An Example of Eliminating a Technical Problem with Only One Single Part Dr.-Ing. Bernd Mittmann Delphi Energy & Chassis Systems Technical Centre Luxembourg bernd.mittmann@delphi.com Introduction This presentation gives an example of a technical problem and its solution that has been developed very quickly and effectively. Only one rejected part had to be measured and this part contained sufficient information to locate the source of the problem. A few additional parts off standard production had to be measured to understand the root cause of the problem. Once the corrective action was implemented the problem was eliminated. Developing the appropriate strategy to attack this problem was the critical step. Without the right strategy this study would not appear as easy as it is presented, now after all the details are understood. The Problem Parts had to be rejected at a rotor assembly line for an automotive generator ever since production has been launched. Typically, rejects had to be scrapped and this meant a big opportunity for improving the financial situation of the plant. Most of the rejects were due to the fact that the magnets could not be fully inserted into the rotor slots as shown in figure 1. This magnet could not be fully inserted due to interference. Slot Width Magnet Width Magnet is pushed into the slot. Figure 1: Problem description Many questions were raised, such as: - Does this problem exist because the magnets are too wide? - Or are the slots too narrow?
- If there is too much variation in the slot width or in the magnet width is it due to the supplied components or is there some variation in the assembly process? - What is the source of this variation? - How to eliminate this problem? Approaching the problem To answer all these questions the Shainin Strategy Diagram for this problem has been developed (figure 2). Side A to Side B (the two ends of the slot) Same Slot A B Design Families Slot to Slot Same Rotor 12 slots total Reproductive Families Rotor to Rotor Same Machine, Time, Figure 2: Strategy Diagram In the first step the focus was set on the design families of variation. These include all sources of variation inherent with the product design, and these can be seen on basically any part. Of course, the differences are more obvious on rejected parts. It is therefore critical to select the right parts for any measurements. Converging to the Red X Suspect A rejected rotor had been selected and care was taken to select one where several magnets couldn t be inserted and some of them didn t go into the slot half way. Any rotor with only one magnet insertion failure or with magnets that almost fit into the slots would not show the important contrast to the same obvious magnitude. Measurements of the rotor slot width and the magnet width were taken on this part according to the strategy. The measurement system used was a caliper which is a quick and inexpensive device. Figure 3 shows the data in a graphical format.
11.3 11.2 Slot Width (mm) 11.1 11.0 10.9 10.8 10.7 10.6 10.5 10.4 1 2 3 4 5 6 7 8 9 10 11 12 Side A Side B Distribution of Magnet Width Slot Nbr Figure 3: Data of one rejected part The following facts were immediately obvious: - The caliper is an appropriate measurement system. It is possible to discriminate between wide and narrow slots and the measurement variation is significantly smaller than the difference in the items that have been measured. - Further efforts must focus on the rotor. The magnet width varies only very little, whereas the rotor slot width has too much variation. An option to eliminate the problem through the magnets would be to shift the entire distribution of the magnets down by about 0.1 mm, i.e. to change the specification. Obviously the supplier has a very capable process but of course adapting the supplier s process would not come without any charges. Moreover, these activities typically take quite some time. - For any position (i.e. slot number), there is no measurable difference between the slot widths on the two ends. - There is an obvious pattern in the rotor slot width: Narrow slots and wide slots are alternating. Most likely, the problem is created in the staking process. This is the operation that assembles the segments to the shaft and where the components are positioned relative to each other. The alternating pattern suggests a rotational off-set between the two segments. As shown in figure 4, there were two machines doing this operation, the old one and the new one. Other Operations Point of Creation Old Staking Machine New Staking Machine Other Operations Point of Discovery Magnet Insertion Figure 4: Process Flow Diagram Other Operations
A Multi-Vari Study had been performed, looking at the reproductive families of variation (i.e. variation inherent with the process): - Select three consecutive parts off the old machine. - At the same time, select three consecutive parts off the new machine. - Measure the slot widths at either end of all the rotor slots of these six parts. - Repeat this at a later point in time twice. Tuesday 9.00 am 11.3 11.2 11.1 11.0 10.9 10.8 10.7 10.6 10.5 10.4 Old Staking Machine New Staking Machine Tuesday 11.00 am Wednesday 10.30 am Very similar patterns observed. (Not all data shown due to complexity.) Almost the entire range observed on the 1 rejected part has been captured. Figure 5: Multi-Vari Chart The conclusion (see figure 5 for the graph) is that the problem definitely is created in the staking process; it can actually be assumed that the old machine is producing virtually all rejects. The reasons for this statement are: - All three parts coming off the new machine had very little slot width variation, and this was true at all three points in time (so with a total of nine parts), - Out of the nine parts produced by the old machine, about half of them exhibited a very strong alternating pattern. - The supplied components go randomly into the old and new machine; therefore the root cause of the problem is not in the components but in the old staking machine. It is important to note that almost the entire range of smallest slot width to largest slot width has been captured during the Multi-Vari Study. Consequently, the root cause of the problem has been present when sampling the parts; it just was not strong enough to produce a reject. But since a fairly small sample size has been used chances of sampling rejects are very small and it can only be expected to capture some 80% of the entire range. Again, a number of questions were open: - Would it be necessary to replace or upgrade this machine? - Should the old machine no longer be utilized and the demanded number of parts produced with additional shifts on the new machine? Either option was not acceptable due to the cost involved; in the end this study was aimed to improve the plant s financial situation.
In fact, carefully observing this particular assembly step of staking and comparing the two machines pointed out the Red X (root cause) for the variation in slot width. First, it had been noticed that there was a gap between the lower segment and the fixture s pins (see figure 6). The segment was allowed to be rotated by some degrees to touch either the left or the right pin. However, this was true for both, the old and the new machine. Since the new machine produced good parts, this gap could not be the root cause for the problem. Tool to stake the upper segment. Right Pin Left Pin Figure 6: The staking machine At this stage, the staking process had been carefully observed in manual mode (low speed). It has been observed that on the new machine the tool to stake the upper segment moved downwards to hold this segment and also the upper tool s pins were guiding the lower segment before the staking operation was performed. In contrast, on the old machine the staking process started before the tool s pin did touch the lower segment. Ultimately, for the new machine it was ensured that the components would be properly aligned before the staking operation started, whereas on the old machine the components alignment was not properly ensured by the tool. The tool s pins had been measured and the ones on the old machine had found to be shorter than the ones of the new machine. The Red X of the problem seemed to be the length of the tool s pin. Confirmation of the Red X Gap allowing the part to be aligned in extreme positions. To confirm the Red X being the pin length it would be necessary to replace the pins by longer ones. Alternatively, it has been chosen to produce parts on the old machine in the two extreme positions when rotating the lower segment in the lower fixture (as indicated in figures 6 and 7). This was easier and quicker to perform and in case the thoughts so far were correct it should be possible to deliberately produce extremely good and bad parts. If this was not confirmed then some of the work so far
was incorrect and no time and money would have been wasted on replacing the pins where maybe it wouldn t be necessary. Therefore, three parts have been produced where the segment had been put on the fixture so it touched the right pin (indicated in green color in figure 7) and another three parts were made so the segment touched the left pin (red color). 11.3 Good Parts Bad Parts 11.2 11.1 11.0 10.9 10.8 10.7 10.6 10.5 10.4 Figure 7: Confirmation The entire range observed on the 1 rejected part has been captured. Figure 7 demonstrates the set-up as well as the test results. It is obvious that the two extreme alignments gave the extremely good and extremely bad parts. None of the good parts was in risk of yielding a reject but all of the bad parts were very likely to end up as rejected parts. The process output could be predicted reliably! As with the Multi-Vari Study it is important for the confirmation to note that the entire range of smallest to largest slot width has been captured. This is giving the rational to state that the factor controlled in the experiment is not only producing some variation of the output but it is producing the extremes that have been seen on the rejected part at the beginning of this investigation. Therefore, it can be concluded that the root cause of the problem has been understood. Apply the corrective action The corrective action an increased pin length on the old machine - was implemented and the problem had been eliminated, see figure 8. Reject Rate: Corrective action implemented. January February March April May June
Figure 8: Reject rate vs. time Basically, one single part contained all the answers needed to eliminate the problem and the reject rate dropped by 80% within a few days only and with virtually no investment! Acknowledgements I want to say a big Thank You! to my colleagues - Eric Glandus and - Désiré Djomani who made sure that this study resulted in a success. Also I am thankful to - Tobias Mack and - Tim Nelson of Shainin LLC from whom I learned the tools applied in this study, among many other useful tools. Shainin and Red X are registered trademarks of Red X Holdings LLC. Strategy Diagram is a trademark of Red X Holdings LLC.