Aviation supplier PFW talks about Artificial Intelligence (AI) in industrial X-ray testing

Lennart Schulenburg: How do you see the current state of AI in everyday life and regarding the work in the field of quality?

Markus Gutensohn: On the part of PFW, we have not yet seen any real AI application within our daily inspection routine. There are many established structures in the aviation industry, some of which are decades old. These quality assurance systems were established to ensure safe aviation. Even if the AI ​​technology is available in a timely manner, acceptance and qualification are major challenges that we and the entire aviation industry face.

Lennart Schulenburg: Do you see the qualification and standardization of AI as the big challenge?

Markus Gutensohn: Absolutely. If you imagine that safety, or the understanding of safety, is suddenly placed in the hands of an algorithm, that is a big step. We now use specially educated and trained staff for this, and now we should all such safety-related tasks hand over to an algorithm? It is a possible scenario, but it is currently a giant step and those who are not ready for it should prepare.

We will all learn to deal with it. And trust will grow, but this has to build up over the near future. Employees have to be taken along; old processes have to be changed. Everything should be considered. Historically grown structures have proven themselves in many ways and there are new ways that are likely to prove to be better.

Lennart Schulenburg: Why does PFW want to use AI in quality assurance and what advantage or added value do you expect from it?

Markus Gutensohn: Today PFW has a very large amount of testing efforts. Now we can see that our quality assurance staff can really use the whole data from an increasingly digitized environment. We expect significant added value if data -driven decisions can be made by an algorithm. Therefore, some time ago, as part of a funded program, PFW began to investigate whether digital and automated X-rays as a supporting evaluation could relieve our staff of strenuous activities. Finding small, fine pores on a monitor for six to eight hours a day may be ambitious, but it is undoubtedly a burden.

Lennart Schulenburg: So, for you, AI is more of an assistance system, which supports the user in making faster and better decisions?

Markus Gutensohn: For now yes, because we have to learn to deal with these systems – see and understand how reliably something like AI works. There will likely be a long phase in which AI is used as an assistance system. Important, safety-critical questions first must be answered by trained auditors. Of course, for one or the other the goal is to have an autonomous AI system that reliably makes such decisions itself and can fully relieve the staff of the decisions. Until then, it is first necessary to demonstrate the performance of the algorithms.

Lennart Schulenburg: What experience does PFW have with automation in X-ray inspection?

Markus Gutensohn: PFW has a large number of X-ray systems running because many of our manufactured components, e.g., where we use welding technology, have to be subjected to an X-ray test. By investing in new systems and programs to adapt to the increased production rates, PFW has increasingly procured partially or fully automated X-ray inspection technology from VisiConsult in recent years. We also use classic, analog X-ray inspection technology with film because there are certain requirements, e.g. for an objection recording of a tube which is not possible with a detector. For instance, when a piece of film is rolled up and pushed into the tube for a single wall inspection. Film technology is particularly advantageous for this, but if an application can be carried out digitally, then the analog variant reaches its limits comparatively quickly.

Lennart Schulenburg: Which priority has automation in terms of robotics and AI at PFW?

Markus Gutensohn: This is a growing and increasingly interesting topic, because the products that PFW produces are manufactured parts. It’s about small quantities, high variability, and a long product life cycle. However, in the future we expect a further increase in the number of units, so that intensive checks are now being carried out to determine where and to what scope automation or robotics is economically viable and technologically possible. PFW is increasingly using robotics in production, e.g. to stay competitive. Robotics is also becoming more and more relevant when it comes to testing of components. Here we rely on the systems of our X-ray partner VisiConsult.

Lennart Schulenburg: What improvements are expected by PFW with regard to the quality of materials testing through automation and AI?

Markus Gutensohn: Today we already have a very high level of quality assurance, which enables us to deliver the highest quality very reliably. From my point of view, improving this even further is a great challenge. If at all, the traceability and objectivity of the decisions can be improved by an AI.

Lennart Schulenburg: If you look at accident statistics for self-driving cars, self-driving cars cause a hundred thousand fewer accidents than a human driver.

Markus Gutensohn: Despite the nice statistics: a single accident in the aerospace industry is one accident too many. We know the consequences of human error. But what if an AI causes the accident?

Lennart Schulenburg: The appropriate legal framework must be set for this, but standardization and qualification are also important aspects. VisiConsult therefore works intensively in standardization, so that these AI test processes can continue to develop in a controlled manner through standards.

We work closely with research groups in the aerospace industry on the subject of AI. How do you perceive the acceptance of your customers?

Markus Gutensohn: PFW has to put itself in the position of the customer in the aerospace industry: If the customer is supplied with components today having no quality problems in the past, then it is a difficult and lengthy process to break away from the established standards, albeit gradually. Placing trust in an algorithm will not be a sure-fire success, even under the aspect of the cooperation between human auditors and AI, when e.g., AI pre-sorts everything. There could be skepticism about whether the examiner is still sensitive enough for the final exam. The human factor is simply extremely important.

On the other hand, I already mentioned at the beginning: You can now clearly see that there are many processes that provide you with data. And our customers will see the possibilities of data mining in order to benefit from it themselves. It is on everyone’s lips and I believe everyone will learn, see and understand for themselves where their real benefit is. Many are currently on the way to find out in order to make data more usable for themselves. The process is now accelerating.

Lennart Schulenburg: You just mentioned data mining – process improvements based on data. Do you already draw conclusions about the welding quality or changes in welding parameters from your checked data? Do you see potential in AI and closely interlinking the process?

Markus Gutensohn: We recently dealt intensively with this topic and will look at it even more intensively from next year as part of a project and within a consortium. The aim is to compare process data and test results in order to show correlations that are otherwise not visible and to draw the right conclusions from them. In the end, of course, it is about reducing the testing effort. We are only really getting started now because we have not yet had any approaches in quality assurance to use AI. The topic of digital twins would then be a single process. But the big goal is of course to completely map the entire process chain.

Lennart Schulenburg: When do you expect a breakthrough in the use of AI in the German quality department of the aerospace industry?

Markus Gutensohn: PFW is an innovative supplier to the aviation industry, and when I consider that I believe we still have a long way to go before AI actually makes its way into quality assurance. It will probably take some time in the industry, but we are now taking a first, but crucial, step with VisiConsult. It is a representative example carried out at a point where it can really be said that there is a lot to be gained. If we manage to develop a solution that increases efficiency in a relevant manner, then it will certainly also work for other areas that are being examined. I assume that this is a process that will take a few years.

Lennart Schulenburg: How long have you been working with VisiConsult and what are your experiences so far?

Markus Gutensohn: With VisiConsult we started to switch to digital X-ray ten years ago. It was successfully approved by our customer Airbus. With this system, we have not only taken the first step into the digital world, but for the first time also automated the entire test using digital test programs. The collaboration has intensified over the past few years and a highly innovative robot solution from VisiConsult has been introduced. The consequence now is to take the data one step further. The bottom line is that it is a successful and constructive cooperation, so that we can always plan the future together. VisiConsult’s self-evident fact that it is always eager to take up and implement customer requirements and wishes during the development phase is also positive. It is an important point for us that VisiConsult not only masters standard systems, but that developments are driven forward according to requirements. Yes, it works very well. Thanks for that.

Lennart Schulenburg: Why was VisiConsult selected as a partner to develop an AI solution?

Markus Gutensohn: Because of the growing cooperation and trust in VisiConsult. Based on the approach that we have successfully developed an automated X-ray inspection technology together and that we are now happy to continue working together with the digital data and its evaluation. From my point of view, it is a logical step to pursue this next evolutionary stage with VisiConsult.

Lennart Schulenburg: What would you advise us to focus on and not underestimate?

Markus Gutensohn: As soon as the first viable AI version is available, it must be presented to a selected group of customers so that the group of customers can get more contact with the topic and learn why it is relevant for their own future. But the feedback is even more important: Such a round table brings a lot of new knowledge that will give the next AI version an immense development boost.

Lennart Schulenburg: How is your team’s acceptance of AI?

Markus Gutensohn: It seems to be clear to everyone in this field that this is how we have to position ourselves. We have to master the future while increase production rates somehow. In order to do this, the team is convinced that it will need appropriate support software – and it’s no secret that Quality Assurance has a hard time finding new colleagues when it matters. So, it is important to consider every possibility.