AI-Powered Inspection

Estimated reading time: 3 minutes

Peter Krippner is the COO of Viscom AG, a manufacturer of automatic inspection systems including 3D AOI. Prior to that, he was the manager of software development and for 17 years, head of the company’s Serial Products (SP) business unit.

In an interview with SMT007 Magazine, Krippner discusses artificial intelligence (AI) and its impact in the electronics manufacturing industry.

Stephen Las Marias: What can you say about AI and its role in manufacturing?

Peter Krippner: Whereas practical solutions like optical character recognition have become routine technologies and are rather not seen as AI, a very good example for artificial intelligence is autonomous driving, where individual vehicles make driving decisions self-sufficient and independent. In manufacturing, customers have voiced an increasing interest in operating SMT lines with fewer personnel; more precisely, with a maximum of two persons per line for all associated activities including AOI operation and verification. In such a scenario, AI has the potential to be very helpful.

Las Marias: Which part of the electronics manufacturing process will greatly benefit from AI?

Krippner: From our perspective, first, this is going to be the support and automation of AOI verification and then AOI programming. Different levels of automation will be available, so the customer can also proceed systematically. Initially, the AI can monitor human verification and report any deviating results (e.g., by issuing a message that it would classify a potential error differently than the operator). In Stage 2, the AI can assume more and more responsibility, and finally verify all components fully automatically.

As for AOI programming, we are now working on automatic detection of housing types for the creation of inspection programs. With the help of “deep learning” mathematical methods, such as those derived from neural networks, the assignment of components on a PCB should gradually become largely autonomous.

Las Marias: How will AI change the electronics manufacturing/electronics assembly landscape?

Krippner: In the future—because of the rising difficulty in finding qualified SMT line personnel as well as general cost pressures—there will be either fewer specialists or more personnel with less experience in the AOI area. Nonetheless, first pass yield (FPY) will have to be over 90% right from the start and the number of escaped defects, zero. Preventing defect escape and false calls by personnel during verification then becomes a logical tactic. This is where Viscom steps in with new AI software developments, which should be ready for market over the course of next year.

Las Marias: What key challenges are you seeing in adopting AI in manufacturing?

Krippner: An immense amount of data—actually, big data—is needed to implement AI. The basic technologies and workflow related to AI-assisted verification have already been completely integrated in the Viscom software. Now data—as much actual data from the field as possible—is needed to train the AI algorithms. This training phase is currently in progress. All this is proceeding step by step, since each classifier trained in this fashion must be validated before use. This requires additional data that are not used for the training.

Over the long term, both the AI verification as well as the AOI programming will be automatic, achieving better results than before from the start. At Viscom, we follow the strategy of growing profits through innovation and technological leadership. One very influencing factor is our customers, who we want to continue to enthuse with progressive solutions.

Las Marias: What manufacturing trends will be greatly addressed by AI?

Krippner: Especially at high resolutions due to miniaturization, fast imaging systems and evaluation methods are necessary. Miniaturization will certainly go hand in hand with more computational power, which then will be available for even better inspection solutions. AI implies that machines can learn continuously and then succeed in solving new problems. A key issue is complete automatization. We will see more so-called lights-out factories in the future where human employees are no longer part of the process and machines can therefore produce.

Las Marias: Are you seeing wide implementation of AI in electronics manufacturing right now?

Krippner: Today’s modern electronics manufacturing is Industry-4.0 driven. This explicitly includes Viscom’s intelligently linked inspection solutions. Information is automatically processed and exchanged with other systems on the production line, which then can provide necessary corrections. Machines react autonomously to process fluctuations, exchange information with one another and then adapt their processes. This shows that even though AI is a relatively new trend here, electronics manufacturing is the right industry to successfully advance it.

To read the full article, which appeared in the September 2018 issue of SMT007 Magazine, click here.