Strojniški vestnik - Journal of Mechanical Engineering 54(2008)1, 68-76 UDC 658.562.4:669.71 Paper received: 19.12.2007 Paper accepted: 14.1.2008 Laser Supported Optical Control of High Pressure Aluminium Cast Products Valter Gruden1 - Drago Bracun2 - Janez Mozina2* 'Hidria-Rotomatika Ltd, Slovenia 2University of Ljubljana, Faculty of Mechanical Engineering, Slovenia We present a new method for surface quality control of aluminium high pressure cast products. By this method it is aimed to improve efficiently the existing practice of castings being only visually checked for surface defects such as laminations, non-fills and cold shots. The method is based on laser triangulation principle. The measured cloud of points is analysed using software designed specifically for automatic detection of surface defects. The paper describes a measurement system, measuring procedure focussed on the detection of surface defects and the comparison of the results with a visual inspection. © 2008 Journal of Mechanical Engineering. All rights reserved. Keywords: aluminium pressure casting, surface quality control, laser 3D-measurement, optical control 0 INTRODUCTION Today, the automotive industry requires complete traceability of each part used. This means that data on the exact date of production, production conditions and quality have to be provided for each part made [1] and [2]. The quality of a die-casting is determined by checking the precast final dimensions, casting quality, in particular with regard to porosity and surface defects, and the chemical composition of materials. The existing mechanical methods used for the purpose are the following: mechanical standard measurements, go-no-go gauges and 3D coordinate measuring machines (precast final dimensions), X-ray (porosity) and emission spectrometers (chemical composition of materials) [3] and [4]. In high pressure casting of aluminium, surface defects are relatively common. Normally, these occur on thin casting walls as laminations, non-fills and cold shots [5]. Surface defects may result from incorrect settings of the high pressure die-casting process, and the wear and cracks on die-casting tools. The traditional method of inspecting the quality of a die-casting is that a machine operator visually checks the quality of the product. The method depends on the skill of the operator and is as such highly subjective. The paper proposes a new method for inspecting the surface quality of die castings, which is based on an optical measurement system and appropriate software to allow fast accurate measurement analyses. The new method shall be demonstrated on a cast product built for automobiles as an electronic component support (Fig. 1). The support casting is made from aluminium alloy AlSi12, and the casting dimensions are 190 x 150 x 10 mm with a minimum mean wall thickness of 1.5 mm. The inner surface of the casting has to be smooth to allow adhesion of an Fig. 1. Electronic component support made by aluminium high pressure die-casting *Corr. Author's Address: University of Ljubljana, Faculty of Mechanical Engineering, Aškerčeva 6, SI-1000 Ljubljana, 68 Slovenia, janez.mozina@fs.uni-lj.si electronic circuit board. Surface defects (e.g. small humps, slivers) protruding from the surface of the casting may cause short-circuit in the attached electronic circuit board and lead to the assembled set being rejected. According to the specifications, the maximum curvature of a casting shall not exceed 0.4 mm and the size of a surface defect shall not exceed 0.08 mm. Currently, three dimension prototypes are used in assessing casting quality: height, width and flatness. Assessment of surface defects, however, is entirely dependent on the visual capabilities of the assessor. On account of being highly subjective, this method is being replaced with a laser based measurement system which both ensures precise measurements of the casting surface quality and improves quality traceability. High pressure die-casting is a modern casting technique for mass production of complex components characterised by thin walls and good mechanical characteristics. The mentioned die-casting is made on a horizontal pressure die-casting machine with a closing force of 4000 kN with a cold chamber [5]. In pressure die-casting, temperature plays a very important role. It affects the time of solidifying, the quality of molten metal structure and the lifetime of pressure die-casting tools. Thermal cracks, which may occur on the surface of the main tool inserts, reduce the lifetime of the tools. These cracks are caused by high temperature differences (up to 500 °C) between the aggressive aluminium alloy and the cavity surface, alloy velocity at the gate area and the shape of the gate area. The most common surface-visible die-casting defects include cold shots (Fig. 2-a), non-fills (Fig. 2-b) and laminations (Fig. 2-c, d). Cold shots are normally caused by high surface tension between the alloy-and the inserts as well as from long filling time. In such a case, the alloy cools down and starts to solidify before the entire cavity has been fulfilled. Often the same reason also accounts for the occurrence of non-fills. Laminations are most commonly observed in die castings after sand blasting. This surface defect occurs when the top layer peels from the rest of the casting. It is caused by high velocity of the alloy or by a preheated surface of the cavity. The most common defects in the die-casting of electronic component supports occur on the thinnest walls of the casting as cold shots or non-fills (cold die-casting tools). Laminations as another type of surface defect may occur as a result of preheated tools and is seen as humps or blisters appearing on the surface. 1 LASER SUPPORTED OPTICAL MEASUREMENT SYSTEM The measuring system (Fig. 3) consists of an optical measuring instrument using the laser triangulation principle [6] and [7] and a movable table that changes the position of the measured product relative to the instrument. The casting contains precast blind holes which lean against three support balls when mounted onto the movable table, thereby ensuring repeatable mounting of the casting onto the table. The system set-up is made up of a laser projector which illuminates the casting with a thin laser plane, and a CDD camera (Basler 101f). At the point where the laser plane hits the casting, a thin bright line can be seen (intersection curve). The image is captured by the camera positioned at an angle with respect to the direction of illumination (laser triangulation). The process points to the 3D profile of the intersection curve. The image is then translated into a personal computer file, and the intersection curve is extracted and its maximum and minimum values are identified [8] and [9]. The result of one measurement is a profile representing a cross- Fig. 2. A cold shot (a), non-fills (b), laminations (c, d) section of the light plane and the illuminated surface (X,Y)., where X and Y are the coordinates of a particular point, X along the intersection curve and Y perpendicular to the surface; index i runs from 1 to n, where n is the number of points within a profile (e.g. 1300). By changing the position of the table or the measured casting in several steps, new profiles are acquired and joined into a 3D surface image referred to as the cloud of points (X,Y,Z)y. The Z-coordinate is the position of a particular profile in the direction of table positioning and index j runs from 1 to m (number of profiles). On account of the measurement technique applied, the cloud of points has a matrix structure {n, m}, which simplifies further operations for modelling and presentation of the measured surface. The measurement system (Fig. 4) incorporates built-in optics which magnifies the image 4.5 times more in the direction perpendicular to the casting surface than in the direction along the intersection curve. The optics is assembled from a spherical lens and two cylindrical lenses making up the cylindrical Keplerian telescope [10]. The optical system was set up to the measuring range of 200 x 35 mm (width x height) and a mean resolution of 0.15 mm along the intersection curve and 0.025 mm perpendicular to the measured surface. Measurement uncertainty in a particular measured point is 0.02 mm (a). The existing system can capture 15 profiles per second, the speed being determined by the image size (1.3 MB) and bandwidth of the 1394 output (alias fire-wire). The measurement system has been set up to capture 750 Fig. 3. A schematic of the new method for inspecting the die-casting surface quality Fig. 4. Keplerian cylindrical telescope increases measurement resolution in the direction perpendicular to the die-casting surface. profiles in 50 seconds. Further increasing of the speed of the system is only possible through development of a new camera supported by modern programmable logic, according to which the image will be processed by the camera and only the profile will be translated along the fire-wire. 2 DATA PROCESSING In developing software for automated detection of surface defects, special emphasis was placed on the speed of action, since a measurement (an average measurement consists of 2 x 106 points) is to be made and processed within the casting cycle (approx. 60 seconds). The flowchart for detection of surface defects is presented in Figure 5. Detection of die-casting defects is only performed in several different non-overlapping regions of interest (ROI). The number, shape and position of the studied regions of interest are namptv! A N J Coltrol limit M j V O Non acceptable Non acceptable Fig. 5. The flowchart for detection of surface defects selected with regard to the functionality or purpose of various casting surfaces. To this aim, the cloud of points is first divided into several sub-regions ROIk = {(X, Y, Z) j ij e k}, where k = 0, 1,..., K (K is the number of all ROI). Figure 6 shows OI1, for which the process of detecting casting defects will be demonstrated in the continuation of the paper. Each ROI is defined with a boundary consisting of four fixed points, positioned into a rectangular pattern. Fixed points are used to ensure Fig. 6. A particular region of interest (ROIJ is defined with four fixed points. that the casting measured maintains the same position at all times, which significantly speeds up data processing as it enables us to avoid the invariant analysis. In the first processing step, the region of interest is filtered. A mean filter [11] is used to clean up isolated points protruding from the die-casting surface, which have probably been caused by second reflection of the transmitted laser beam on the die-casting. The first important parameter used in assessing the die-casting surface quality is global curvature (GU) of a particular ROIk. In calculating this parameter, the plane (Rk) is first adjusted to the ROIk, using the method of least squares [11]. In the next processing step, perpendicular distances Dk are calculated from the points of ROIk to the plane Rk (Fig. 7). Global curvature GUk is then determined as a distance between the maximum and the minimum of the Dk. According to the specifications, GUk of a particular ROIk may not exceed 0.4 mm. In the second processing step, detection of surface defects is performed. To this purpose, ROIk is sub-divided into small segments (Fig. 8), the surface area of which corresponds in size to a measured points Fig. 7. Definition of global curvature (GU) typical casting defect (e.g. 5 x 5 mm). The underlying idea of segmentation is to perform narrow-band filtering on ROIk surface to detect geometrical shapes the size of the segment. Figure 8 shows a sample segmentation of ROIk. In detecting surface defects, certain parameters are first calculated for each segment and then a statistical analysis of the values obtained is carried out for the entire ROI,. Calculation of k parameters by segment is similar to the calculation of global curvature GU. On every segment Skvs = {(X, Y, Z) j ij e (k a v a s)}, (v = 1, 2, 'J V (number of rows) and s = 1, 2, .. S (number of columns)) a plane R is fitted in a least-square sense, and the perpendicular distances Dvs of the S points to the plane R are calculated. k,v,s r r v,s The first parameter to be calculated for each segment is local curvature of the matrix segment (SU ) which is defined similarly as the global curvature GU , as a distance between the maximum k and the minimum of D . Local curvature of the matrix v,s segment SUvs defines the size of a surface defect. The second parameter describes the average deviation PD of perpendicular distances D v,s L L v,s PDVS(1), nPtSv,s where nptsvs is the number of points within a particular matrix segment Skvs. Both parameters are then statistically analysed along the entire ROIk. The research has shown that defect identification through the average deviation PD is considerably more reliable than the detection based on the local curvature of a matrix segment Fig. 8. ROIl divided to a matrix of 28 x 4 segments whose average size is 5 x 5 mm SU . In the light of this fact, the mean value of PD is first calculated for all the matrix segments within a ROIk, generating the average value APDk, whose multiple M is set as the upper control limit for detection of surface defects. In conclusion, a surface defect is likely to be located in a segment where PDvs > M * APDk. The multiple M of the upper control limit is determined experimentally; in our case M = 2. The defect found is confirmed by the parameter SUvs indicating an absolute defect size. When SUvs exceeds 0.08 mm (specification of surface defect size), the matrix segment {v, s} contains a surface defect, and the casting is marked as »Not acceptable«. Otherwise, the defect is on the verge of acceptance; the casting is marked as "Limit" and needs to be inspected using an alternative method (also visual inspection). 3 EXPERIMENTS The new method is demonstrated through the shape measurement and surface quality control of a group of 32 castings, which had been visually inspected following production. The method's performance is demonstrated on two castings, focusing in both only on the first region of interest (ROI1). The first casting E_2 successfully passed visual inspection in the production. Using the new method, the calculated global curvature is GU1 = 0.15 mm (Fig. 9). The process of averaging PD generates the value APD1 = 0.0032 mm; perception threshold for surface defects is 0.0064 mm (M = 2). The graph of perpendicular distances PD (Fig. 10) shows that Fig. 9. Visualization of Dl on casting E_2. (light: +0.7 mm above Re dark: -0.7 mm below Rk) Fig. 10. Control chart of PD (segment number = (v-1)*S+s) all PD are lower than 0.0064 mm indicating there are no surface defects. In cases where such results are obtained, no further analysis of the local curvature SU is performed. The second casting E_4 is marked as conditionally acceptable in the production visual control. The region of interest ROIj shows signs of laminations (cf. Fig. 2 c and d). Using the new method, global curvature is calculated at GUj = 0.08 mm (Fig. 11), and found to be within the control limits (< 0. 4 mm). The mean of PD values is APD1 = 0.0038 mm; perception threshold for surface defects is 0.0074 mm (M = 2). The graph of the parameter PD (Fig. 12) indicates that PD exceeds the perception threshold in segments 46, 53 and 74, indicating occurrence of potential surface defects in these segments. In such cases, the local curvature parameter SU (Fig. 13) is analysed further. It can be observed that in the Fig. 11. Visualization of Dk on casting E_4. (light: +0.04 mm above Rk, dark: -0.04 mm below Rk) Fig. 12. Control chart of PDv Fig. 13. Control chart of SUv< segment 53 SU exceeds the control limit of 0.08 mm and is therefore marked as containing a surface defect. In segments 45 and 74 SU is below 0.08 mm and these segments are marked as defect limit. As the matrix segment 53 contains a surface defect, the casting is marked "not acceptable" (defective). The same procedure is used for analysing the remaining die-castings. The results are given in Table 1. The second column presents visual inspection results. Castings containing no defects are marked "Acceptable"; conditionally acceptable castings are marked "Limit" and defective castings are marked "Not acceptable". The results of the new method are given in the third, fourth and fifth columns. The third column indicates segments with surface defects, the fourth column states global curvature and the fifth column states the decision regarding the surface quality of the casting. A comparison into the columns Visual Inspection and Decision shows that the new method is stricter in inspecting the castings. All the castings which have been marked as defective in the production, have been recognized as such by the new method; additionally, the method yielded several castings with small defects and marked Fig. 14. Segments with surface defects (46, 53, 74) them as "limit" defects. The reason for stricter control lies in currently low control limits (M = 2 and SU = 0.08 mm). If these are increased, the strictness of quality control decreases. Stricter control causes production costs to rise as it increases reject rates and requires additional checking of conditionally acceptable die castings. The results call for an additional fine setting of control limits in the phase of introducing the new method for casting quality control into the production process. 4 CONCLUSIONS The paper presents the development of a new method for control of die-casting surface quality, which is based on a 3D measurement of the die-casting shape with a laser supported optical measurement system and on analysing the 3D measured shape. In using the new method, the measured shape is sub-divided into several non-overlapping regions of interest, whose size and position are selected with respect to the casting functionality. For each of these regions, global curvature and the position and size of surface defects are calculated. Casting surface quality is assessed by checking whether the calculated values fall within the set control limits. Through this approach, casting surface quality is assigned numerical values and is time-independent, unlike the currently used visual inspection where a worker decides on the Table 1. Comparison of the classic (visual) and a new method for quality control New method Casting Visual inspection Segment with defects Global curvature [mm] Decision E 1 Acceptable 15 0.14 Limit E 2 Acceptable 0.15 Acceptable E 3 Acceptable 0.11 Acceptable E 4 Limit 46,53,74 0.08 Non acceptable E 5 Non acceptable 19,48,49 0.15 Non acceptable U 1 Non acceptable 74 0.16 Non acceptable U 2 Acceptable 0.11 Acceptable U 3 Acceptable 27 0.17 Limit U 4 Non acceptable 46,54,67,74 etc. 0.22 Non acceptable U 5 Non acceptable 2,8,9,10,12 etc. 0.33 Non acceptable U 6 Acceptable 75 0.14 Limit U 7 Acceptable 0.15 Acceptable U 8 Non acceptable 58 0.20 Non acceptable U 9 Non acceptable 10, 28, 53, 54 etc. 0.33 Non acceptable U 12 Acceptable 0.11 Acceptable U 13 Acceptable 8 0.10 Limit U 14 Acceptable 0.10 Acceptable A 1 Acceptable 0.11 Acceptable A 2 Acceptable 0.11 Acceptable A 3 Acceptable 0.11 Acceptable A 4 Acceptable 0.12 Acceptable A 5 Limit 76 0.11 Limit A 6 Acceptable 0.10 Acceptable A 7 Acceptable 0.12 Acceptable A 8 Acceptable 0.09 Acceptable A 9 Acceptable 0.10 Acceptable A 10 Acceptable 0.10 Acceptable A 11 Acceptable 0.12 Acceptable A 12 Acceptable 0.08 Acceptable A 13 Acceptable 0.11 Acceptable A 14 Acceptable 81, 82 0.10 Non acceptable A 15 Acceptable 37, 65 0.14 Non acceptable acceptability of a cast product relative to his/her subjective experience. Tests have proven that the measurement system achieves sufficiently high measurement precision and accuracy (0.025 ± 0.02 mm perpendicular to the casting surface) and that the current speed of measurement (15 profiles per second) is too low to accommodate introduction of the new measurement method into the production process and will therefore need to be increased. Furthermore, a comparison into the results of the quality of test castings has pointed to the need for an additional fine setting of control limits in the phase of introducing the new method for casting quality control into the production process in order to achieve the balance between technological requirements and increasing of costs on the one hand and stricter casting quality control on the other. 5 REFERENCES [ 1] Chandra M.J. Statistical quality controll. CRC press, 2001. [2] Keller P. Quality engineering handbook, 2nd ed. Marcel Dekker, Inc., 2003. [3] Leopold J. et al. New developments in fast 3D-surface control. Measurement, 2003, 33 p. 179-87. [4] Shull J. P. Nondestructive evaluation: theory, techniques and applications. Marcel Dekker, Inc., 2002. [5] Andresen B. Die casting engineering: A hydraulic, thermal and mechanical process. Marcel Dekker, Inc., 2004. [6] Dönges A., Noll R. Lasermeßtechnik, Grundlagen und Anwendungen. Heidelberg: Hüthig, 1993, 214 p. [7] Girod B., Greiner G., Niemann H. Principles of 3D image analysis and synthesis. Kluwer Academic Publishers, 2002. [8] Bracun D., Jezersek M., Diaci J. Triangulation model taking into account light sheet curvature. Meas. Sci. Technol., 2006, 17, p. 2191-96 [9] Bouguet J. Y. Camera calibration toolbox for Matlab, URL http://www.vision.caltech.edu/ bouguetj/calib_doc/index.html (09.10.2007) [10] O'Shea D. Elements of modern optical design. John Wiley &Sons, 1985, 81 p. [11] Numerical Recipes: The Art of Scientific Computing, 3rd ed. Cambridge University Press, 2007.