Artificial intelligence (AI) is now becoming a common household word. Almost every day there is news about how much money is invested on AI development. AI was known a few decades ago as neural networks and its application was found in many areas, including controlling automatic transmission shifts depending on driver demand for power. What changed now is the massive increase in computing power due to the development of high power semiconductor chips, and the costs have dropped significantly. The performance of AI or the accuracy for prediction depends a lot on the quality of data fed into the AI model. 

I was wondering how AI could be effectively utilized for automotive design, engineering productivity improvement, problem solving, etc. I am sure there are teams exploring AI applications in these areas. I wanted to focus here on AI application on solving automotive engineering problems. Automotive companies have nearly a century worth of data in these areas, but to utilize it today requires availability of these data in digital form. I think that is where the problem lies. Although corporate memory retention related to problem solving has been always emphasized, it depended on engineers to do a good job of documenting it. There are established processes to capture how the problem was solved, but there are many others that were not captured. This is mostly because the engineers had to move on to other things and not enough time was spent on documentation. As a result, I observed the reinvention of wheels many times. Every automotive company has its own search engines, but they do not necessarily function as desired. With a good AI model, not only will the search function be more effective and efficient, but it will also analyze existing data to come up with recommended solutions. That is where the strength lies.

There are commercial AI models like ChatGPT, which does a reasonably good job in answering general questions like travel planning, tax laws, etc. These models are fed by data that are available in the public domain. However, most company data are private and not available in the public domain. Therefore, ChatGPT will not be very helpful. Let me explain with a couple of examples.

Often malfunction of throttle body sensor in an engine is linked to a tribological issue, mainly degradation of grease. The throttle body sensor allows air intake in the engine according to driver demand of power. In essence, there is a set of pointed metal fingers that slides back and forth against a resistive track (metal strip) and are lubricated by grease (thus a tribological contact). As the demand for power goes higher, the fingers move further, drawing more current and opening the butterfly valve more to let more air in. When I asked ChatGPT for a solution to this problem, it simply says, “Over time, this track wears out”—no mention of grease failure. Another example related to gelling of engine oil in diesel engine—this is sometimes caused by addition of vegetable oil (in particular, higher triglyceride content) by some customers in the engine oil pan. When I asked ChatGPT, the possible reasons mentioned are, “oxidation and thermal degradation,” “contamination by soot or fuel,” “coolant/moisture contamination,” “wrong oil viscosity” and “additive drop out.” Of course, these are all possible reasons but misses on higher triglyceride content. 

Therefore, to take advantage of AI, companies need to generate documentation of solved problems in digital form. There would be a slow start, but if we keep emphasizing and/or incentivizing engineers, gradually the digital database will grow, and bigger benefits could be reaped in the future through use of an AI model.

Dr. Arup Gangopadhyay is retired from Ford Motor Co. and is based in Novi, Mich. You can reach him at arup.gangopadhyay@sbcglobal.net.