Automotive Spells Quality With AI

As eloquently said by Illinois’s governor, JB Pritzker, during a viral commencement speech at Northwestern University this summer, “When we see [something different than ourselves] the first thought that crosses almost everyone’s brain is rooted in either fear, or judgement or both. That’s evolution. We survived as a species by being suspicious of things that we aren’t familiar with.” This might be true of Artificial Intelligence (AI) articles such as “Are We Facing AI Armageddon?” or “War, Pestilence, Extinction and Artificial Intelligence: Technology Leaders Sound A New Alarm.” Maybe it’s also at the root of a portion of this last week’s political actions to limit AI around the world (e.g., U.S. Vice President, Kamala Harris, “Warns That the ‘Existential Threats’ of AI Are Already Here”). Let’s face it (no pun intended), it might even have been your first reaction when you saw the picture above.

But let’s hold the phone for a minute: AI has many, seriously-positive aspects within our technology-centric world, which we need to understand, celebrate and protect in the midst of politicians taking aim with restrictions and regulations. For example, just automotive quality – which, mind you, can strongly influence the functional safety and cybersecurity of the vehicles that carry your loved ones – has multiple threads, with three good examples worth understanding: predictive maintenance, end-of-line inspection and anomaly detection.

Predictive Maintenance

Predictive maintenance is essentially spotting something via analytics which will eventually malfunction and fixing it before it does. Computers not only spit out information and identify trends in data from possibly dozens of sensors, but then conclude likely results and train themselves on future responses. Such predictive maintenance has been used for decades in aerospace to evaluate when engines must be pulled off airplanes (e.g., the 1940s Royal British Air Force rethought scheduled maintenance of WWII planes using such methods), which is a very expensive and timely proposition given “uptime” of the fleet.

Automotive started using predictive maintenance as of the 1990s in mostly heavy machinery. 15-20 years later, Chevy claims to have launched the first ever driver warning using predictive maintenance (PM) in 2016. Fast forward six years later and the 2022 automotive PM market is $18.9 billion USD with a Compounded Annual Growth Rate (CAGR) of 18.6% through 2032. What’s more: those numbers don’t include all of the untold cost avoidances such as injuries, lost productivity, etc. from failed parts that cause accidents, deaths, etc.

End of Line Inspection

Visual inspection of both the plant and the parts within the plant has existed since the beginning of manufacturing, but human inspection has only been shown to be 70-80% effective. Additionally, cameras and sensors can inspect simple parameters, but are only as smart as the engineers programming them.

Artificial intelligence can use a dataset of past failures and associated sensor data to learn what results in downtime, a part failure, or a less-than-optimal ending.

“Downtime on the shop manufacturing line is extraordinarily expensive,” states Jim Keller, Quantphi’s AWS Global CEO. “One of our customers had fifteen [intertwined] root causes of failures on their line, and was looking to accurately predict when the line would require maintenance and necessitate taking the line down. We captured the data from their sensors, leveraged AI and provided them predictions 10 hours prior to a potential downtime incident so they could deploy resources to remediate that situation, redirect that line or redirect the resources of that line.”

Anomaly Detection

In theory, a vehicle should know exactly how quickly it’s traversing the highway since there are multiple sensors on the vehicle (e.g., wheel sensor, GPS positioning) which measure speed and communicate that across its internal network. However, sometimes those speeds differ slightly due to calibrations, but not significantly enough to cause automatic, emergency braking to falsely activate.

That is, as long as a hacker hasn’t cloaked a malicious message and remotely injected it into the vehicle’s network to create an accident. And for many legacy vehicles using an age-old “CAN network,” the normal coding wouldn’t know the difference. It’s just looking for a message that’s been formatted correctly and acts expeditiously. And, even worse, traditional systems wouldn’t inform the manufacturers (or enforcement agencies) of the attacks.

Artificial intelligence creates a more robust system by looking for the combination of variables and associated data that’s unusual (e.g., comparing the wheel sensors to GPS data to <x>), and flags suspicious behavior to cybersecurity engineers at a Vehicle Security Operations Center. This certainly isn’t the end-all-be-all of cybersecurity detection systems, but adds another arrow to the quiver.

“There’s no magic pill for any of these Artificial Intelligence applications,” states Keller. “It’s really about being practical about how this technology can be leveraged, and solving what matters. Companies like Quantiphi want to do that in a way that’s demonstrable, provable and backed by data.”

That doesn’t sound so scary to me.

Author’s Note

COVID has affected our lives in ways that none of us can imagine. I semi-joke that all of us will have behavioral effects akin to our Great-Depression-surviving grandparents picking half-eaten toast out of the trash while uttering “This is still perfectly good food.” Our kids will suffer from reduced social interaction, young adults will be handicapped from poor workplace integration and mentoring (e.g., in a hybrid environment, it’s harder to look over someone’s shoulder), and the Gen X’ers will ache from a strong distrust of politicians.

I am a Gen X’er.

By no means do I wish to politically sojourn within an automotive article and suffer either the wrath of my Editor or hundreds of comments. But I do want to pronounce that I’m inherently suspicious of politicians cordoning off things that are technologically beyond Capitol Hill advisors and lobbyists alike, especially when it could hamper the quality of important, safety-related, societal tools such as cars.

Of course, though, maybe I’m suspicious because politicians are different than me, so my first thoughts are either fear, judgement, or both.