“What’s this thing going to show me that I don’t already know, sonny?” asked Bill, a nineteen-year veteran of a multinational manufacturer of major drive-train assemblies, on the first day of our proof-of-concept. We had just loaded many years of their warranty claims data, tear-down reports, shipment information, and sales information. Because my name isn’t Sonny, it didn’t seem to be a promising beginning.
Bill follows the industry practice of using Pareto charts to identify the top issues for the products they produce. They ship their products to a number of original equipment manufacturers (OEMs), which install them in multiple factories, into a final product.
I decided to run one of our most popular algorithms, Discovery, which looks for anomalies, correlations, relationships, and patterns in data. After only a few seconds, the algorithm delivered a number of results. Because I did not have deep knowledge of their data, I showed Bill the first result and asked him if it made sense. The result showed that they had been shipping the same product model to a number of OEM factories, but when they shipped it to one particular OEM factory, the product had a higher than expected failure rate in the field. This issue had apparently been going on for years but because it was a low-frequency failure it never got into their Pareto chart of top issues so no one really noticed the problem.
Bill looked at the result and said, “Can’t be!” He then took some notes and disappeared. Three hours later, he came back into the conference room and told me he had found the problem. He said that particular factory was their customer’s oldest factory, and unlike the other three factories, they had a legacy contract with their union that required the product to be shipped to that factory without the output device attached. It turned out that the workers at that particular factory were not always attaching and tightening the output device to the correct torque specification, and that error resulted in premature seal failure at the output device. He went on to tell me that the cost just to remove and replace the assembly – not including the cost of rebuilding it – was $800 each replacement. The fix was easy; they informed their customer’s plant management about the reason for the failures, and they put procedures in place to correct the problem.
Because of something I call intellectual curiosity, Bill didn’t dismiss the result PolyVista had discovered out of hand, but instead, he went on to track down the issue’s root cause. In the end, he realized that PolyVista was able to show him something he didn’t already know.
