main problems for most scrap. Further investigation was performed by investigating the recycle sand. It was found that water content of the moulding material was highly associated with the used sand moisture and temperature (Figure 11). Apparently, the mouding sand mixer was programmed to add water in relation to the recycled sand moisture and temperature. Hence, it is important to stabilize the recycled sand moisture and temperature to reduce the moisture correction effort of the mouding sand mixer, to achieve more Zakljucek • Ceprav zbrani podatki o proizvodnji ne omogocajo sledljivosti, specificni za del, je mogoce pridobiti model za napovedovanje mehanskih lastnosti z zadovoljivo natancnostjo napovedi skozi uporabo dnevnih povprecnih vrednosti ali izvirnih izmerjenih vrednosti. • Trdoto analiziranega dela GJV je mogoce natancno napovedati glede na kemicno sestavo (C, Si, Mn, P, Cu, Mg in S) z uporabo metode MLR (180±2 zlitine HB). • Na podlagi rezultatov metode MLR je vsebnost fosforja v primerjavi z drugimi kemicnimi elementi pokazala najvecji ucinek na vrednost HB rotacijsko simetricnega dela GJV. • Model MLR za napovedovanje vrednosti Rm je povezan z manjšo natancnostjo napovedovanja (430±15 MPa) in bi ga bilo treba izboljšati z vec podatki. • MLR je bila najprimernejša metoda za napovedovanje mehanskih lastnosti, in sicer zaradi svoje preprostosti pri ustvarjanju in možnosti preskušanja pomembnosti vdelanih parametrov. • Z modelom lahko napovemo lastnosti dela 100 minut pred dejansko meritvijo, kar lahko pomaga prepreciti maksimalno izgubo v seriji 800 delov. • Pomembno je stabilizirati vlago in temperaturo recikliranega peska, da bi se izognili korekciji vlage v mešalniku peska. • Ustrezne delovne meje za te parametre, pomembne za izmet, so bile dolocene z opazovanjem vrednosti parametrov iz proizvodnih serij, v katerih so bili izdelani kakovostni ulitki. V prihodnje je pomembno optimizirati procesne modele za natancnejši nadzor parametrov. Verjamemo, da sta za to stable moulding material properties and thereby mitigate casting scrap. 5 Conclusion • Although the collected production data does not allow part-specific traceability, the mechanical property prediction model with satisfied prediction performance can be obtained using day-average values instead of original measured values. • Hardness of the analyzed GJV part can be well predicted based on the chemical composition (C, Si, Mn, P, Cu, Mg and S) using the MLR method Alloys (180±2 HB). • Based on the results of the MLR method, phosphorus content showed the highest effect on HB of the rotationally symmetric GJV part among other chemical elements. • The MLR model for predicting Rm showed lower prediction performance (430±15 MPa) and should be improved with more data. • The MLR was the preferred method for predicting the mechanical properties because of its simplicity in creation and possibility to test the significance of inline parameters. • With the model, we can forecast part properties 100 minutes before the real measurement, which can help prevent maximum serial loss of 800 parts. • It is important to stabilize the recycled sand moisture and temperature to avoid strong moisture correction by the mouding sand mixer. • Proper working boundaries for those scrap relevant parameters were determined by observing parameter values from production batches which produced good cast parts. kljucna natancnejša sledljivost procesov in povecanje obsega zbiranja podatkov. Vendar je treba te rešitve uresniciti na podlagi poštenih stroškov. Metodologija za sledenje posameznih litoželeznih delov ali manjših skupin litoželeznih delov bi lahko bila predmet prihodnjega raziskovalnega dela. Poleg tega bi lahko bila podrobna preiskava priprave materiala forme in meritve na liniji, npr. vlažnosti koncnega peska, še ena pomembna tema novega raziskovalnega dela, saj ti parametri mocno prispevajo k zmanjšanju kolicine odpadkov in s tem k prihranku stroškov. Viri / References In the future, it is important to optimize the process models for more accurate parameter control. We believe that more precise process traceability and more data collection are the keys to this. However, they should be realized at a fair cost. A methodology for tracing either individual or smaller groups of cast iron parts cheaply could be a future research work. In addition, detailed investigation of moulding material preparation and inline measurements, e.g. finished sand moisture, could be another important topic to work on, for they strongly contribute to scrap rate reduction and therefore cost-saving. [1] Krüger.M, Internal information of Ortrander Eisenhütte GmbH, 2018 Production capacity, scrap information. (2019). [2] M.A. Essam, A.Y. Shash, H. Megahed, E. El-Kashif, Effect of section thickness on microstructure and mechanical properties of compacted graphite iron for diesel engine applications. Heliyon 7. (2021). https://doi.org/10.1016/j.heliyon.2021.e05930 [3] R. Kumar, B. Bihari et al., Effect on the Mechanical Properties of Gray Cast Iron with Variation of Copper and Molybdenum as Alloying Elements. International Journal of Engineering Research & Technology (IJERT). (2014). [4] G. Gumienny, Effect of Chromium on the Solidification Process and Microstructure of Vermicular Graphite Cast Iron. ARCHIVES of FOUNDRY ENGINEERING. (2015). https://doi.org/10.1515/afe-2015-0054 [5] P. A. Heller and H. Jugnbluth, Gießerei 42 (1955). [6] W. Patterson et al., Zerspanbarkeit von Grauguß. Westdeutscher Verlag, Köln und Opladen 1965 (ISBN: 9783663041528), p. 20 (1965). [7] Deike et al., Formula for estimation of hardness of cast iron. (2000). [8] A, Sata et al., New Approach to Casting Defects Classification and Analysis Supported by Simulation. Conference: Indian Foundry Congress 2011. (2011). [9] S.N. Anoli, Optimization of Essential Parameters in Green Sand Process to Minimize Persisting Casting Defects Using Taguchi Approach. Journal of Engineering Science and Technology Review. (2019). [10] A. Denis, J. Antonsson, M. Clementi, SinterCast Tracking Technologies - Industry 4.0 Process Control and Traceability in the Cast Iron Foundry Industry. SinterCast AB. (2018). [11] F. Deng, R. Li, S. Klan, W. Volk, Comparative Evaluation of Marking Methods on Cast Parts of Al–Si Alloy with Image Processing. Inter Metalcast. (2021). https://doi. org/10.1007/s40962-021-00661-0 [12] N. K. Vedel-Smith and T. A. Lenau, Casting traceability with direct part marking using reconfigurable pin-type tooling based on paraffin–graphite actuators. Journal of Manufacturing Systems. (2012). https://doi.org/10.1016/j.jmsy.2011.12.001 [13] DISATAG - Trace and Guidance. https://www.disagroup.com/de-de/foundry-products/ digital-solutions/disa-tag. Accessed 2019 [14] AGentle Introduction to the Rectified Linear Unit (ReLU). https://machinelearningmastery. com/rectified-linear-activation-f. Dostopano 20. 8. 2020 [15] M. Li and J. Wang, An Empirical Comparison of Multiple Linear Regression and Artificial Neural Network for Concrete Dam Deformation Modelling. (2019). https://doi. org/10.1155/2019/7620948 [16] V. Gholami, M.R. Khaleghi, Acomparative study of the performance of artificial neural network and multivariate regression in simulating springs discharge in the Caspian Southern Watersheds, Iran. Applied Water Science. (2019). https://doi.org/10.1007/ s13201-018-0886-4 [17] S.G. Hussein, Porosity in castings. (2019). https://www.researchgate.net/ publication/331974367_Porosity_in_casting [18] V. Ingle, M. Sorte, Defects, Root Causes in Casting Process and Their Remedies: Review. Int. Journal of Engineering Research and Application. pp.47-54 (2017) https:// doi.org/10.9790/9622- 0703034754 [19] W. Ali, Defect Analysis for Sand Casting process (Case Study in foundry of Kombolcha Textile Share Company). International Research Journal of Engineering and Technology (IRJET). (2020). https://www.irjet.net/archives/V7/i1/IRJET-V7I1146.pdf [20] B.S. Kamble, Analysis of Different Sand Casting Defects in a Medium Scale Foundry Industry - AReview. International Journal of Innovative Research in Science, Engineering and Technology (IJIRSET). (2016). https://doi.org/10.15680/IJIRSET.2015.0502017 [21] H. Kambayashi, Y. Kurokawa, Evaluation with surface analysis equipment, of casting defects in cast iron articles (review). Materials Science Forum. 539-543 (2007). https:// doi.org/10.4028/www.scientific.net/MSF.539-543.1110 Z. Zovko Brodarac1, A. Mahmutovic2, S. Zeljko3, L. Zeljko1 1Univerza v Zagrebu, Metalurška fakulteta (HR), 2TC Livarstvo d.o.o., (SI), 3Plamen d.o.o. (HR) / 1University of Zagreb Faculty of Metallurgy (HR), 2TC Livarstvo Ltd, (SI), 3Plamen Ltd (HR) Numericne simulacije za optimizacijo tankostenskega ulitka EN-GJL-200 Numerical Simulation In Optimization Of Thin-Walled EN-GJL-200 Casting Povzetek Uvedba novih strategij in konceptov, kot so »ulitki skoraj koncne oblike« in »takoj v prvem poskusu,« je danes kljucnega pomena za livarsko proizvodnjo. Visok izkoristek materialov pri kar najmanjšem številu formarskih opravil in izogibanje napakam sta glavna cilja proizvajalcev ulitkov. Zato predstavljajo numericne simulacije kot del tehnologij CAD/CAE nepogrešljivo orodje za zagotavljanje konkurencnosti na globalnem trgu. Numericna simulacija litja in strjevanja predstavlja opis fizicnega pojava na podlagi matematicnega modela. Uporaba racunalniških numericnih simulacij, ki temeljijo na kompleksnih in celovitih matematicnih modelih, kot je Fourierov zakon, se pogosto uporablja pri ocenjevanju termicnih procesov v livarstvu. Numericna simulacija omogoca analogno prikazovanje metalurških procesov z izracunavanjem in graficno dispozicijo procesov od litja pa vse do koncnega strjevanja ulitka. Postopek strjevanja je izredno kompleksen proces, razumevanje katerega predvideva obsežno poznavanje vedenja in interakcij materialov in tehnologij. Kompleksnost kovinskih ulitkov sestavljajo interakcije elementov in masni prenos med strjevanjem ter tehnološki razvoj, vkljucno s prenosom toplote. Najkoristnejša numericna metoda je metoda koncnih razlik, in sicer zaradi njene enostavnosti, vendar pa je metoda koncnih elementov natancnejša in ni povezana z omejitvami v smislu kompleksnosti geometrijskih oblik. Ta raziskava se je osredotocala na optimizacijo litja in strjevanja tankostenskih ulitkov EN-GJL-200 s pomocjo numericne simulacije. Kompleksna geometrija tankostenskega ulitka predstavlja izziv zaradi skorjastih delov in upogibanja ulitka pod obremenitvijo v trenutni tehnološki konfiguraciji. Optimizacija numericne simulacije razkriva spremembe tehnoloških parametrov pri ulivnih sistemih, prezracevalnih sistemih in podpornih sistemih. Optimizacija omogoca tudi polnjenje ulitkov z namenom preprecitve izdatne toplotne preobremenitve in posledicno obremenitve konkretnih delov ulitka. Kljucne besede: numericna simulacija, EN-GJL-200, tankostenski ulitek, strjevanje Abstract Implementation of new strategies and concepts such as “Near net shape castings” and “Right for the first time” represents an imperative for nowadays foundry production. High material utilization, with a minimal number of forming operations and defects avoiding, is the main goal for casting producers. Therefore, numerical simulation as a part of CAD/CAE technologies represents an indispensable tool for achieving competitiveness in the global market. Numerical simulation of casting poring and solidification represents a description of physical phenomena based on amathematical model. The application of computer numerical simulations, which are based on complex and comprehensive mathematical models such as Fourier law, found its significant application in consideration of thermal processes in the foundry. Numerical simulation enables the analogical display of metallurgical processes by calculation and graphical disposition of the process from the pouring to the final casting solidification. The solidification process is a very complex process that comprehends knowledge related to material and technology behaviour and interactions. The complexity of metal casting consists of element interactions and mass transfer during solidification, and technological development including heat transfer. The most useful numerical method is the finite difference method due to its simplicity, but the finite element method is more accurate and does not have any limitation concerning the complexity of geometric shapes. The focus of this investigation was the optimization of pouring and solidification process of thin-walled EN-GJL-200 casting using numerical simulation. The complex geometry of thin-walled casting represents the challenge due to crusted parts and stress bending of casting in the existing technological setup. Numerical simulation optimization reveals changes in technological parameters of the gating system, venting system, and support. Optimization enables even filling with castings to prevent an extensive thermal overload and consequently stress of particular parts of castings. Keywords: numerical simulation, EN-GJL-200, thin-wall casting, solidification Uvod Numericne simulacije se vse pogosteje uporabljajo za upravljanje metalurških procesov, kot sta litje in strjevanje. Glede na cas obdelave, ki je potreben za posamezno nalogo, lahko simulacije razdelimo v dve skupini: povezane procese (numericne operacije v zelo kratkem casu, kot so vodenje procesa, priprava peska za formo, odmerjanje v peci, pihanje belega grodlja v pretvornikih, nacin delovanja gorilnika itd.) in nepovezani procesi (reševanje numericnih operacij v zelo dolgem obdobju, kot so litje, strjevanje, pretok toplote, pretok tekocine, deformacija materiala itd.). Obstajajo razlicni pristopi reševanja zgoraj navedenih problemov, npr. eksperimentalni, teoreticni in numericni. Zaradi natancnosti, uporabnosti in stroškov je najbolj sprejemljiv numericni pristop. Splošni problem prevodnosti toplote je v dolocanju temperature na vsaki tocki togega telesa, za katerega sta znana zacetna temperatura in porazdelitev toplotnega toka z dolocenimi omejenimi in 1 Introduction Numerical simulations are increasingly used to control metallurgical processes such as pouring and solidification. Depending on the time processing required for each task, the simulations can be divided into two groups: online processes (numerical operations in a very short time such as process management, preparations of mold sand, dosing of furnaces, blowing of white pig iron in converters, burner operation mode, etc.) and off-line processes (solving numerical operations over a very long period such as casting, solidification, heat flow, fluid flow, material deformation, etc.) There are different approaches to solving the above problems, such as experimental, theoretical, and numerical. The most acceptable approach is numerical due to its accuracy, applicability, and costs. The general problem of heat conduction lies down in the determination of the temperatu­re at each point of a rigid body, for which the initial temperature and heat flux distribution are known with defined limited and boundary mejnimi pogoji po Fourierjevem zakonu. Odvisnost geometrijskih in fizikalnih kolicin v mehaniki kontinuuma je vzpostavljena na podlagi diferencialnega elementa. Rešitev diferencialnih enacb se z ustrezno diskretizacijo zmanjša na rešitev sistema linearnih algebraicnih enacb z neznankami vozlišca. Najpogosteje uporabljene metode diskretizacije diferencialnih enacb prenosa toplote in mase so numericne metode [2]: metoda koncnih diferenc (FDM), metoda koncnih elementov (FEM), metoda kontrolnega volumna, metoda robnih elementov. Najpogosteje uporabljene numericne metode so: metode koncnih razlik zaradi njihove preprostosti in metode koncnih elementov zaradi pomanjkanja omejitev v povezavi s kompleksnostjo geometrijskih oblik [2]. Glede na potrebo zastavljenega problema se ustrezna numericna metoda izbere z uporabo ustrezne programske opreme za numericno simulacijo. Metoda koncnih elementov spada med metode diskretne analize in temelji na fizicni diskretizaciji obravnavane domene kontinuuma. Osnova za izracun je del domene koncnih dimenzij, tj. koncni element, kontinuum pa je mreža koncnih elementov. Glede na nacin izpeljave in oblikovanja enacb, ki opisujejo stanje elementa, obstajajo štirje osnovni vidiki metode koncnih elementov: neposredne metode, variacijske metode, metoda utežnih ostankov in metoda energijske bilance. Pri numericni analizi toplotne prevodnosti se uporabljata metodi utežnih ostankov in variacijska metoda. Za vsak koncni element je znacilna oblika elementa, število vozlišc, število in vrsta neznanih kolicin (parametrov) v posameznih vozlišcih. Iz oblike in neznane velikosti izhajajo razlicni tipi koncnih elementov, in sicer enodimenzionalni (linearni), conditions according to the Fourier law. The dependence of geometric and physical quantities in continuum mechanics is established on the differential element. The solution of differential equations is reduced through appropriate discretization to the solution of a system of linear algebraic equations with nodal unknowns. The most commonly applied method of discretization of differential equations of heat and mass transfer are numerical methods [2]: finite difference method (FDM), finite element method (FEM), control volume method, edge element method. The most commonly used numerical methods are: finite difference methods due to their simplicity and finite element methods due to lack of limitation regarding the complexity of geometric shapes [2]. Depending on the requirement of the posed problem being, an appropriate numerical method is selected, using appropriate numerical simulation software. The finite element method belongs to the methods of discrete analysis and is based on the physical discretization of the considered continuum domain. The basis for the calculation is a part of the domain of finite dimensions, i.e. the finite element, and the continuum is a network of finite elements. According to the way of deriving and formulating equations that describe the state of an element, there are four basic aspects of the finite element method: direct methods, variation methods, the weight residue method, and the energy balance method. In the numerical analysis of heat conduction, the weight residue methods and the variation method are used. Each finite element is characterized by the shape of the element, the number of nodes, the number and type of unknown quantities (parameters) in individual nodes. Depending on the shape and unknown sizes, different types of finite elements dvodimenzionalni (trikotni in štirikotni) in tridimenzionalni (tetraedrski in prizmaticni) elementi. V geometrijah z ukrivljenimi površinami se najpogosteje uporabljajo dvodimenzionalni trikotni in tridimenzionalni tetraedrski elementi [3]. Samozadostne livarne se morajo v današnjem casu prilagajati strogim zahtevam svetovnega trga in zato poiskati rešitve za njihovo izpolnjevanje na podlagi koncepta takoj v prvem poskusu [4]. V zadnjih desetletjih je najpomembnejše orodje upravljanja, ki se uporablja v livarnah, vitka proizvodnja [5, 6]. Nacela takšne proizvodnje so skrajšanje casa do prevzema izdelka, izboljšanje kakovosti in zmožnost prilagajanja zahtevam kupca. Eden gospodarnih nacinov proizvodnje ulitkov je koncept procesa proizvodnje ulitkov skoraj koncne oblike [7]. Kupcu to predstavlja zmanjšanje sekundarnih postopkov na komponentah, ki imajo manjše tolerance, za proizvajalca pa najvecji dobicek. Izpolnjevanje zahtev strank je mogoce doseci samo s celostnim pristopom. To pomeni, da je treba celotno življenjsko dobo komponente obravnavati kot sistem, v katerem je treba posamezne segmente življenjskega cikla preveriti skupaj z vsemi delujocimi dejavniki kot medsebojno odvisne dele celote. Življenjski cikel komponente razlikuje naslednje faze [8]: • razvoj (ideje, skice, konstrukcije, izbor materialov in postopka, izracun, izdelava prototipa), • proizvodnja (priprava, litje, strojna obdelava, montaža), • uporaba (delovanje, vzdrževanje, servisiranje), • recikliranje (razstavljanje, zbiranje, razvršcanje, uporaba, odlaganje odpadkov). are derived, namely one-dimensional (linear), two-dimensional (triangular and quadrangular), and three-dimensional (tetrahedral and prismatic) elements. Two-dimensional triangular and three-dimensional tetrahedral elements are most often used in geometries with curved surfaces [3]. Nowadays, self-sustainable foundries must adjust to the high demands of the global market and therefore seek a solution to meet them based on the concept Right for the first time [4]. In the last decades, the most important management tool introduced in foundries is Lean manufacturing [5,6]. The principles of such productions are to shorten the adoption time of the product, improve the quality, and be adaptable to the customer. One of the economical ways of producing cast components is the Near net shape process concept of casting to an almost finite shape / dimension [7]. For the customer, this means a reduction in secondary operations on components that have a narrower tolerance, and for the manufacturer, maximum profit. Fulfillment of customer requirements can only be achieved with an Integral approach. This means that the total lifetime of a component should be considered as a system in which individual life cycle segments should be reviewed together with all influencing factors as interdependent parts in their entirety. The life cycle of a component distinguishes the following phases [8]: • development (ideas, sketches, constructions, choice of materials and procedure, calculation, prototyping), • production (preparation, casting, machining, installation), • use (operation, maintenance, servicing), • recycling (dismantling, collection, sorting, utilization, waste disposal). Najpomembnejši segment je prva faza, torej razvoj komponente, saj doloca življenjsko dobo ostalih segmentov in faz. Socasni inženiring mocno prispeva k skrajšanju razvojnega casa in izboljšanju kakovosti izdelkov [9, 10]. Ta nacin spodbuja proizvodnjo ustreznih ulitkov s pravilnim pristopom brez napak že v prvem poskusu (takoj v prvem poskusu). S tem pristopom se vecina kljucnih odlocitev sprejme pred izdelavo izdelka, ko je izvajanje sprememb najenostavnejše in najcenejše. Bistveni elementi socasnega inženiringa so: tehnologije CAD / CAM / CAE, hitra izdelava prototipov, centralno upravljanje proizvodnih podatkov, vecfunkcijske ekipe, analiticne metode, ekspertni sistemi in zbirke znanja. Optimizacija procesa litja je dosežena s tehnologijo CAE. Cilj takšne optimizacije je prihranek materiala, zmanjšanje mase, zvecanje dovoljene obremenitve, napetosti, togosti itd. K tej optimizaciji smo pristopili z izdelavo modela na podlagi uporabe razlicnih racunalniško podprtih tehnologij. Za virtualno izdelavo 3D-ulitkov se uporablja tehnologija CAD, na podlagi katere se preverjajo zahtevane lastnosti izdelka: napetosti, deformacije, sposobnost litja itd. [11]. V zadnjih dvajsetih letih je bila uporaba simulacije polnjenja kalupov obvezen segment tehnologije CAE. Numericna simulacija z vkljucenim termotehnicnim izracunom doloca prenos toplote z ulitka v formo, graficni prikaz toplotnega polja in razvoja napetosti. Iz pridobljenih rezultatov je mogoce dolociti strjevanje, stopnjo polnjenja forme, temperaturno in toplotno porazdelitev napetosti, mikrosegregacijo, vroce tocke itd. [12]. Ko virtualno litje izpolnjuje vse predvidene zahteve, je mogoce pristopiti k hitri izdelavi prototipov v razlicnih razpoložljivih materialih, vkljucno z originalnim materialom ulitka in ustrezno tehnologijo ulitka [13]. Na izdelanem The most important segment is the first phase, i.e. the development of the component because it determines the lifespan of other segments and phases. Simultaneous engineering contributes significantly to the reduction of development time and the improvement of product quality [9, 10]. This approach encourages the production of correct castings with a zero defect approach in the first attempt first and correctly (Right for the first time). With this approach, most key decisions are made before constructing a product, when changes are easiest and cheapest. Essential elements of simultaneous engineering are: CAD / CAM / CAE technologies, rapid prototyping, central production data management, cross-functional teams, analytical methods, expert systems, and knowledge bases. Optimization of the casting process is achieved using CAE technology. The goal of such optimization is to save material, minimize mass, maximize the allowable load, stress, stiffness, etc. This optimization is approached by creating a model using various computer-aided technologies. CAD technology is used for the virtual production of 3D castings, based on which the required product requirements are examined: stresses, deformations, casting ability, etc. [11]. In the last twenty years, the use of mold filling simulation has been a mandatory segment of CAE technology. Numerical simulation with incorporated thermal- technical calculation determines the heat transfer from the casting to the mold, a graphical representation of the thermal field, and stress development. From the obtained results, it is possible to determine solidification, mold filling rate, temperature, and thermal stress distribution, microsegregation, hot spots, etc. [12]. When the virtual casting meets all the required requirements, rapid prototyping prototipu je treba opraviti vse zahtevane preskuse, nato pa se zacne proizvodnja komponente. V literaturi je na voljo mnogo primerov podobnega pristopa [14, 15]. Ta raziskava se je osredotocala na optimizacijo litja in strjevanja tankostenskih ulitkov EN-GJL-200 s pomocjo numericne simulacije, ki predstavljata pomemben del socasnega inženiringa. Optimizacija numericne simulacije razkriva spremembe tehnoloških parametrov pri dotocnih siste­mih, prezracevalnih sistemih in podpornih sistemih. Optimizacija omogoca tudi polnje­nje ulitkov z namenom preprecevanja izdatne toplotne preobremenitve in posle-dicno obremenitve konkretnih delov ulitka. Poskusni postopek Livno železo EN-GJL-200 ima zaradi luskaste strukture grafita nizko natezno trdnost, trdoto, žilavost, raztezek in modul elasticnosti [16-19]. Ena glavnih pomanjkljivosti tega materialaje obcutljivost na debelino stene (velikost preseka), kar lahko povzroci znatne razlike v mehanskih lastnostih znotraj ulitka. Ciljni ulitek s kompleksno geometrijo je prikazan na Sliki 1. Obicajni proizvodni proces želenega ulitka je bil izveden z linijo za formanje in litje DISAMATIC®. Kompleksna geometrija tankostenskega ulitka predstavlja izziv zaradi skorjastih delov in upogibanja ulitka pod obremenitvijo v trenutni tehnološki konfiguraciji. Numericna simulacija je bila izvedena z uporabo programske opreme ProCAST na obstojecem materialu in tehnologiji ter z optimiziranimi parametri. Glede na ugotovljene napake se je optimizacija osredotocala na tehnološke parametre, kot so korekcija ulivnega sistema, odzracevalnega sistema in obstojece can be approached in various available materials including original casting material and corresponding casting technology [13]. All the required tests should be performed on the obtained prototype, after which the production of the component is started. A number of examples of similar approach is available in the literature [14, 15]. The focus of this investigation was the optimization of pouring and solidification process of thin-walled EN-GJL-200 casting using numerical simulation as an important part of simultaneous engineering. Numerical simulation optimization reveals changes in technological parameters of the gating system, venting system, and support. Optimization enables even filling with castings in order to prevent an extensive thermal overload and consequently stress of particular parts of castings. 2 Experimental Grey iron quality EN-GJL-200 due to the flaky structure of graphite, possess low tensile strength, hardness, toughness, elongation, and modulus of elasticity [16­19]. One of the main disadvantages of this material is the sensitivity to wall thickness (cross-sectional size), which can result in significant variations in mechanical properties within the casting. Target complex geometry casting is shown in Figure 1. The usual production process of target casting was performed using the DISAMATIC® line for molding and casting. The complex geometry of thin-walled casting represents the challenge due to crusted parts of casting and stress bending of casting in the existing technological setup. Numerical simulation was performed using ProCASTsoftware in existing material and technology set up and with optimized podpore. Izvedena sta bila numericna simulacija in tudi proces litja z originalnimi in optimiziranimi parametri z namenom primerjave vpliva spremenjenih parametrov na nastanek napak. Rezultati in razprava Postopek optimizacije se je zacel z oceno celotne teže ulitka in nacina polnjenja, kar je privedlo do težnje po zagotovitvi laminarnega in enakomernega pretoka po ulivnem sistemu ter odpravo odvecne teže odzracevalnih delov. Kompleksna geometrija preiskovanega ulitka predvideva uporabo podpore, da bi se izognili ukrivljenosti nepodprtih delov. Ta podpora omogoca tudi dodatno polnjenje, ceprav je odvisno od dotocne zmožnosti podpiranja livne votline. Izvirna (V1) in optimizirana razlicica (V2) spremenjene geometrije sta predstavljeni na Sliki 2. Pomikanje nosilca proti rebru ulitka zagotavlja njegovo aktivno vlogo pri enakomernem polnjenju livne votline, kot je prikazano na Sliki 3. Zmanjšanje velikosti zracnika in sprememba geometrije sta parameters. According to identified defects, the focus of optimisation was placed on technological parameters such as correction of inflow system, venting system, and existing support. Numerical simulation and also casting process were performed with original and optimized parameters to compare the influence of changed parameters on defect occurrence. 3 Results and discussion The optimization process started with an evaluation of complete casting weight and filling mode which resulted in a tendency to achieve laminar and uniform flow over the ingate system as well as the elimination of excess weight on the venting parts. The complex geometry of investigated casting requires support to avoid free parts curvature. This support also enables additional filling, although it depends on the ingate ability to support the mold cavity. Original (V1) and optimized version (V2) of changed geometry are presented in Figure 2. Scrolling the support toward rib part casting enables its active role in the uniform filling of the casting cavity as shown in Skupna teža ulitka = 25,7 kg Skupna teža ulitka = 24,3 kg Total casting weight = 25,7 kg Total casting weight = 24,3 kg Slika 2. Originalna (V1) in optimizirana razlicica (V2) spremenjene geometrije Figure 2. Original (V1) and optimized version (V2) of changed geometry V1 24 % (2,5 kg) + 76 % (7,9 kg) V2 35 % (3,2 kg) + 65 % (6 kg) Teža ulitka in odzracevanje = 10,4 kg Teža ulitka in odzracevanje = 9,2 kg Casting and venting weight = 10,4 kg Casting and venting weight = 9,2 kg Slika 3. Nacin polnjenja originalne (V1) in optimizirane (V2) razlicice Figure 3. Filling mode for both original (V1) and optimized (V2) version povzrocila zmanjšanje skupne teže, kar je ostala prvotna vloga. Glede na enake zacetne parametre litja: temperatura litja Tp= 1400 °C in cas polnjenja t=5 s v mešanico svežega peska in bentonita, je numericna simulacija litja V1 Figure 3. Reduction in the vent size and changing the geometry resulted in total weight reduction remaining in its original role. According to the identical initial casting parameters: pouring temperature Tp= 1400 V2 Slika 4. Cas do distribucije solidusa originalne (V1) in optimizirane (V2) razlicice Figure 4. Time to solidus distribution for both original (V1) and optimized (V2) version V1 V2 Slika 5. Izracun pojavnosti poroznosti originalne (V1) in optimizirane (V2) razlicice Figure 5. Porosity occurrence calculation for both original (V1) and optimized (V2) version in strjevanja pokazala razlicno toplotno porazdelitev pri litju, kar je vodilo v razlike v casu do solidusa v pomembnih kriticnih vozlišcih, kot je prikazano na Sliki 4. Obmocje v spodnjem delu ulitka je na splošno povezano s krajšanjem casa strjevanja. Cas do solidusa je najdaljši, a še vedno precej enakomeren, tj. 80–110 s, saj je to obmocje v neposrednem stiku z dovodnim kanalom. Vendar pa je bilo zabeleženo skrajšanje casa strjevanja za 2 s v kriticnih debelostenskih odsekih. Rebra in podpora so povezana s casom 30–60 s do konca strjevanja. Zgornji del ulitka je prav tako povezan s krajšim casom strjevanja, kar razkriva širše površine, za katere je znacilna casovna oznaka pribl. 60 s. Na splošno omogoca hitrejše strjevanje krajši cas segrevanja in prepreci nastanek skorje. Zmanjšanje velikosti in volumna odzracevalnih odprtin ter razširitev velikosti dotoka in pomikanje opore v smeri proti rebrom omogocajo znižanje volumna taline in s tem teže ulitka kot tudi hitrejše polnjenje in strjevanje. Vroce tocke so pokazale pojav poroznosti, kar je bilo potrjeno s simulacijo v graficni predstavitvi na Sliki 5. Izracun poroznosti ni nakazal razlike v položaju videza, ceprav je bila velikost v optimizirani razlicici manjša. Tudi ta položaj je sprejemljiv, ce je poroznost v notranjosti ulitka. Ker smo premaknili oporo, potrebno za preprecevanje deformacije prosto litih delov, je bila izvedena tudi simulacija deformacije pri litju v formo ali zrak, kot je prikazano na Sliki 5. Simulacija deformacije je pokazala popacenje ulitka v formi tudi naknadno v zraku, in sicer zaradi kompleksne geometrije, sestavljene iz tankih in debelih sten ter reber. Ker je ta ulitek del celotnega izdelka, popacenje ni sprejemljivo in ga je treba zmanjšati na najmanjšo raven. Ceprav se popacenju ni mogoce popolnoma izogniti, °C and filling time t=5 s in green sand bentonite mixture, numerical simulation of pouring and solidification revealed different thermal distribution in casting and therefore resulted in a different time to solidus in critical nodes as presented in Figure 4. The area in the lower part of casting indicated shortening of solidification time in general. Time to solidus is the longest but still rather uniform 80-110 s since this area is in direct junction to inflow. Although, shortening of solidification time for 2 s has been noticed in critical thick- walled sections. Ribs and support indicated 30-60 s to solidification end. The upper part of casting also shows shortening of solidification time revealing wider areas dedicated to cca 60 s time mark. In general, faster solidification enables shorter time of heating avoiding crust appearance. Lowering the vents size and volume and widening the inflow size as well as scrolling the support toward ribs enable lowering the melt volume, and therefore casting weight and also faster filling and solidification The hot spots indicated porosity occurrence which has been confirmed with simulation in graphical representation in Figure 5. Porosity calculation did not show difference in position of appearance, although the size in the optimized version was smaller. Also, this position is accept­able if the porosity is inside the casting. Since the support needed for avoiding the deformation of free casting parts was moved, the deformation simulation for casting in the mold and the air was also performed, as shown in Figure 5. Deformation simulation indicated distortion of casting in mould also afterward in the air due to complex geometry consisting from thin and thick-walled sections and ribs. Since this casting is one part of thecomplete product distortion is not acceptable and should be lowered to minimum. Although, Slika 6. Numericna simulacija deformacije v razlicnih casih; a) v formi (50 min po vlivanju), b) v zraku Figure 6. Numerical simulation of deformation in different time marks; a) in mold (50 min after pouring), b) in the air je njegov videz maksimalno zmanjšan, kot je prikazano na Sliki 6. Poleg predstavljenih rezultatov so bili izracunani tudi feritna frakcija, izkoristek in natezna trdnost, razmik dendritnih vej in trdota. Nobena od teh lastnosti ni pokazala bistvenega odstopanja. Numericna simulacija omogoca uporaben pregled delovanja sistema talina/litje – forma. Ker so pridobljeni rezultati zagotovili dobro optimizacijo v zvezi z litjem in polnjenjem forme, vendar pa niso bili tako zanesljivi v povezavi s strjevanjem in pridobljenimi deformacijami, je nastala potreba po dodatni optimizaciji. 4 Zakljucki S to preiskavo smo potrdili pomen nove strategije in konceptov. Cas izracuna ter the distortion is not completely avoided its appearance is minimized as shown in Figure 6. Beside presented results ferrite fraction, yield and tensile strength, dendrite arm spacing and hardness were also calculated. Neither of those properties did show a significant deviation. Numerical simulation enables useful review in the melt/casting – mold system behaviour. Since obtained results achieved good optimisation relating to pouring and mold filling, but not so satisfying relating to solidification and obtained deformations, it imposed the need for additional optimization. 4 Conclusions The importance of new strategies and concepts was confirmed through this ProCAST. Poudarek je bil na optimizaciji procesa litja in strjevanja z modeliranjem ulivnega sistema, zracnikov in podpore. Optimizacija nakazuje enakomerno polnjenje z boljšo porazdelitvijo dovodne aktivnosti in zmanjšano toplotno preobremenitev na posameznih položajih ulitka. Teža ulitka se je znatno zmanjšala in celoten cas strjevanja se je skrajšal. Kompleksna geometrija tankostenskega ulitka predstavlja izziv zaradi nagnjenosti k deformacijam. Optimizacija ni v celoti uresnicila cilja preprecevanja deformacij, vendar razkriva pozitivne povratne informacije za pomembne položaje (referencne oznake). Ceprav predstavlja optimizirano litje izboljšanje, zahteva ta geometrija ulitka vecjo optimizacijo tehnološke nastavitve, da bi že v prvem poskusu dosegli pravo rešitev. Viri / References was optimized using ProCAST software. The focus was placed toward optimizing the pouring and solidification process though modelling the inflow system, vents, and support. Optimizition indicates uniform filling with better distribution of inflow activity and reduced thermal overload in particular positions of the casting. Casting weight was significantly reduced and total solidification time was shortened. The complex geometry of thin-walled casting represents the challenge due to deformation tendency. Optimization did not completely reach the goal of deformation avoidancebut it reveals positive feedback for significant positions (reference marks). Although, optimized casting revealed improvement, this casting geometry requires more optimization technological setup to achieve right for the first time. [1] J. Crnko, Modeliranje procesnih operacija u metalurgiji, Metalurški fakultet, Sisak 1990. [2] J. H. Lienhard IV, J. H. Lienhard V, A heat transfer textbook (3rd edition), Phlogiston Press, Chambridge, Massachusetts, 2003. [3] E. R. Champio, Finite element analysis in manufacturing engineering; McGraw-Hill Inc., New York, 1992. [4] E. Ambos, I. Behm, Ganzheitliche Entwicklung-Quelle fur innovative Produkte, Processe und Werkstoffe, Giesserei (1993)3, 35-39 [5] I. Hrgovic, Temeljne postavke lean proizvodnje, Ljevarstvo 45 (2003)3, 89-92 [6] G. Engels, H. J. Heine, Shifting Goals for Foundries in German & Worldwide, Foundry Managment & Technology 1995,5, p.34 [7] P. M. Bralower, Near Net Shape Processes Needed Now, Modern Casting, 1987, 3, 124 [8] R. Sahm, N. Hansen, Towards Integrated Modelling for Intelligent Casting, Livarski Vestnik 48 (2001)2, 34-44 [9] E. Ambos, I. Behm, M. Brahmann, T. Hornig, I. Hofmann, Gießerei und Modellbau im Blickfeld der Hoch Technologien, Giesserei 84 (1997)16, 15-18. [10] I. Hrgovic, Uvod u simultano inžinjerstvo, Livarstvo 45 (2003)1, 5-14 [11] G. Hartmann, P. Bernbeck, V. Kokot, Gießereien als Entwicklungspartner der OEMs, Giesserei 90 (2003)6, 44-55 [12] M. Scheider, S. Andersen, Use Simulation to Analyze Macrosegregation, Hot tears, Heat Tretmant in Steel Castings, Modern Casting 2000, 5, 39-43. [13] I. Pfeifer-Schaller, F. Klein, Zerstorungsfrei Baulteilprufng an Aluminium-und Magnesium Druckgussteilen mit Hilfe der Computertomografie, Giesserei-Rundschau 50 (2003)5/6, 109-116. [14] Z. Zovko Brodarac, M. Targuš, N. Dolic, M. Radoš, Optimization of grey cast iron casting technology by numerical simulation, 47th International October Conference on Mining and Metallurgy Proceedings, ed. A. Kostov, M. Ljubojev, Bor: Mining and Metallurgy Institute Bor, 2015, 355-358 [15] Z. Zovko Brodarac, M. Targuš, A. Mahmutovic, Optimizacija proizvodnje odljevaka primjenom simultanog inženjerstva, International Conference MATRIB 2015 Materials, Wear, Recycling; Proceedings, ed. D. Coric, I. Žmak, Zagreb: Hrvatsko društvo za materijale i tribologiju, 2015, 76-90 [16] D. B. Craig, M. J. Hornung, T. K. McCluhan, Gray Iron, poglavlje u knjizi Metals Handbook, Volume 15, Casting, ed. ASM International Handbook Committee, ASM International, Metals Park, Ohio, 1988., 629-646. [17] HRN EN ISO 945-1:2018, Mikrostruktura željeznih lijevova -- 1. dio: Razredba grafita vizualnom analizom (ISO 945-1:2008; EN ISO 945-1:2008), Državni zavod za normizaciju i mjeriteljstvo, Zagreb. [18] HRN EN 1560:2012, Ljevarstvo -- Sustav oznacivanja željeznog lijeva -- Simboli materijala i odbrojcavanje materijala (EN 1560:2011), Državni zavod za normizaciju i mjeriteljstvo, Zagreb. [19] HRN EN 1561:2011, Ljevarstvo -- Sivi željezni lijevovi (EN 1561:2011), Državni zavod za normizaciju i mjeriteljstvo, Zagreb J. Fercec1, A. Slana1, A. Šibila1 1TALUM d.d. Kidricevo, Tovarniška cesta 10, 2325 Kidricevo (SLO) Karakterizacija napak v ulitkih iz aluminija, proizvedenih z nagibnim gravitacijskim litjem, v povezavi s procesnimi parametri Characterization of Defects in Aluminum Castings Produced by Tilt Gravity Casting in Relation to Process Parameters Povzetek V procesu proizvodnje ulitkov iz aluminija se velikokrat soocimo z napakami, ki jih zaznamo, ko so ulitki že sestavljeni oziroma tik pred uporabo. Med napake sodi tudi netesnost. Ta napaka plinu ali tekocini omogoca, da preide skozi stene ulitkov, zato je za nekatere ulitke kljucna karakteristika tesnost. Med takšne ulitke sodijo ohišja, pokrovi itd. Za doseganje stabilnega proizvodnega procesa in razumevanje osnovnega vzroka napak je treba izvesti raziskave oziroma karakterizacijo napak z racunalniško tomografijo, opticno mikroskopijo in vrsticno elektronsko mikroskopijo. V clanku je predstavljena sistematicna raziskava netesnih kosov, ulitih s postopkom nagibnega gravitacijskega litja, za dolocitev vzroka za netesnost. Vzrok za netesnost smo želeli povezati s procesnimi parametri, ki povzrocijo napako, ki vodi do netesnosti. V raziskavi smo ugotovili, da sta glavna vzroka za netesnost kosov, ulitih z nagibnim gravitacijskim litjem, vkljucki oziroma oksidne kožice in krcilna poroznost. Pri nastanku krcilne poroznosti smo se osredotocili na vpliv temperature orodja. Meritve temperature orodja smo opravili s termicno kamero. Za dolocitev vpliva temperature orodja na strjevanje ulitka smo izvedli simulacijo litja. Kljucne besede: nagibno gravitacijsko litje, napake v ulitkih, procesni parametri Abstract In the production process of aluminum castings, we often face defects that are detected when castings have already been assembled or just before final use. These defects include leaks that allow gas or liquid to pass through the walls of castings. Thus, for some castings tightness represents the key feature. Such castings include housings, covers, etc. To achieve a stable production process and to understand the root cause, it is necessary to perform research or characterization of defects with computer tomography, optical microscope, and scanning electron microscopy. In this paper, we performed systematical research of leaking parts cast with tilt gravity casting to determine the cause for the leakage. We wanted to relate the causes of leakage to process parameters affecting defects that lead to leakage. In the study, we found that the main reason for leakage of cast parts cast by tilt gravity casting are inclusions or oxide film and shrinkage porosity. For the formation of shrinkage porosity, we focused on the influence of mold temperature. For the measurement of mold temperature, we used a thermal imaging camera. To determine the impact of mold temperature on the solidification of cast parts we used a casting simulation. Keywords: tilt gravity casting, casting defects, process parameters kjer se livno orodje odpre. Proizvedeni ulitek se nato vzame iz stroja [1, 2]. V proizvodnem procesu gravitacijskega litja lahko nastanejo razlicne napake. Na to vpliva veliko parametrov, kot so aluminijeva zlitina, geometrija ulitka, orodje itd. Neka­tere napake vplivajo na videz ulitka, druge pa povzrocijo poslabšanje njegovih lastnosti [3, 4]. Namen clanka je raziskati nastanek netesnosti ulitka, proizvedenega z nagibnim gravitacijskim litjem. Netesnost pri nagibnem gravitacijskem litju ni pogost pojav. Ulitki, liti s to tehnologijo, imajo debelejše stene v primerjavi z ulitki, izdelanimi z visokotlacnim litjem, pri katerem je netesnost pogostejša. Ta napaka se zaznava s preskusom netesnosti, tocna lokacija pa se doloci s preskusom z mehurcki. Netesnost v ulitkih pomeni, da pride do nastanka poti, ki notranjo steno ulitka povežejo z zunanjo. V clanku so predstavljeni trije razlicni pojavi netesnosti v ulitkih. Eksperimentalno delo V raziskavi smo uporabili ulitke, pri katerih smo netesnost poiskali s preskusom. Ulitki so bili izdelani iz livne zlitine AlSi8Cu3. Za dolocitev natancne lokacije netesne poti v ulitku je bila uporabljena racunalniška tomografija (CT). Lokacijo netesnosti smo analizirali z opticno mikroskopijo (OM) in vrsticno elektronsko mikroskopijo (SEM in EDX). S simulacijo litja s programom Magmasoft smo raziskali osnovni vzrok za krcilno poroznost. Za meritve dejanske temperature smo uporabili termicno kamero za spremljanje površinske temperature orodja, in sicer Chem-Trend ter Inprotect IRT. tool is opened. The produced cast part is then removed [1, 2]. In the production process of gravity casting, various defects can occur. There are many parameters with possible influence such as aluminum alloy, the geometry of casting, die, etc. While some defects affect the appearance of the casting there are defects that can cause deterioration of the product properties [3,4]. This paper aimed to study the formation of leakage of cast parts produced by tilt gravity casting. Leakage defect is not a common defect for the tilt gravity casting procedure. Castings cast by this technology have thicker walls compared to castings cast by high-pressure die casting, where leaks are more common. This defect is detected by the leakage test while the exact location is detected by the bubble test. Leakage of casting means there is a leakage path provided connecting the inner wall to the outer wall. This paper presents three different leakage occurrences of the castings. 2 Experimental Work In our research, we used cast parts where a leak was found with a leakage test inspection. Cast parts were produced with cast alloy AlSi8Cu3. Computed tomography (CT) was used to determine the exact position of the leakage path within the cast part. The leak location was analysed with an optical microscope (OM) and a Scanning Electron Microscope (SEM) EDX. We used a casting simulation with the Magmasoft program to investigate the root cause for the shrinkage porosity. To measure the actual temperatures, we used a thermal imaging camera to monitor the surface temperature of the tool. Specifically, we used Chem-Trend & inprotec IRT thermal imaging camera. 3 Rezultati 3.1 Racunalniška tomografija (CT) V industrijskem okolju se rentgensko slikanje pogosto uporablja za pregledovanje ulitkov, saj omogoca hitro izvedbo analize, vendar pa s to analizo ni mogoce pridobiti zahtevane informacije o vzroku netesnosti. Zato smo uporabili 3D-racunalniško tomografijo, s katero je mogoce pridobiti informacije o velikosti por in razdaljah med njimi. Analiza CT, ki smo jo izvedli na vec vzorcih, je pokazala, da netesnost povzrocijo vkljucki in krcilna poroznost. Na Sliki 2 so prikazani trije razlicni vzroki za netesnost, ki so bili zaznani pri analizi CT. Na Sliki 2a je prikazano, da poroznost ni bila zaznana kot napaka, vidna pa je napaka, ki je videti kot vkljucek. Na Sliki 2b je prikazana kombinacija vkljucka in poroznosti. V tretjem primeru smo zaznali netesnost zaradi velike pore (Slika 2c). 3.2 Preiskava z opticnim mikroskopom (OM) Na podrocju, na katerem smo z uporabo analize CT zaznali napake, smo izvedli metalografsko preiskavo. Rezultati so prikazani na Sliki 3. Na Sliki 3a je prikazana razpoka (pot za netesnost – rdeca pušcica) oziroma oksidna kožica. Podobna situacija je predstavljena na Sliki 3b, vendar z zaznano poroznostjo. Kombinacija teh dveh pojavov na ulitku povzroci netesnost. Na Sliki 3c je prikazana potrjena prisotnost velike pore, ki je bila vidna z analizo CT. Netesnost povzroci velika krcilna poroznost na steni ulitka. 3 Results 3.1 Computed Tomography (CT) Scan In an industrial environment, the 2D radiographic method (x-ray) is commonly used to inspect cast parts since the analysis can be performed quickly. However, with this analysis, we cannot obtain the required data for the cause of the leakage. For these reasons, we use 3D computed tomography scans which can provide information on pore sizes and interpore distances. CT analysis was performed on several samples, which showed that leakage occurs due to inclusions and shrinkage porosity. Figure 2 shows three different occurrences of leakage which we detected on CT. In Figure 2a the CT does not recognize the defect as porosity. However, there is a visible defect that looks like inclusion. Figure 2b shows a combination of inclusion and porosity. In the third case, we detected a leak due to a large pore (Figure 2c). 3.2 Optical Microscope Examination On locations where we detected defects using CT analysis, we performed a metallographic examination. Results are presented in Figure 3. Figure 3a shows a crack (leakage path - red arrow) or oxide skin. A similar situation is presented in Figure 3b, however, there is also a noticeable porosity. A combination of the two effects causes leakage of the cast part. Figure 3c confirms the large pore which has been seen on the CT scan. A large shrinkage porosity in the wall of the casting causes the leakage of the cast part. Slika 2. Slike precnih presekov, izvedene z racunalniško tomografijo, s prikazanim podrocjem netesnosti: a) vkljucek v vzorcu, b) poroznost + vkljucek, c) krcilna poroznost Figure 2. Images of Computed Tomography cross-section of samples displaying the leak location, a) inclusion in samples; b) porosity + inclusion; c) shrinkage porosity Slika 3. Rezultati metalografske preiskave podrocij na ulitku z racunalniško tomografijo: a) oksidne razpoke oziroma kožice, b) oksidna kožica + krcilna poroznost, c) velika poroznost Figure 3. Results of the metallographic examination of the cast parts by the tomographic scan; a) oxide crack or skin; b) oxide skin + shrinkage porosity; c) large porosity 3.3 Vrsticna elektronska mikroskopija (SEM) Glavni namen karakterizacije z vrsticnim elektronskim mikroskopom (SEM) je bila analiza morfologije razpok in poroznosti ter opredelitev vrste vkljucka. Na razpoki smo izvedli analizo EDX. Na Sliki 4 so prikazana podrocja, na katerih smo izmerili kemicno sestavo napak. Meritev EDX (Slika 5) je razkrila kemicno sestavo vkljucka, ki vsebuje ogljik (C) in kisik (O). Rezultati potrjujejo naše domneve iz preiskave z OM, da so v ulitku prisotne oksidne kožice. 3.3 Scanning Electron Microscopy (SEM) The main goal of the characterization technique with Scanning Electron Microscopy (SEM) was to analyse the morphology of cracks and porosity and define the type of inclusions. We performed EDX analysis on the cracks. Figure 4 presents locations where we measured the chemical composition of defects. EDX measurement (Figure 5) revealed the chemical composition of inclusion consists of carbon (C) and oxygen (O). This result confirms our assumptions from OM (crtkani krog). Termicna analiza prikazuje nestabilno temperaturo na karakteristicnih podrocjih, ko je operater izvedel vzdrževalne posege. 3.5 Simulacije litja Eden pomembnejših procesnih parametrov, ki vplivajo na krcilno poroznost, je temperatura orodja oziroma kokile. Študije in reference [5] kažejo, da temperatura orodja vpliva na nastanek krcilne poroznosti, ki se veca z narašcanjem temperature orodja. Krcilna poroznost je napaka, ki nastane zaradi pomanjkljivega napajanja. Velike pore, prikazane na Slikah 2c in 3c, zelo redko nastanejo v stabilnem procesu, vendar se pojavljajo in želeli smo ugotoviti njihov vzrok. Pri preiskavi s termicno kamero smo opazili, da se v stabilnem ciklu pojavijo spremembe v temperaturi orodja na karakteristicnih lokacijah, in to do 20 °C (crtkani krog na sliki 6b). To je bil razlog, da smo raziskali mogoce spremembe strjevalne fronte in potencialno tveganje za nastanek krcilne poroznosti s simulacijo litja. Simulacijo smo izvedli z obicajno temperaturo orodja 380 °C in s temperaturo nad obicajno. Na Sliki 7b je prikazana strjevalna fronta na karakteristicnem podrocju (tocka 1 na Sliki 6a) s temperaturo 361 °C (Slika 7a). Zaznali smo (Slika 6b, graf), da pride do odstopanj, saj se temperatura lahko poviša do 380 °C. Zato smo simulirali temperaturo orodja 400 °C, pri cemer smo na karakteristicnem podrocju dosegli temperaturo približno 380 °C (Slika 7c). Na Sliki 7d je prikazana strjevalna fronta, ki se je v primerjavi z obicajnim ciklom spremenila. Višja temperatura povzroci pocasnejše strjevanje v ulitku. Ta sprememba v strjevalni fronti lahko vodi do nastanka krcilne poroznosti na karakteristicnem podrocju. after major maintenance work, temperature rises (dashed circle) at a characteristic location (point 1). Thermal analysis showed the unstable temperature at characteristic locations when the casting operator performed maintenance interventions. 3.5 Casting simulation One of the process parameters influencing shrinkage porosity is the casting tool or mold temperature. Astudy in reference [5] shows that mold temperature influences formation of shrinkage porosity which increases with increasing mold temperatures. Shrinkage porosity is a defect that occurs due to a failure in effective feeding. Large pores as shown in Figures 2c and 3c are a very rare occurrence in a stable process. However, they do turn up and we wanted to identify the root cause. Examining the thermal images, we observed that in stable cycles the change in tool temperature can reach up to 20 °C at a characteristic location (dashed circle in Figure 6b). For this reason, we examined the possible change of solidification front and potential risk of shrinkage porosity formation with the casting simulation. We performed casting simulations with normal mold temperature at 380 °C and with a temperature above normal. Figure 7b shows solidification fronts on the characteristic location (Point 1 in Figure 6a) with a temperature at 361 °C (Figure 7a). Figure 6b (graph) indicates a deviation at other locations where the temperature increases up to 380 °C. For this reason, we simulated mold temperature at 400 °C whereby we achieved a temperature near 380 °C at the characteristic location (Figure 7c). Figure 7d shows solidification fronts that are changed compared to the normal cycle. Higher mold temperature influences slower solidification of the cast part. This Slika 7. Simulacija litja ulitka: a) temperatura orodja 380 °C s temperaturo 361 °C na karakteristicnem podrocju, b) strjevanje pri temperaturi orodja 380 °C, c) temperatura orodja 400 °C s temperaturo 377 °C na karakteristicnem podrocju, (d) strjevanje pri temperaturi orodja 400 °C Figure 7. Casting simulation of casting part; a) mold temperature380 °C with temperature 361 °C on the characteristic location; b) solidifications for mold temperature 380 °C; c) mold temperature 400 °C with temperature 377 °C on the characteristic location; d) solidification for mold temperature 400 °C 4 Zakljucki V raziskavi so bile raziskane napake za netesnost ulitkov, proizvedenih z nagibnim gravitacijskim litjem. Ceprav so v proizvodnjo ulitkov uvedeni ukrepi za preprecitev vkljuckov, je pojav oksidnih vkljuckov v ulitkih vedno mogoc. Debelejše oksidne kožice lahko v procesu nagibnega gravitacijskega litja preprecimo z uporabo keramicnega filtra, ki ga vstavimo v orodje pred zacetkom cikla. S spremljanjem uporabe keramicnega filtra za vsak cikel je change in the solidification front can lead to the formation of shrinkage porosity at the characteristic location. 4 Conclusions This study investigated leakage defects in cast parts produced by tilt gravity casting. Although casting production involves measures to prevent inclusions there is always a possibility for an oxide inclusion occurrence in the casting. A thicker oxide mogoce zmanjšati tveganje za debelejše oksidne kožice v ulitku in potencialno netesnost. Velika krcilna poroznost lahko nastane zaradi razlicnih procesnih parametrov. V clanku smo se osredotocili samo na temperaturo orodja. S termicno kamero smo na karakteristicnih podrocjih opazili spremembe v temperaturi orodja, in to do 20 °C. S simulacijo litja smo simulirali spremembe temperature orodja za dolocitev vpliva na strjevanje. Spremembe, ki lahko vplivajo na nastanek krcilne poroznosti, smo opazili na poteku strjevalne fronte. Prav zato je pri nagibnem gravitacijskem litju pomembno nadzorovati temperature orodja, ceprav je to v industrijskem okolju velik izziv. Viri / References skin in the tilt gravity casting process can be prevented with a ceramic filter which is put on the mold before the start of a cycle. By monitoring the utilization of the ceramic filter for each cycle, it is possible to reduce the risk of a thick oxide skin in the cast part and its potential leakage. Large shrinkage pores which can be formed due to many process parameters are a different issue. In the present paper, we focused only on the mold temperature. By the thermovision camera, we observed there are changes in mold temperatures up to 20 °C at characteristic locations. With a casting simulation, we simulated a change in mold temperature to examine the influence on solidification. We observed a change in the solidification front that can have an impact on the formation of shrinkage porosity. For this reason, it is important to control the mold temperature at tilt gravity casting although this represents a major challenge in an industrial environment. [1] Eric Riedel, Philipp Köhler, Mostafa Ahmed, Benjamin Hellmann, Ingo Horn, Stefan Scharf. Industrial suitable and digitally recordable application of ultrasound for the enviromentally friendly degassing of aluminium melts before tilt casting. Procedia CIRP, V 98, P: 589–594, 2021. [2] Daniel Molnar, Csaba Majoros, Laszlo Varga. Control volume simulation of gravity die casting. MultiScience – XXXI. MicroCad International Multidisciplinary Scientific Conference. April 2017. [3] George-Christopher Vosniakos, Titos Giannakakis, S. Mylhäuser. Intelligent systems for evaluation of gravity casting quality. International conference on virtual engineering applications for design and product development, Dublin 2003, Ireland. [4] Xinjin Cao, John Campbell. Oxide inclusion defects in Al-Si-Mg cast alloys. The Canadian Journal of Metallurgy and Materials Science. V: 44, I: 4, 2005. [5] Longfei Li, Daquan Li, Junzhen Gao, Yongzhong Zhang, Yonglin Kang. Influence of mold temperature on microstructure and shrinkage porosity of the A357 alloys in gravity die casting. Advances in Materials Processing. 2018. Peter Kirbiš1,2, Ivan Anžel2, Mihael Bruncko1 1SIJ Metal Ravne d.o.o. (SI), 2Univerza v Mariboru, fakulteta za strojništvo (SI) / University of Maribor, Faculty of Mechanical Engineering (SI) Kontinuirno litje visokoogljicnega nanostrukturnega bainitnega jekla Continuous Casting of High Carbon Nanostructured Bainitic Steel Povzetek V prispevku smo obravnavali strjevanje in razvoj mikrostrukture med navpicnim kontinuirnim litjem v laboratorijskem okolju. Izbrana zlitina je visokoogljicno nanostrukturno bainitno jeklo s kemijsko sestavo (0,7C-5,5Mn-1Cr-1,5Al-0,6Mo). Nanostrukturna bainitna jekla dosegajo izjemne kombinacije mehanskih lastnosti, ki hkrati povezujejo visoko trdnost in žilavost. To omogoca izjemno fina mikrostruktura, kjer znaša debelina plošc bainitnega ferita zgolj okrog 10 nm. Fina mikrostruktura se doseže z baininto transformacijo jekla pri zelo nizkih temperaturah od 200 °C pa vse do sobne temperature. Da bi zagotovili tako nizke transformacijske temperature, je treba znižati temperaturo zacetka tvorbe martenzita Ms pod sobno raven. Posledicno imajo ta jekla visoko vsebnost ogljika in so hkrati legirana z Mn, Cr, Ni in Mo. Dodatno se za zakasnitev oz. zavrtje izlocanja karbidov tem jeklom posamezno ali v kombinaciji dodajajo visoke kolicine Si in Al. Zaradi kompleksne kemijske sestave predstavljajo ta jekla tehnološki izziv za izdelavo s tehnologijo kontinuirnega litja. Trenutna preiskava je pokazala, da je novo razvito jeklo primerno za izdelavo s kontinuirnim litjem, kadar je mogoce zagotoviti stabilne razmere strjevanja. Kljuce besede: nanostrukturno bainitno jeklo, kontinuirno litje, segregacije Abstract The current research work describes the solidification and microstructure formation during vertical continuous casting of a high carbon nanostructuredbainitic steel (0.7C-5.5Mn-1Cr­1.5Al-0.6Mo) in laboratory conditions. Nanostructured bainitic steels achieve exceptional combinations of mechanical properties due to their very fine microstructure where the bainitic ferrite plate thickness is only about 10 nm. This very fine structure is obtained by bainite formation at very low temperatures below 200 °C and down to ambient level. To ensure such low transformation temperatures and suppress the formation of martensite below room temperature these steels have a high carbon content and are also alloyed with Mn, Cr, Ni, and Mo. Additionally, in order to suppress the precipitation of carbides during bainite formation, they contain high amounts of Si and Al either separately or in combination. Such a complex composition is challenging from the viewpoint of production using continuous casting. It was observed that the newly developed steel can be successfully continuously cast, provided stable solidification conditions can be maintained during the casting process. Keywords: Nanostructured bainitic steel, continuous casting, segregations 1 Uvod Na letni ravni se s tehnologijo kontinuirnega litja izdela 1,2 milijarde ton jekla. Povecano povpraševanje po naprednih materialih zahteva nenehne izboljšave in dvig produktivnosti. Zagotavljanje potreb trga posledicno vodi v razlicne izzive za proizvajalce in kontinuirno litje jekel z vedno bolj zahtevnimi kemijskimi sestavami [1]. Med ta jekla sodijo tudi visoko trdnostna jekla s transformacijsko podprto plasticnostjo (TRIP); te vrste jekel vsebujejo visok delež Al za stabilizacijo dolocenega deleža zadržanega avstenita [2]. S povišanjem vsebnosti ogljika in drugih legirnih elementov se zniža temperatura martenzitne premene pod sobno temperaturo in tako omogoci tvorbo nizkotemperaturnega bainita. Kadar poteka tvorba bainita pri temperaturah pod 250 °C, se debelina plošc bainitnega ferita giblje v nano obmocju. Visoka vsebnost Al povzroci povišanje viskoznosti jekla, kar otežuje litje in lahko privede do neenakomerne površine ali celo površinskih napak [3], zato je potrebna visoka hitrost litja. Hkrati pa visoka vsebnost ogljika zniža toplotno prevodnost jekla [4], kar praviloma zahteva nižjo hitrost litja, da se tvori zadostno debela skorja. Višja vsebnost ogljika poviša tudi tendenco jekla k tvorbi mikro segregacij [5] in hkrati poviša trdoto po litju kot tudi obcutljivost jekla na zarezni ucinek. Razpoke, nastale iz površinskih napak, tako lažje napredujejo v notranjost. Namen trenutnega prispevka je ovrednotenje potenciala za nova visokoogljicna nanostrukturnabainitna jekla z visoko vsebnostjo aluminija. 2 Materiali in tehnologije Jeklo s kemijsko sestavo 0,7C-5,5Mn-1,5Cr­ 1,5Al-0,6Mo je bilo vertikalno kontinuirno lito na laboratorijski napravi. Visoka vsebnost 1 Introduction Each year about 1.2 billion tons of steel are produced into semi-finished shapes using a continuous casting process. The demand for producing high-performance steel has increased which requires manufacturing via continuous casting to increase productivity and reduce production costs. These factors force steel producers to face different challenges to meet the customers’demands by producing steels of high quality with ever more demanding chemical compositions [1]. One such steel group is high strength transformation induced plasticity steels (TRIP), grades contain high Al to achieve the desired amount of retained austenite [2]. At increasing carbon content the transformation temperatures are decreased allowing for the formation of bainite at low temperatures. When bainite is formed at temperatures below 250 °C the plate thickness is in the nanostructured range. The high Al content is known to increase the steels viscosity during pouring thereby leading to the formation of surface depressions [3] therefore requiring a high casting speed. Whereas the high carbon content decreases the steel thermal conductivity [4] and it is, therefore, necessary to cast slowly to obtain a sufficiently thick outer shell. Additionally, a higher carbon content makes the steel more susceptible to micro and macro segregations [5] as well as increases the cast hardness and notch sensitivity of the cast billet thereby cracks can easily propagate from depressed regions. The current work aims to evaluate the potential for successful continuous casting of a novel high carbon, high aluminium TRIP steel. Al poviša viskoznost taline, medtem ko je po strjevanju jeklo nagnjeno k nastanku razpok zaradi visoke vsebnosti ogljika, kar predstavlja izziv za uspešno izvedbo kontinuirnega litja. Poskus je bil izveden na napravi za laboratorijsko vertikalno kontinuirno litje Technica-Guss GMBH 30 E. Sklop sestoji iz okrogle šobe premera 10 mm, kokile, vhodne šobe, vodno hlajene bakrene kokile in potisne palice, kot je shematsko prikazano na Sliki 1. Jeklo je bilo indukcijsko pretaljeno v vakuumu, pred litjem je bila komora napolnjena z argonom, in sicer za zmanjšanje poroznosti. Pred litjem je bila temperatura taline 1560 °C. Postopek kontinuirnega litja poteka po korakih, pri cemer se potisna palica odmakne za doloceno razdaljo-korak, kar talini omogoca, da stece v šobo in vodno hlajeno kokilo, kjer se zadrži za dolocen cas in pri tem strdi. Kljucni parametri hitrosti litja so dolžina koraka in cas zadrževanja, saj dolocajo stabilnost procesa in kakovost površine lite gredice. Stabilni parametri so bili doseženi pri dolžini koraka 2 mm in casu zadrževanja 6 s. Iz gredice smo izrezali vzorce v vzdolžni in precni smeri vzdolž srednje ravnine gredice. Vzorci so bili zaliti v bakelitno maso in metalografsko pripravljeni ter jedkani z barvnim jedkalom LePera. Mehanske lastnosti gredice v litem stanju so bile dolocene z nateznim preskusom v vzdolžni smeri. Rezultati in diskusija Znotraj stabilnih parametrov litja je strjena mikrostruktura, sestavljena iz globulitne cone na robu, ki preide v usmerjeno dendritno strjevanje proti sredini palice. Vidni so pasovi pozitivne in negativne segregacije, kot prikazujeta Slika 2 na precnem prerezu in Slika 3 na vzdolžnem prerezu. Kot prikazuje Slika 2, je mikrostruktura znotraj pozitivno segregiranih obmocij sestavljena 2 Materials and methods The steel of composition 0.7C-5.5Mn-1.5Cr­1.5Al-0.6Mo, was continuously cast in a laboratory setup. Due to the high Al content, the steel has reduced fluidity whereas the high carbon content increases the as cast hardness and the tendency for cracking, making this steel challenging for the continuous casting process. The laboratory setup on continuous casting machine Technica-Guss GMBH 30 E, consists of a crucible, inlet nozzle, water-cooled Cu die, and push rod as shown schematically in Fig. 1. The steel was induction melted in vacuum, before casting the chamber was filled with pure argon to reduce porosity. Before casting the melt temperature was 1560 °C. The continuous casting process is stepwise whereby the pushrod is withdrawn a certain length (step) allowing molten steel to flow into the inlet nozzle and water cooled crucible where it is held for a period of time to solidify. The parameters of holding time casting speed and step length are the most crucial in determining the stability of the cast billet. The casting step was 2 mm with a 6 s holding time. Samples for metallography were taken in the transverse and longitudinal direction along the midsection of the bar, mounted in resin and metallographically prepared, followed by tint etching using LePera reagent. The as cast ductility was determined using tensile testing in the longitudinal direction. 3 Results and Discussion Within stable casting parameters, the solidification structure consists of positive and negative segregation bands visible in both the transverse and longitudinal directions in Fig. 2 and Fig. 3. Within regions of positive segregation an austenitic dendritic structure forms, poroznost v osrednjem delu gredice, kar je bilo pricakovati glede na slabo fluidnost taline. Smer strjevanja je pod kotom, kot je razvidno iz Slike 3. To nakazuje, da se gredica pricne strjevati na sticni površini s kokilo, medtem ko je osrednji del segret zaradi stika z zgornjo talino, ki bo ulita v naslednjem koraku. Na površini gredice so prisotne vdolbine, ki so deloma zalite in zato vidne Additionally, Fig. 2 shows some centreline porosity, which is to be expected with this production process in steel with low fluidity. The solidification direction is at an angle as clearly visible in Fig. 3, this occurs as the steel solidifies first at the die surface whereas the central region is heated by the molten metal cast in the next step. As can be expected depressions and subsurface porosity occur at the junction Slika 4. Površinske napake in podpovršinska poroznost, (jedkano z reagentom Le Perra) Figure 4. Surface depression and subsurface porosity, etched with Le Perra reagent Slika 5. Trakasta mikrostruktura v notranjosti kontinuirno lite gredice. Martenzit je obarvan temno, dendriti avstenita pa rumeno/modro (jedkano z Le Perra reagentom) Figure 5. Banded microstructure within the continuously cast bar, Positive segregation bands contain martensite (dark) and negative segregation bands consist of austenitic dendrite structure (yellow), etched with Le Perra reagent kot podpovršinska poroznost kot prikazuje Slika 4. Ta pojav je posledica slabe fluidnosti taline, je pa znano, da se lahko v praksi v veliki meri prepreci z uporabo ustreznih livnih praškov [6]–[8]. To pomeni, da se kljub visoki vsebnosti Al lahko izdelajo gredice z ustrezno površino za nadaljnje valjanje. Notranjost kontinuirno lite gredice ima trakasto mikrostrukturo, sestavljeno iz martenzita, ki se tvori znotraj obmocij negativne segregacije, na Sliki 5 je obarvana temno. Predpostavlja se, da je prišlo do segregacije predvsem C in Mn. Izracunana temperatura zacetka martenzitne premene Ms za to jeklo znaša -58 °C, kar je pod sobno temperaturo, zato je pricakovati moc avstenitno/bainitno mikrostrukturo. Jekla, ki imajo srednje vsebnosti Mn (okoli 5 %) in visok delež ogljika, so podvržena tvorbi izrazito trakaste mikrostrukture [9], ki je posledica hkratne segregacije C in Mn [10]. Ta elementa imata visoko medsebojno afiniteto in sta visoko mobilna med potekom strjevanja. Kljub visoki vsebnosti C je trdota jekla v litem stanju samo 52 HRC. Na liti gredici so bile dolocene mehanske lastnosti z nateznim preskusom, kjer so bili izmerjeni natezna trdnost 1476 MPa in raztezek A5 13 % ter kontrakcija Z 32 %. Deformacija je najverjetneje lokalizirana znotraj avstenitnih obmocij, visok delež avstenita in deloma tudi poroznost pa sta glavna vzroka relativno nizke trdote, še posebej pa nizke trdnosti v primerjavi s trdoto. Zakljucki Novo razvito jeklo je primerno za proizvodnjo s tehnologijo vertikalnega kontinuirnega litja, pri cemer je treba zagotoviti stabilne razmere litja. Segregacije in osrednja poroznost so znotraj pricakovanih okvirjev za takšno vrsto jekla in se lahko v veliki meri odpravijo point of the stepwise casting process as shown in Fig. 4. However, these have been shown to be successfully mitigated using appropriate mold powders[6]–[8] resulting in billets of good surface quality for further processing by rolling. The interior of the continuously cast bar exhibits a banded microstructure with martensite forming within bands of negative segregation as these contain less Mn and C. The calculated Martensite start (Ms) temperature for this steel composition is -58 °C, which is below room temperature therefore in the absence of segregation an austenitic/bainitic microstructure could be expected. Steels containing medium manganese and high carbon content are known to be prone to banding [9], due to co-segregation of Mn and C [10], which have a high affinity towards each other and are both highly mobile during solidification. Despite the high carbon content the as cast microstructure achieves a hardness of 52 HRC, During tensile testing we obtained a UTS of 1476 MPa and an elongation A5 13% and 32% contraction. The discrepancy between hardness and obtained UTS is likely due to casting defects most notably porosity. 4 Conclusions The steel can be successfully vertically continuously cast provided stable conditions can be maintained, segregation and porosity content are within expected regions, and likely to homogenize during high-temperature annealing treatment before hot rolling. No coarse precipitates or large inclusions were observed within the microstructure. It would seem that the surface depressions cannot be entirely mitigated using vertical continuous casting z visokotemperaturno homogenizacijo pred nadaljnjo plasticno predelavo. Znotraj mikrostrukture gredice ni bilo prisotnih grobih izlockov ali velikih nekovinskih vkljuckov. Zdi se, da površinskih napak brez uporabe livnih praškov ni mogoce prepreciti med vertikalnim kontinuirnim litjem. Znotraj obmocij negativne segregacije se tvori martenzit, ki pa ne zniža celotne duktilnosti gredice, saj deformacija potece predvsem znotraj avstenitnih obmocij, nastalih znotraj obmocij pozitivne segregacije. Viri / References without applying a suitable mold powder, the formation of martensite within negative segregation bands does not impair the overall ductility of the continuous cast bar due to the high austenite content within positive segregation bands which provide sufficient overall plasticity. [1] E. Commission, Continuous casting of high-carbon steels in billet and bloom sections at sub-liquidus temperatures. [2] B. C. De Cooman, “Structure-properties relationship in TRIPsteels containing carbide- free bainite,” Curr. Opin. Solid State Mater. Sci., vol. 8, no. 3–4, pp. 285–303, 2004, doi: 10.1016/j.cossms.2004.10.002. [3] H. Cui et al., “Formation of surface depression during continuous casting of high-Al TRIP steel,” Metals (Basel)., vol. 9, no. 2, pp. 1–9, 2019, doi: 10.3390/met9020204. [4] K. Kinoshita, Y. Yoshii, H. Kitaoka, A. Kawaharada, K. Nishikawa, and O. Tanigawa, “Continuous Casting of High Alloy Steels.,” Metall. Soc. AIME, TMS Pap. Sel., vol. 60, pp. 429–444, 1983. [5] A. Nicholas Grundy, S. Münch, S. Feldhaus, and J. Bratberg, “Continuous Casting of High Carbon Steel: How Does Hard Cooling Influence Solidification, Micro - And Macro Segregation?,” IOP Conf. Ser. Mater. Sci. Eng., vol. 529, no. 1, 2019, doi: 10.1088/1757-899X/529/1/012069. [6] W. Yan, A. McLean, Y. Yang, W. Chen, and M. Barati, “Evaluation of mold flux for continuous casting of high-aluminum steel,” Adv. Molten Slags, Fluxes, Salts Proc. 10th Int. Conf. Molten Slags, Fluxes Salts 2016, pp. 279–289, 2017, doi: 10.1007/978­ 3-319-48769-4_30. [7] K. Zhang, J. Liu, and H. Cui, “Investigation on the slag-steel reaction of mold fluxes used for casting al-trip steel,” Metals (Basel)., vol. 9, no. 4, 2019, doi: 10.3390/met9040398. [8] H. Steel, “Advances in Molten Slags, Fluxes, and Salts: Proceedings of the 10th International Conference on Molten Slags, Fluxes and Salts 2016,” Adv. Molten Slags, Fluxes, Salts Proc. 10th Int. Conf. Molten Slags, Fluxes Salts 2016, no. January, 2016, doi: 10.1007/978-3-319-48769-4. [9] S. a. Khan and H. K. D. H. Bhadeshia, “The bainite transformation in chemically heterogeneous 300M high-strength steel,” Metall. Trans. A, vol. 21, no. 3, pp. 859–875, 1990, doi: 10.1007/BF02656570. [10]Y. N. Dastur and W. C. Leslie, “Mechanism of work hardening in Hadfield manganese steel,” Metall. Trans. A, vol. 12, no. 5, pp. 749–759, 1981, doi: 10.1007/BF02648339. AKTUALNO / CURRENT Pregled svetovne livarske proizvodnje v letu 2020 Porocevalci livarske revije “Modern Casting” porocajo, da je “pandemija covid-19 imela obcuten vpliv na livarsko proizvodnjo, glede na to, da so skoraj vse države – razen Kitajske, porocale o padcu proizvodnje.” V tem prispevku povzemamo tabelarne podatke in komentar iz decembrske številke ameriške revije “Modern Casting” iz leta 2021, in sicer iz clanka “Fewer Castings Made in 2020” (Padec livarske proizvodnje 2020). Opozarjamo, da podatki za slovensko livarsko industrijo niso povsem tocni, ker so, iz nam neznanih razlogov, izloceni podatki za temprano litino. Dejanska skupna livarska proizvodnja za Slovenijo je znašala v letu 2020 172.840 ton. Znotraj tega zneska je tudi livarska proizvodnja temprane litine v višini 3.100 ton. V letu 2020 so države po celem svetu obcutile ucinke vladnih zapor in restrikcij pri omejitvah dela zaradi pandemije covid-19. Skoraj vsaka država je za leto 2020 porocala o obcutnem padcu proizvodnje, v primerjavi z letom 2019. Ena od redkih izjem je bila Kitajska, ki je vpliv covida obcutila prej kot ostale države, vendar je potem relativno hitro dosegla kar 6% rast proizvodnje. Korejska livarska proizvodnja je ostala na približno enakem nivoju, brez padca ali rasti. Nova v popisu livarske proizvodnje je letos Indonezija Glede na porocanja Indonezijskega združenja za livarsko industrijo (APLINDO), je država proizvedla 589,779 ton ulitkov, kjer je litje aluminija v ospredju. Vodilnih 10 držav po livarski proizvodnji je prikazanih v priloženi tabeli. Zopet, Kitajska, Indija in ZDAzavzemajo prva tri mesta. V pregledu celotne svetovne proizvodnje je razvidno, da je ta v letu 2020 znašala 105.505.602 ton, kar je 3% manj kot v letu 2019. Podatke, ki so navedeni v pregledu livarske proizvodnje, so zagotovila livarska društva ali sorodni predstavniki te industrije vsake posamezne države, prav tako pa tudi WFO – Svetovna livarska organizacija in CAEF – Evropsko združenje livarn. Tabela 1. Najvecjih deset držav po livarski proizvodnji v letu 2019 Najvecjih 10 proizvajalcev ulitkov Država Proizvodnja v tonah Gibanje na leto 2019 Kitajska 51.950.000 6,00% Indija 11.314.360 -1,50% ZDA 9.748.811 -13,7% Rusija 4.200.000 ni novih podatkov Nemcija 3.482.883 -29,60% Japonska 3.446.903 -34,7% Mehika 2.855.650 ni novih podatkov Južna Koreja 2.380.200 0,00% Turcija 2.170.759 -6,0% Brazilija 2.073.173 -9,40% Proizvodnja ulitkov v 2020 (v tonah) Država Siva litina Nodularna litina Temprana litina Jeklo litina Baker Aluminij Magnezij Cink Druge neželezne Skupno Avstrija 33.400 91.700A - 9.600 - 106.798 4.504 - - 246.002 Belgija 43.000 3.900A - 5.500 - - - - - 52.400 Brazilija 1.148.123 468.952 - 269.512 20.524 160.464 4.534 1.064 - 2.073.173 KanadaB 330.841 - - 90.091 14.237 211.374 - - - 646.543 Kitajska 21.750.000 15.300.00 630.000 6.350.000 870.000 6.800.000 - - 250.000 51.950.000 Hrvaška 19.465 6.161 - 120 202 65.606 - - 131 91.685 Ceška 117.000 34.500A - 41.000 16.000 77.400 300 800 - 287.000 DanskaC 28.900 58.100A - - 1.188 2.224 - - 112 90.524 Finska 17.300 23.100A - 6.700 2.415 1.730 - - - 51.245 Francija 431.900 593.600A - 41.900 16.118 293.529 - 18.880 2.180 1.398.107 Nemcija 1.618.700 957.100A - 138.000 46.076 652.738 20.489 49.761 19 3.482.883 Madžarska 16.500 58.000A - 2.000 729 118.900 250 1.662 99 198.140 Indija 7.911.763 1.095.522 50.000 912.893 - 1.344.182 - - - 11.314.360 Indonezija 77.783 42.060 86.940 128.724 46.086 182.586 - - 25.600 589.779 Italija 534.400 300.600A - 58.000 38.168 540.296 3.676 75.834 1.235 1.552.209 Japonska 1.598.113 1.169.743 29.439 153.000 57.019 343.651 - 13.792 82.146G 3.446.903 Južna Koreja 881.400 670.700 - 150.000 24.100 642.200 12.000D - - 2.380.400 MehikaE 816.160 560.270 - 336.250 215.500 832.770 - 79.500 15.200 2.855.650 NorveškaC 8.800 22.300A - - - 6.526 - - - 37.626 PakistanC 181.000 24.540 - 48.750 14.200 21.200 - - 2.730 292.420 Poljska 360.000 124.000A - 40.000 4.800 272.000 - 6.000 2.400 809.200 Portugalska 26.100 76.100A - 4.100 16.203 31.966 - 2.165 - 156.634 RomunijaC 15.000 1.500 - 3.500 3.000 60.000 2.000 250 90 85.340 RusijaC 2.184.000E - - 1.134.000 117.600 588.000 75.600 - 100.800 4.200.000 Slovenija 59.300 39.800A - 17.600 990 44.610 - 7.477 - 169.785 Južna AfrikaC - - - - - - - - - 443.000 Španija 283.100 582.800A - 65.300 15.279 101.317 - 7.304 683 1.055.783 Švedska 126.000 51.000A - 20.200 - 56.400 - - - 253.600 Švica 8.400 11.900A - 2.500 2.023 10.815 - 762 - 36.400 Tajvan 543.617 179.697 - 55.007 27.368 481.593 6.237 - - 1.293.519 Turcija 617.300 854.700A - 192.000 24.851 449.503 761 31.644 - 2.170.759 UkrajinaB - - - - - - - - - 1.560.000 Velika Britanija 128.400 195.600A - 41.600 8.300 102.522 2.000 7.300 - 489.722 ZDA 7.616.824 - - - 304.279 1.425.120 - 47.786 354.802 9.748.811 Skupna svetovna proizvodnja 49.532.589 23.597.945 796.379 10.317.847 1.907.255 16.028.028 132.351 352.773 838.27 105.505.602 A - vsebuje temprano litino B - podatki iz 2015 C - podatki iz 2019 D - vsebuje magnezij E - podatki iz 2017 G - vsebuje aluminijevo tlacno litino Tabela 2. Pregled svetovne livarske proizvodnje v letu 2020 Vir: Podatki in tabeli iz Ameriške livarske revije Modern Casting. mag. Mirjam Jan-Blažic AKTUALNO / CURRENT Seje organov Društva livarjev Slovenije Društvo livarjev Slovenije vsako leto, v mesecu februarju, na isti dan sklicuje svoj redni Obcni zbor, Izvršni odbor ter Nadzorni odbor. Tako je bilo tudi letos, s tem, da so vse tri seje organov morale zaradi še vedno neustrezne epidemiološke situacije potekati po video konferencah, in sicer dne 17. 02. 2022. Najprej je zasedal Nadzorni odbor, ki je obravnaval porocilo o delu in financnem poslovanju Društva livarjev Slovenije v letu 2021. Sprejel je sklep, da je financno poslovanje Društva v letu 2021 vodeno v skladu z veljavnimi racunovodskimi standardi za društva. Poraba sredstev je potekala v skladu s programom dela Društva za leto 2021, ki je bil sprejet na rednem Obcnem zboru dne 25. 02. 2021. Po koncani seji Nadzornega odbora se je sestal Izvršni odbor Društva, potem pa še Obcni zbor. Osrednji tocki obeh organov Društva sta bili: 1. Porocilo predsednice o delu in financnem poslovanju Društva za leto 2021 2. Program dela Društva za leto 2022 Vsi clani Izvršnega odbora in delegati Obcnega zbora so s sklicem seje prejeli pisna porocila predsednice o delu in financnem poslovanju Društva za leto 2021 in dopolnjen program dela Društva za leto 2022. Iz porocila predsednice o delu in financnem poslovanju Društva za leto 2021 je razvidno, da so vse osrednje programske tocke iz programa Društva za leto 2021 bile realizirane, razen programa sodelovanja in udeležb z drugimi društvi in livarskimi organizacijami v tujini. Izpad tega dela programa je nastal zaradi pandemije korona-virusa. Velika vecina seminarjev je bila izvedena on-line zaradi pandemije korona-virusa. Za razliko od leta 2020, ko je nastopil prvi val pandemije, je bila izvedba vseh nacrtovanih seminarjev v celoti realizirana. V nadaljevanju so navedeni vsi izvedeni seminarji v letu 2021, z navedbo clanic Društva, ki so se seminarja udeležile. Seminarji za železove livarne • Dne 11. 05 in 12. 05. 2021, dvodnevni video seminar v organizaciji izvajalca ÖGI – Avstrijskega livarskega Inštituta iz Leobna, pod naslovom »Cast iron and casting defects«. Udeležba 12 slušateljev iz podjetij: EXOTERM-IT d.o.o., KOVIS-LIVARNA d.o.o., OMCO METALS SLOVENIA d.o.o., ETA d.o.o. CERKNO, CIMOS d.d., VALJI d.o.o., Katedra za livarstvo NTF, LIVARNA GORICA d.o.o. • Dne 20. 10. 2021, enodnevni video seminar v organizaciji Katedre za livarstvo Naravoslovno-tehnicne fakultete, pod naslovom »Strjevanje sivih litin in napake povezane s strjevanjem«. Udeležba 9 slušateljev iz podjetij: CIMOS d.d., LIVARNA VUZEMICA, ETA d.o.o. CERKNO, KOVIS-LIVARNAd.o.o., LIVARNAGORICA d.o.o., OMCO METALS SLOVENIA d.o.o. • Dne 10. 06. 2021, enodnevni video seminar v izvedbi GZS, Službe za varstvo okolja, pod naslovom »Okoljevarstvena problematika za livarne«. Udeležba 12 slušateljev iz podjetij CIMOS d.d., DIFA d.d., HIDRIA d.o.o., ISKRA ISD –STORITVE d.o.o., LTH CASTINGS d.o.o., MAHLE ELECTRO DRIVES KOMEN d.o.o., MARIBORSKA LIVARNA MARIBOR d.d., TALUM d.d., TELKOM d.d. Seminarji za neželezove livarne • Dne 05. 05. in 06. 05. 2021, dvodnevni video seminar izvajalca ÖGI – Avstrijskega livarskega Inštituta iz Leobna, pod naslovom: »High pressure die castings«. Udeležba 12 slušateljev iz podjetij CIMOS d.d., DIFAd.d., HIDRIAd.o.o., ISKRAISD – STORITVE d.o.o., LTH CASTINGS d.o.o., MAHLE ELECTRO DRIVES KOMEN d.o.o., MARIBORSKA LIVARNA MARIBOR d.d., TALUM d.d., TELKOM d.d. • Dne 15. 09. 2021 v organizaciji RWP GmbH iz Nemcije, seminar pod naslovom: »Electro mobility«. Udeležba 17 slušateljev iz podjetij: CIMOS d.d., HIDRIA d.o.o., ITOCHU EUROPE PLC, LTH CASTINGS d.o.o., MARIBORSKA LIVARNA MARIBOR d.d., TALUM d.d. • Dne 26. 10. 2021 v organizaciji Katedre za livarstvo Naravoslovno-tehnicne fakultete, seminar na temo »Strjevanje livarskih aluminijevih zlitin, razvoj mikrostrukture in povezane napake«. Udeležba 11 slušateljev iz podjetij: DIFA d.o.o., HIDRIA d.o.o., LTH CASTINGS d.o.o., MAHLE ELECTRIC DRIVES KOMEN d.o.o., MARIBORSKA LIVARNA MARIBOR d.d., TALUM d.d. Osrednja livarska prireditev v Portorožu – 61. IFC Portorož 2021 pa je z udeležbo presegla tisto iz leta 2020. O tem osrednjem livarskem dogodku Društva v letu 2021 smo že podrobno porocali v Livarskem vestniku št. 3/2021 in št. 4/2021. Iz porocila o financnem poslovanju Društva livarjev Slovenije v letu 2021 je razvidno, da je Društvo ustvarilo skupne prihodke v višini 102.734,72 €. Delež prihodkov iz naslova clanarine je znašal samo 22 % ali 22.727 €. Zaradi omejenih clanskih prispevkov in manjšega obsega pokroviteljskih sredstev ter vecjih stroškov, ki jih je zahteval covid-19, so stroški v letu 2021 presegli prihodke za 3.023,29 €. Porocilo predsednice o delu in financnem poslovanju Društva v letu 2021 je Obcni zbor soglasno sprejel. V programu dela Društva za letošnje leto so predvideni naslednji seminarji: Seminarji za železove livarne • Dvodnevni nadaljevalni seminar izvajalca ÖGI-Livarskega Inštituta v Leobnu, v Avstriji, na njegovi lokaciji, pod naslovom »Cast iron and casting defects«. Gre za nadaljevalni seminar, ki je bil s strani Inštituta izveden lansko leto preko video konference. Komisija je ocenila, da je potrebno temo še bolj poglobljeno predstaviti slušateljem z vkljucitvijo prakticnega dela v preizkusnih laboratorijih ÖGI. Rok izvedbe seminarja je 28. 06. in 29. 06. 2022. Število udeležencev je omejeno na obicajnih 12. • Seminar na temo »Odstranjevanje fluoridov iz odpadlih livarskih peskov« v izvedbi podjetja GTP Schaefer in Exoterm-IT d.o.o ter po možnosti tudi drugih izvajalcev. Ta seminar je že bil nacrtovan v letu 2020, vendar pa ni bil izveden zaradi pandemije koronavirusa. Društvo išce drugega kompetentnega izvajalca seminarja, ki bi širše pokril tematiko odpadnih livarskih peskov. • Seminar na temo okoljevarstvene problematike za livarne v izvedbi Gospodarske zbornice Slovenije. Rok izvedbe je v drugi polovici junija. Število udeležencev ni omejeno. Seminarji za neželezove livarne • Dvodnevni nadaljevalni seminar izvajalca ÖGI-Livarskega Inštituta v Avstriji, na njegovi lokaciji, na temo »Tehnologije mazanja in vplivi mazanja na cistost vlitkov, površinsko napetost in efekte v nadaljnji uporabi vlitkov«. Ta tematika je že bila obravnavana na video-seminarju lansko leto, in se bo, v Leobnu, nadgradila s prakticnim delom v preizkusnem obratu za tlacno litje in laboratorijih Inštituta. Predlagano je, da se program prilagodi predvideni strukturi slušateljev in predlogom livarn clanic Društva. Rok izvedbe seminarja je 11. 05. in 12. 05. 2022. Število slušateljev je omejeno na obicajnih 12. • Trodnevni seminar za livarne tlacnega litja, izvajalca Buehler, Švica, na lokaciji firme Buehler v Uzwilu. Gre za seminar, ki je namenjen predvsem skupini visoko usposobljenih, že izkušenih livarskih strokovnjakov. Rok za izvedbo seminarja je v 43. koledarskem tednu, število udeležencev pa je omejeno na 10. • Enodnevni seminar izvajalca RWP GmbH Nemcija (Dr. Konrad Weiss), na temo »Livni materiali, toplotno obdelavo in nacini za doseganje zahtevanih mehanskih lastnosti«. Seminar bo izveden 14. 09. 2022 - to je en dan pred uradnimpricetkom 62. IFC Portorož 2022. Število udeležencev je omejeno na 12-15. • Enodnevni video ali seminar v živo v izvedbi Katedre za livarstvo Naravoslovnotehniške fakultete Univerze v Ljubljani, na temo »Vpliv necistoc v ingotih in krožnem materialu na kakovost izdelkov in mehanske lastnosti«. Število udeležencev na seminarju ni omejeno. Termin izvedbe bo s Katedro za livarstvo dolocen naknadno, predvidoma pa v jesenskem casu. • Vsakoletni seminar za vse vrste livarn na temo »Okoljevarstvena problematika za livarne« v izvedbi Gospodarske zbornice Slovenije, Službe za varstvo okolja. Rok izvedbe je v drugi polovici junija. Število udeležencev ni omejeno. Na predlagane dopolnitve programa dela Društva za leto 2022 ni bilo pripomb, zato je Obcni zbor soglasno sprejel sklep, da se dopolnjen program dela Društva livarjev Slovenije za leto 2022 sprejme. Na organih Društva je bil sprejet tudi sklep o minimalnem clanskem prispevku za Društvo v letu 2021, do katerega so upraviceni clani Društva, ki so v prisilni poravnavi, kot tudi sklep o clanskem prispevku za ostale clane Društva, ki se zaradi narašcajocih stroškov in inflacije zvišujejo pri vseh clanicah za 100 €. mag. Mirjam Jan-Blažic AKTUALNO / CURRENT STEM D.O.O. bogatejši za nove proizvodne prostore Nova proizvodna hala STEM d.o.o. Clan Društva livarjev Slovenije, STEM d.o.o. iz Nove Gorice, ki je eden izmed najvecjih proizvajalcev peskalnih strojev v Sloveniji ter eden izmed vodilnih proizvajalcev peskalnih strojev v Evropi, nas je obvestil, da je zaradi rasti poslovanja razširil svoje proizvodno­razvojne zmogljivosti z novimi, najsodobnejše opremljenimi prostori, ki vkljucujejo tudi testno-razvojni center. Vodja prodaje, g. Igor Bitežnik še dodaja, da ta vecmilijonska naložba pomeni dodatno zagotovilo za nemoteno nadaljnjo rast in širitev podjetja na obstojecih in novih trgih. AKTUALNO / CURRENT Pregled livarskih prireditev v 2022 Datum dogodka Ime dogodka Mesto in država 28. - 29.04. 2022 64. avstrijska livarska konferenca Leoben, Avstrija 19. - 23.06. 2022 6. konferenca »Steels in Cars und Trucks« Milano, Italija 21. - 23.06. 2022 CastForge Stuttgart, Nemcija 22. - 23.06. 2022 1. Forum litega železa poteka v okviru CastForge Stuttgart, Nemcija 14. - 16.09. 2022 62. IFC Portorož 2022 Portorož, Slovenija 5. - 7.10. 2022 Konferenca litja cinkove Matrice – Evropa Koblenz, Nemcija 16. - 20.10. 2022 74. mednarodni livarski kongres Busan, Korea Naslov / Address: KS Kneissl & Senn Technologie GmbHMühlgraben 43b A-6343 Erl Direktor: Mag. Franz Senn T: +43(0)5373/76020-0F: +43(0)5373/76020-20 E: info@ks-tech.athttps://ks-tech.at POPRAVEK / ERRATA V adremi Livarskega vestnika št. 4/2021 s podatki za podjetje KS Kneissl & Senn Technologie GmbH je naveden napacen naslov spletne strani: www.calderys. com. Pravilen spletni naslov glasi: https://ks-tech.at/. Za neljubo napako se iskreno opravicujemo podjetju KS Kneissl & Senn Technologie GmbH ter bralcem Livarskega vestnika. In No. 4/2021 of the Foundry Journal, containing information about the company KS Kneissl & Senn Technologie GmbH, a wrong website address was stated: www. calderys.com/. The correct website address is: https://ks-tech.at/. We sincerely apologise to the company KS Kneissl & Senn Technologie GmbH and the readers of the Foundry Journal for this unwanted mistake. DRUŠTVO LIVARJEV SLOVENIJE Vabilo za 62. IFC PORTOROŽ 2022 z livarsko razstavo 14. - 16. SEPTEMBER 2022 Kontakt: DRUŠTVO LIVARJEV SLOVENIJE, Lepi pot 6, p.p. 424, 1001 Ljubljana T: +386 1 2522 488, F: +386 1 4269 934 drustvo.livarjev@siol.net, www.drustvo-livarjev.si