This morning, a news article about the recall of more than 5.5 million vehicles by Honda and Toyota caught my attention. One recall is due to a fuel pump failure issue, whereas the other pertains to airbag defects. While this was the first-ever fuel pump-related issue for Honda, GM recalled thousands of vehicles earlier this year for fuel pump-related issues. Airbag defects are among the top reasons for recalls by almost all automakers. All that information made me wonder if we can use AI to minimize these recalls.

Apart from the safety concerns, the monetary effects of these recalls are gigantic. In 2021, Ford had to set aside $4B to cover warranty costs. While eliminating recalls entirely is next to impossible, Ford and other automakers can spend a fraction of that warranty cost to build capabilities that can help them significantly reduce their warranty costs. But why do I say that recalls can never be eliminated? Because the real world is (and please excuse my language) a b**ch.

Automakers today are now being supplied by hundreds (in some cases, thousands) of multi-tier supplier networks. While quality control aspects are in place, it is next to impossible to perform comprehensive quality checks on all of them. The good news is that they do not need to do quality checks on all of them. Since reading the news article this morning, I have been digging into the top parts that lead to significant recalls. If you apply the Pareto principle, the number of parts that lead to 80% of recalls can be counted on the fingers of your hands. Yet, even for these, you can never eliminate every possible recall. Why?

At a high level, the part failures and recalls can be classified into two major buckets (there are some additional buckets as well, but we will disregard them). These buckets are defects from wear and tear/usage and design defects. Critical parts are designed and tested for a peak life duration wear and tear. But as I indicated earlier, the world is a b**ch. When assembled in the automotive, it is made to work in tandem with other parts and influenced by aspects like the gentle or harsh treatment of the car by the user, local weather, and whatnot.

The lowest hanging fruit here is design issues. An example is a crucial recall reason for seatbelt latches or seatbelts. Pick any leading automobile maker, including EV manufacturers like Tesla, and you will find at least one seat belt-related recall, with some as elementary as seatbelts being hard to use. There are two ways to attack this problem with AI. One is by leveraging a customized Generative Design AI solution that has also factored data on every single recall made for that part by any automobile manufacturer. The automotive industry is no stranger to Generative Design. GM has been experimenting with Generative Design, including in the seat belt bracket design arena, as you can read in this article, Driving a lighter, more efficient future of automotive part design.

In case you are wondering, Generative design is a 3D modeling technology that leverages cloud computing and AI. Engineers set up requirements for the design model, like manufacturing processes, loads, constraints, etc., and then the software generates designs that meet those requirements. So, if the likes of GM are already using Generative Design, why do they still run into design issues? In fact, the probability that GM will issue a seat belt-related recall in 2024 is very high. What is not aligning here?

It is the solution design. As indicated above, custom aspects, like possible design issues that have led to recalls, need to be included in the “intelligence” and configuration of the solution. The good news is that the list of problems is not very long and complicated. What is defined simply in the definition above, “like manufacturing processes, loads, constraints, etc.” hides a mind-boggling amount of details. And those are the details that need to be taken care of when designing a solution that can help reduce the cost of recalls by half.

Now, the other area is recalls due to wear and tear/usage. Simulation powered by AI should be a component of a Generative Design solution and should factor this into the design phase itself. But let us say that we want to add another layer of AI-enabled capability to further bring down the recall costs. And let us try to understand this using an example of seat belt latches.

Issues due to usage may lead to malfunctions in seat belt latches and recalls. This is where an AI solution can use periodic data from the real world to learn how those parts change with usage. An example of a data-point source is when a car comes for servicing at a dealership. A picture of the part gets uploaded and analyzed by a deep-learning solution. Data can be collected more frequently through voluntary participation (by offering perks). This is just a high-level overview, and many more details will go into designing such a solution. But at the core of this solution will be a customized simulation tool, coupled with deep learning.

Building and implementing this is not easy. But once in place, ROI can be less than a year. Most leading automakers have already invested in building AI capabilities. The time is right to leverage them to attack issues that have plagued them for decades.