AI in Automotive Production

This blog was co-written with Matthias Bastian.

Key takeaways

• Visual inspection requires human judgement, but the process can suffer because of human limitations.
• AI and AR provide tools that support the human inspector, resulting in increased productivity and decreased errors.
• Implementing this high-tech solution is fast, easy, and already happening in leading-edge companies.

Artificial intelligence is about to take quality inspection and end-of-line inspection in the automotive industry to a new level. The effort involved is manageable, and the results are great.

Visual inspection is crucial for high quality in maintenance, service and final acceptance. A keen eye and human judgement are irreplaceable. However, studies show that errors occur in one-fifth of inspections.

Complex products, long-term demands on worker concentration, and unclear instructions make errors unavoidable. Anyone who spends a shift checking whether weld seams are clean or holes are drilled accurately will find it difficult to do this 100% of the time. It’s an inescapably human problem.

Or is it?

If inspection engineers can be supported in their work by artificial intelligence, they could certainly succeed. But how?

PTC's new Vuforia Step Check solution combines augmented reality (AR) and artificial intelligence (AI) to provide an unprecedented visual inspection tool. Once the tool has been set up for a specific component or product line, it guides inspectors through the inspection process with the help of AR, so that even inexperienced specialists can reliably perform multistep checks.

Making training and quality inspection more intelligent

One of the first Step Check users is Nascote Industries, a division of Magna International, one of the world's largest suppliers to the automotive industry. This industry is facing the challenge of vehicles becoming increasingly complex, resulting in an increasingly complex assembly process.

There are dozens of steps that need to be carried out in a specific order and within a specific time frame. Training at a green table with 2D paper instructions is time-consuming and does not always lead to the desired result.

The process becomes clearer and easier to understand when learners are guided step-by-step to the inspection points on the real object with AR support and instructions - explanations are displayed on the tablet or AR glasses. This can also be done directly during the actual quality inspection.

Consider the example of plug connections: Even if the functional test does not reveal any faults, a connector that is not fully engaged can come loose over time due to vibration while driving and may paralyze the vehicle.

Nascote Industries has already tested Vuforia Step Check during end-of-line inspections of electrical connectors in the field, as well as to speed up training processes. The manufacturer is paying particular attention to such "soft connections". An AI-supported visual inspection prevents these from being overlooked.

How control with AI works

Inspection engineers either start by manually entering an ID number for the part to be inspected or they scan the barcode. An overview of all the steps that need to be completed appears on their tablet. A gray check mark indicates which inspection points have not yet been completed.

A green tick indicates passed inspection points, while a red tick indicates failed checks. Tapping each item in the list displays further information or instructions.

A virtual signpost guides the inspection engineer to the next checkpoint, even if that checkpoint is outside the current field of view. This means that nothing is accidentally overlooked.

During the check, Vuforia Step Check compares a live camera feed with the stored 3D reference model to check whether, for example, components have been fitted correctly. If not, the inspection engineer can obtain further information to either rectify the problem on-site or mark the inspection point as "Failed".

The system can be extremely meticulous here, so that even insufficiently tightened screws, deviating gap dimensions, or minimally slipped drill holes are detected. Almost everything that can be recognized by a visual inspection can be checked - the AI becomes a third eye, so to speak.

The advantages of AI control

While the inspection engineer is testing and checking, Vuforia Step Check is already creating a report. As soon as all inspection points for a procedure have been completed, the report is sent automatically - making work easier for employees. All automatically transmitted reports (which can be enriched with comments on any errors and their causes) are recorded in Vuforia Insights so that managers can see at any time what is going well and what is not.

The software also improves with every use. It learns, if only because individual components are captured by the operators from different angles and in different lighting conditions, which makes the AI even more "seeing". The more often the tool is used, the more accurate the recognition becomes because the algorithm learns with each use.

Vuforia Step Check does not require months of programming and training. The initial effort is limited: To get started, it is sufficient to feed the system with a CAD model already available from a product’s initial design phase.

The Vuforia Editor provides a guided workflow for setting up inspection points and instructions, and adding additional resources to help the inspection engineer who will ultimately use the procedure in the Vuforia Vantage app. The app allows them to view work instruction procedures in the environment in which they are performing the tasks.

Companies like Nascote Industries that integrate such a solution into existing quality control processes can increase throughput on existing products and shorten ramp-up times on new ones - while achieving an extremely high level of quality. This is because test errors are largely prevented, thus further increasing customer satisfaction. It is also essential that costs due to rejects, rework or recalls are reduced.

Finally, the data collected on the store floor can be analyzed more easily so that future processes and products can be improved more quickly. As you can see: AI is becoming tangible. It has arrived in the production halls.