AUTOMATED INSPECTION SYSTEM OF ELECTRIC MOTOR STATOR AND ROTOR SHEETS

Transcription

1 9th International DAAAM Baltic Conference "INDUSTRIAL ENGINEERING" April 2014, Tallinn, Estonia AUTOMATED INSPECTION SYSTEM OF ELECTRIC MOTOR STATOR AND ROTOR SHEETS Roosileht, I.; Lentsius, M.; Mets, O; Heering, S; Hiiemaa, M. & Tamre, M. Abstract: In manufacture of machine parts, an automated product inspection can save time and resources. Large scale electric motor stator and rotor sheets are manufactured by stamping. The sheet metal is usually stored outdoors and brought into the factory before processing. The difference in temperatures and other factors create the need to inspect the produced sheets to verify that they meet specifications. This is often done by hand, but this operation is time consuming and is prone to human error. The aim of this work is to explore the possibility of using a machine vision system to solve this task. Images of the sheets are analysed by the custom software, taking into account the thermal expansion of the material. Key words: Machine vision, measuring system, 1. INTRODUCTION A problem has risen in the production of electric motor stator and rotor sheets on how to keep those large-scale parts inside strict precision requirements. The temperatures fluctuations, the ware of the stamping tools and different material batches create a need for strict quality control. At the moment the parts are measured manually after stamping and only the outer and inner diameters are checked. This method isn t satisfactory for the high precision needed possibility for measuring error is great and it can t be assured that the necessary tolerances are kept. The certainty that the tolerances are met is essential at this stage of production it will prove very costly when a preassembled rotor wouldn t fit inside the stator at the later stages of production. Moreover, it takes a lot of time and effort from the machine operators to measure such large-scale parts manually. Due to large difference in part sizes and small tolerances a special coordinate-measuring machine would be otherwise needed and it is considered too expensive. Thus it was decided to look into the possibility of constructing a machine vision based measuring device for that task. The stator and rotor sheet parts (later referred only as sheets) are manufactured from a 0.5 mm thick electrical steel. Sample technical drawings (see Fig. 1) were acquired from manufacturer and are the basis of our work so far. Another objective stated in the assignment was the possibility to measure sheets with different diameters and shapes the system needs to be quite versatile and reconfigurable for different part sizes this would be done once every couple of days Fig. 1 Stator sheet part 283

2 Dimensions of the sheets are dependent on the temperature during and after the stamping. According to the manufacturer the temperature is one of the key issues for keeping the tolerances within limits. This was also considered when developing the measuring system. Moreover, the logging of all the measuring data and statistical analysis of it should also be incorporated into the system. 2. INSPECTION SYSTEM 2.2 System overview As the main goal of this work was to study the possibility of using a machine vision for making precise measurements in described situation the actual measuring device was not constructed to every last detail it is dependent on the concrete specifications needed from the company and will be designed based on those. Despite that, the outline of the inspection system was designed, appropriate components were chosen and a software algorithm was created. For research, a smaller scaled system was designed and produced, more on that in Section 3.1. For the full-scale system, it was decided to use following concept: two cameras are placed over the work area so that segment of the sheet from each end would be in a frame. This allows measuring the inner and outer diameter as well as the grooves of the sheet. As the image frames are not overlapping, it raises the question of distance measurement between the cameras. To overcome that, following concept was invented: the distance measurements are bound to the worktable through precisely fabricated reference sheet part. The edge of the reference sheet is caught into the frames of both cameras and is used as an origin for taking the measurements. Since the length of the reference sheet is precisely known, the full measurements can be calculated. A precise positioning of the cameras in relation to the measured sheet isn t necessary as long as the camera s distance from the measured object is fixed. That means the sheet could be positioned arbitrary on the work area and we would still get accurate measurements. Additional task was to measure sheets with different diameter. With this concept it is easily done reference sheet is changed and cameras moved. For added ergonomics the system can also be made automatically adjustable. The system consists of two cameras, a computer, two flash units, temperature sensors and a diffuser (Fig. 2) Fig. 2 Overview of the system The object is lit from beneath by two flashes, one for each measuring area, which are directed at a diffuser to provide uniform lighting (Fig. 3.). The flashes produce a very intense light, overpowering any ambient light and giving a clear outline of the object. Tests showed that the area lit by the flashes needs to be sufficiently concealed from the rest of the room to prevent unnecessary reflections. Fig. 3 Lighting setup Another important task of the system is taking into account all the temperatures involved. To accurately verify if the sheets fit inside strict precision requirements the thermal expansion needs to be considered. Firstly, calculations show that in the case of the sheet s outer diameter being

3 millimetres, a 10 C temperature rise would result in 0.11 mm expansion which exceeds the required tolerances. For purposes described, sheet s and reference sheet s temperatures are measured during the inspection (see Fig. 2). Secondly, the temperature differences before, during and after stamping process lead to the part not complying with specifications and monitoring those would be really helpful. Measuring this data needs to be integrated into the system so it could be evaluated beforehand if the finished part would be precise enough to pass the inspection. It would help to decrease wasted production time and reduce scrap material. The information necessary to decide if the material temperature meets the allowable limits comes from the data logging statistics described in Section System components The main components of our inspection system include cameras, lenses, flashes and temperature sensors. Next, a preliminary selection of those components is made. For the cameras we used quite new approach to the machine vision systems commercially available digital SLR cameras. DSLRs fit perfectly to our situation large requirements are put on image resolution and the measuring process isn t time-critical. Furthermore, almost no effort is needed from us to integrate different parts of the system, as lenses, flashes and a computer. DSLRs offer supreme image quality to most machine vision cameras. The camera we will use for this task is the Canon EOS 100D. It has the largest image resolution in it s class with 18 megapixel 5184 by 3456 resolution. For the camera lens we will use the Canon 50mm f/1.8 II. It offers distortion free and sharp images. For the flashes, almost any professional xenon tube flash will do, as an example the one s used while testing were Canon Speedlite 430EX and 580EX II. Temperature of the sheet is measured with an infrared temperature sensor. To get an accurate measurement, the emissivity of the material and possible background reflections need to be taken into account. Suitable temperature sensor for this task would be PC151MT-0 from Calex. It has configurable emissivity settings, reflected energy compensation and a built-in RS485 interface [ 2 ]. 2.3 Algorithm and user interface A custom software was designed during this project using NI s Labview and it s Vision environment. Also, a preliminary user interface was made (Fig. 4). The algorithm designed is capable of measuring the parts automatically or manually, images are calibrated accordingly (Fig. 5), large number of parameters can be modified during system setup, and all the data is logged. Fig. 4 User interface Fig. 5 Calibration matrix 2.4 Data logging A computer based measuring system gives us one additional opportunity it is really easy to implement the collection and statistical analysis of all the data gathered. 285

4 The measured values, calculated uncertainties, temperatures of the sheet before and during stamping and during the measurement, inspection system s reference temperature, the exact time and part numbers currently measured are logged by the algorithm. Additionally, even the material supplier and batch details, sheet s tracing code, stamping process settings and any another relevant information could be logged. As described before, such statistics are really useful for future quality improvements. The producer could even take into account how the material from different suppliers influences the stamping process and set up the die according to the sheet properties and thus provide more uniform quality. Additionally, any issues with production quality can be precisely traced down to initial process parameters. 3. RESULTS 3.1 Test rig To evaluate the precision capabilities of our inspection system to be developed, we need an estimate of the uncertainties involved using such technology. In order to do that, we constructed a test rig and estimated it s measuring. Then we would get an estimate of the precision our fullscale system is capable of. Compared to the designed inspection system, our test rig uses different camera, instead of Canon 100D we used Canon 350D an older model from Canon lineup. It has 8 megapixel sensor with image size of 3456x2304 pixels instead of 18 megapixels of 100D. Otherwise, camera lenses and flashes were exactly the same. To assume the limiting factor of the system precision is only the camera resolution, it would mean that the full-scale system should be roughly 1.5 times more precise than our test rig. Of course in addition to the resolution many parameters have an impact on uncertainties, but those have roughly the same magnitudes on the full scale system. 3.2 Results obtained To obtain the data necessary for calculations, four different reference objects were measured repeatedly three of them are shown in Fig. 6. Fig. 6 Reference objects For the purpose of uncertainty calculation, the reference objects were measured with micrometer caliper. The lowest uncertainty was achieved from reference object nr. 4 and was the basis of calculations X ref = mm, u X ref = mm. Ten measurements of the reference object were taken in different parts of the work area. First 5 results of the measurement series can be seen in the table, no bias correction made. Measurement Value [mm] x x x x x mean value x Table 1 Object nr 4 measurements 3.2 Uncertainty calculation A direct method for measurement uncertainty determination consists of the following steps: Investigation of precision; Investigation of trueness (bias); Correction of bias; Determination of measurement uncertainty [ 3 ]. In our work we used the procedure One- Point Protocol approach it is applicable if it can be taken for granted that the result obtained on the reference object can be 286

5 considered as a representative point for the entire measuring range. In our case this is so there should be no precision difference along the measuring range for the machine vision system. The procedure goes as follows: the reference object will be measured repeatedly (at least n = 6 times) under normal operation conditions. From those measurements appropriate standard deviation is obtained. This will thereafter be called "procedural standard deviation" and the symbol s v will be used. [ 3 ] Next we should investigate the bias of the measurements. A machine vision measurement system is in most cases prone to some bias it is dependent on how the algorithm is detecting the edges of the parts. The bias could be minimized by either changing the edge detection algorithm or mathematical correction of it. Correction of bias is reasonable because in case of machine vision system it can be taken for granted that the bias is constant over the entire measuring range. What is special for a machine vision measurement system is the fact that presence of bias is dependent on measurement taken. Fig. 7. Measuring bias As seen in Fig. 7, if a measurement is taken from rising edge to another rising edge there is virtually no bias present. Currently we will consider measuring bias in uncertainty calculations as this is consistent to our test measurement conditions, but in case of normal operation, the measuring algorithm must take into account the measuring condition and automatically decide if bias correction is needed. If a constant absolute deviation can be expected, the mean deviation will be subtracted from measurement result: = x meas x ref y corr = y meas (1, 2) Here x meas is the mean value of the calibration measurements, x ref the reference value of the measurand, y meas the measurement result from the algorithm during normal operation and y corr the corrected measurement result. [ 3 ] The standard uncertainty of the corrected measurement result is calculated according to the rules of uncertainty propagation. For a correction according to (2) this gives us u(y corr ) = s 2 v + s v 2 + u(x n ref) 2, (3) where s v is procedural standard deviation and u(x ref ) is standard uncertainty of the reference value. [ 2, 3 ]. So, to determine the measurement uncertainty we followed the procedure. First we must know the s v reference object was measured repeatedly and standard deviation was calculated. Two different procedures exist for the determination of that. The Type A procedure is calculated by using statistical methods and the Type B procedure is used for the estimation of uncertainties caused by individual factors, due to systematic effects. To find out the standard deviation of the reference measurement series we used Type A evaluation procedure - experimental standard deviation s of the measurement series can be used as an estimate of the standard deviation. [ 2, 3 ] s = n i=1 (x i x ) 2 (4) n 1 x is arithmetic mean, x i is i-th measured value and n is number of measurements. According to the procedure we can equate that to s v. The expanded uncertainty of measurement is obtained by multiplying the standard uncertainty by a coverage factor k. A value of k = 2 is recommended, which roughly corresponds to a confidence level 95%. [ 2 ] 287

6 U = k u (5) From the data gathered (see Table 1) we can calculate the standard deviation s (4): s = s v = mm As discussed in Section 3.1, during testing we used a camera with lower image resolution and quality. For the full-scale system s v would roughly be 1.5 times less. Standard uncertainty of the full-scale system s one measuring area calculates (3), u X ref = mm. (Section 3.2) : u(y 1.1 ) = = mm and the uncertainty of the whole measurement: customizable and can be improved upon in almost all aspects with additional resources. In addition to accurate measurements, the statistics are gathered which are really useful for future quality improvements. 6. REFERENCES 1. CALEX PC151MT-0 Technical Data anuals/pyrocouple%20manual_c.pdf ( ) 2. Guide to the expression of uncertainty in measurement. International Organization for Standardization, Geneva, Guide to the Evaluation of Measurement Uncertainty for Quantitative Test Results. EUROLAB Technical Report, Paris, 2006 u(y 1 ) = u(y 1.1 ) 2 + u(y 1.2 ) 2 + u(y ref ) 2, (6) where y 1 is the sheet s diameter, y 1.1 and y 1.2 measurements from the reference sheet to the part s edge and u(y ref ) is the standard uncertainty of the reference sheet. So: u(y 1 ) = = mm And for expanded uncertainty we get (5): U = = mm In the future, additional uncertainty components could be added which were not taken into account now due to their small contribution, as an example from temperature measurement errors and fluctuations. We can confidently say that the current precision is more than enough to meet our assignment. 5. CONCLUSION Proposed system is a viable option for lowcost automatic measuring of stator and rotor sheets, moreover, the technology can be used for accurate measurement of any flat sheet-metal part. It is highly 7. ADDITIONAL DATA ABOUT AUTHORS 1) Ingmar Roosileht, master s student Tallinn University of Technology,, Ehitajate tee 5, Tallinn 19086, Estonia ocvatorium@hotmail.com ) Oliver Mets, master s student mets.oliver@gmail.com ) Marek Lentsius, master s student lentsius@hotmail.com ) Siim Heering, master s student siimheering@gmail.com ) PhD. Maido Hiiemaa maido.hiiemaa@ttu.ee 6) PhD. Mart Tamre Mart.Tamre@ttu.ee 288