Automotive assembly is notoriously complicated. A single millimeter of error can bring things to a standstill, but worse than that, parts aren’t always the same, and customization is a growing demand from customers.
Quality control used to mean human inspectors squinting at panels under fluorescent lights. But competition from China has meant that production needs to be faster than ever, and also cheaper than ever, meaning your variable costs must come down.
One way to do this has been to reduce the number of humans involved during the manual checks. Machine vision is precisely how this is being achieved, with systems from specialists like Eines helping replace subjective human observation with artificial intelligence.
Human error and recalls
Even if we take our human inspectors to have 20/20 vision, they still fall victim to drift. This is a phenomenon of psychology about the gradual decline in a human’s ability to spot anomalies at work.
Over time, fatigue and eye strain are real, as are headaches, boredom, and the many environmental distractions. All can contribute to more mistakes throughout the day. Data from the National Highway Traffic Safety Administration (NHTSA) shows that 30 million vehicles are recalled annually due to manufacturing flaws.
These recalls cost brands billions in damaged trust, but also operational costs too. They must track down faulty units once they leave the factory, and the costs of not spotting it during production have increased by several orders of magnitude.
It’s clear that manual checks aren’t giving the 100% repeatability needed for a 24/7 manufacturing cycle, and end up as a false economy. While the upfront cost is larger, machine vision solves this by never tiring and maintaining the same standard of scrutiny throughout.
What can machine vision achieve?
Machine vision isn’t just about sticking a camera on the end of a robotic arm (though it can be that). It’s an entire army of high-res sensors and IoT connectivity. Behind the scenes are complex deep learning algorithms that, while repeatability is high, can self-improve.
These vision systems have been found to detect defects with up to 90% more accuracy than humans. Much like a self-driving car, they can see what we cannot see. Before AI was used, it was rules-based systems, and these struggled with variations in lighting or part positioning – everything had to be highly repeatable.
But the neural networks used by the likes of Eines mean that the vision systems analyze images, finding minute misalignments and even microscopic paint chips.
The knock-on effect of having all inspections digitized and logged is an unprecedented amount of traceability. There’s a permanent record for every chassis – any mistakes can be investigated and identified as to what went wrong. It means surgical, VIN-specific recalls can be done instead of mass-market withdrawals.
Areas where machine vision outperforms manual checks
Investment in machine vision is increasing as the return on investment becomes clearer to see. In the press and body shops, for example, it’s the vision systems that measure gap and flush alignments.
They can do so with sub-millimeter precision before panels are attached, and it reduces stacking errors (where small misalignments early on compound into major fitment issues down the line).
Once the vehicle comes to the paint shop, the optical tunnels scan each and every inch of the chassis – and do so at the same time, making it extremely fast. Even dust can be found, and orange peel textures. The assembly line becomes self-aware and self-correcting.
Machine vision can be viewed as the fail-safe for the human elements remaining on the line, too. It verifies that the correct trim pieces and even internal components are installed for specific vehicle configurations. Again, this is more important in a time when customized products are growing in demand.
By checking both the presence and integrity of parts without slowing down the conveyor belt, the check-while-moving workflow is exceptionally efficient at improving throughput.
Creating systems that think like humans
The true revolution here is actually in deep learning rather than automation as we knew it. Early automated scanners were often finicky, and a slight change in factory lighting, or just a dusty lens, could create false positives.
AI vision systems are very different – they’re more human-like. They use deep learning to genuinely understand and recognize what a perfect part looks like. Unlike a human, it has infinitely more data points.
AI can distinguish between a genuine defect and a harmless variable. For example, a reflection on a drop of water.
Yes/no logic caused too much downtime through false positives, while a nuanced understanding can instead create acceptable variance, much like a master craftsman who knows which minor deviations are harmless and which will compromise the final product. This is important for the bottom line.
Human-like reasoning and self-learning allow the system to adapt to new vehicle models or part variations without needing manual reconfiguration. As the algorithm processes more images, they become better, this is a return on investment that arguably increases over time, rather than depreciates or amortizes.
The business case is clear. While upfront costs may be high for replacing manual checks, variable costs come down (fewer errors and recalls) while throughput and productivity increase.
It’s worth noting that, in times of high wage inflation, switching to automation can save more than initially forecasted. Waste is also reduced, and this can help with ESG reporting and investor sentiment.
Moving to automated inspection
The automotive industry is no longer a place for manual, unreliable quality control, especially when competing with ultra-cheap Chinese companies, which are growing in trust and popularity.
Moving to machine vision is to reduce defects. For any recalls that do occur (of which there should be fewer), the digitized logs of each part can help pinpoint the exact products that need to be recalled, which also saves money and goodwill.
As AI improves, updates can be made, and the vision system becomes more accurate over time. The workers on the line can operate faster and with less stress, as they know mistakes will be caught.