I am having a debate. For those who know me, that won't come as a surprise. I love to debate about things that I'm passionate about. The current debate that I'm having is about product design. The specific topic is about how to approach determining part tolerances for a new design.
The debate is prompted by problems we are having getting a supplier to achieve the tolerances we specified for the parts they make for us. We have been through three rounds of back and forth with this supplier trying to get good parts only to finally decide to do an experiment and learn that we could have given more tolerance to the supplier to begin with.
Of course, my suggestion to rectify this problem is well designed and planned Design of Experiments (DOE) for the purpose of determining the edges of the performance window. The counterpoint is that DOE takes too long and we don't have time or the samples to do it. This post is not a how to on DOE, I can cover that another time. This post is about what is the best business decision to make.
So, lets analyse the sides of this debate. On one side the arguement is that DOE takes too long, we don't have the samples and we can't test everything. I would agree with one of those viewpoints, we can not test everything. We should only test the important things. How to determine the important things, for that you need a QFD (Quality Function Deployment) tool. We'll save that for another time also.
The other side of this debate is that NOT doing DOE takes too long, costs more money and reduces the chances of success of the design.
What are the facts? DOE can take a while to complete IF too much is thrown into the mix to test. DOE matrices can grow to hundreds of replicates and samples if not contained. There are strategies to whittle down the many factors to the critical few, among the best I have used are Taguchi Designs. Taguchi design matrices allow for testing many factors at two or three levels for the purpose of screening factors for importance. I have completed simple two factor two level designs in less than a week from start to finish, as recently as this summer. Can DOE take a long time, yes, if we dont plan it well. So what is the cost in time and money associated with NOT doing Designed Experiments? Those costs are real and painful. In the real life debate I am having, we have been working with a supplier since April of 2010 to try to get good parts, now, in October, we still don't have good parts. This has delayed the intended release of the design several times and forced us to decide between another delay and accepting a 40% scrap rate for these parts. This is only half of the story though. By not doing experiments to determine the real needed specification, we have asked the supplier to make a part that costs more to produce because of the extra care required to obtain the tight tolerances we ask for. Additionally, the tight tolerances require that we measure the parts on a sofisticated, highly precise measurement machine that takes quite a few minutes to make the measurement for one part, and finally because we can not get good parts, we must measure every part, further adding to cost and slowing the process. The final cost associated with this decision is in the manufacture of the design. Due to all of the issues, Assemblers will probably have to rework these products in order to make sure they function properly before leaving the factory, taking more time that could be spent doing other things.
To me, the decision is easy. Determine the proper specification upfront by building samples to test the boundaries of performance on critical dimensions, incorporate those learnings into the tolerance for the design, give the suppliers as much room as possible to make good parts, and measure it as simply as possible with acceptable precision. Seems worth the investment of a few weeks effort to me.