This long-haul truck part was nearly impossible to see once assembled, but the consequences of missing a defect would be impossible to ignore.
At Daimler’s Mount Holly Truck Plant, one of the more difficult quality inspections involved a component that disappeared from view once it was assembled.
The joint connects the steering column to the truck frame, where the steering wheel is attached. A bolt has to pass through two holes in the correct orientation and be properly torqued. Once installed, however, the operator can’t see the joint.
The stakes were higher than a typical quality defect. The steering-column joint connects the steering column to the truck frame, so a failure could leave a truck careening down the road without control of its steering.
The plant relied on what Joanna Cooper calls a “four-eye” inspection process: one inspector checked the assembly inline, followed by another person who inspected it offline and signed off before the truck could be released.
Cooper was general manager of the Mount Holly plant at the time, with 18 years of manufacturing experience. The workforce was also undergoing a significant change, with average seniority cut in half.
“When you bring in new people, when you’re doing changing shift models, and then you have attendance challenges, those are the days that you really run the risk of, are all the eyes actually working or are only some working,” Cooper said.
The plant had a paper trail, she said, “but not necessarily the confidence that everything 100% of the time was done.”
Cooper wanted to close that gap. Through Women in Manufacturing— the only global trade association dedicated to supporting, promoting, and inspiring women in the manufacturing sector—she connected with Priyansha Bagaria, founder and CEO of Loopr Looper AI, a developer of AI-powered quality intelligence platform designed to replace manual inspection workflows.
Bagaria began working with the company and Mount Holly’s manufacturing engineering team to identify a practical application for AI, and the steering-column joint became the test case.
Taking the camera to the part
Cooper wanted a tightly defined project rather than an open-ended AI initiative.
“We really needed to understand, is this actually giving us the result that we wanted to get?” she said.
The joint was particularly important because it was safety-critical and difficult to inspect. Engineers had already been looking for a solution but had not found one that adequately mitigated the risk.
Loopr initially focused on fixed-camera vision systems. But when Bagaria examined the application, she found that approach wouldn’t work. The component was assembled on a moving line, and there was no practical location from which a fixed camera could see the joint.
“We realized there’s no way a fixed camera can ever solve that problem,” Bagaria said.
Instead, Loopr developed a tablet-based system using a borescope. Rather than bringing the part to a camera, the inspection could move with the operator.
The system checks the joint’s fit and orientation and whether the required torque is present. Inspection results can then be recorded and fed into Daimler’s manufacturing control systems.
That also addressed limitations with the existing offline inspection. Workers could climb onto stands to inspect the joint, but trucks varied in height.
“No two trucks were built the same,” Cooper said. “Some were higher than others.”
The physical inspection was only part of the challenge. Different truck configurations could produce situations in which individual components were within tolerance but the completed assembly was still difficult to install.
Cooper described the issue as “stackup.”
“They can all be in tolerance, but if you’re on the high end of one tolerance on oe part and a lower end of tolerance on the other, you could run into a more challenging installation,” she said.
Loopr recorded videos of different configurations and used them to show the system what a good assembly looked like.
When the system encountered something it wasn’t sure about, it flagged the result for review by an engineer rather than leaving the decision to the operator.
“If it was right, then they could say ‘check and the system’ would learn from that,” Cooper said.
The plant did not want operators simply accepting uncertain results. That human review remains a critical part of the system. Operators indicate whether they agree or disagree with the result, creating a feedback loop.
“The only way I see AI doing extremely well and in future … is going to be a human in the loop,” Bagaria said.
From inspection to evidence
For Cooper, the system’s value extended beyond identifying a problem on the line. Because the joint was safety-critical, she wanted serialized information showing that the work had been completed correctly.
The inspection data can provide that record and potentially serve as evidence in the event of a warranty claim. It can also reveal patterns in the manufacturing process, including how often workers miss particular components and where additional training might be needed.
The initial application eventually led to other inspections, particularly involving steering components. Cooper said the reaction from workers changed as they saw the system operating.
“Once that happens and people are able to see it, now you have less of a push and more of a pull,” she said.
For Cooper, that experience reinforced a basic rule for manufacturing AI projects: start with a specific problem, involve the people who understand it and determine whether the technology actually produces the intended result.
“There are so many solutions out there, often people get territorial with their own solutions and not necessarily having the best solution,” Cooper said.
The steering-column joint wasn’t necessarily the plant’s most frequent quality problem. But it was a problem that mattered, was a critical safety concern and had resisted other solutions.
For Cooper, that made it worth solving.
