Artificial intelligence (AI) is also rapidly establishing itself in photonics. As the person responsible for research and development projects at the Precitec Group, Dr. Markus Kogel-Hollacher is one of the protagonists of this trend, and as Executive Director of the Laser Technology Working Group (AKL e.V.), he witnesses how its members from industry and research are increasingly driving forward AI projects. In this PHOTONICS interview, Kogel-Hollacher explains how Precitec is using AI to create tangible customer benefit, why training algorithms is a tough but rewarding task, and how cleverly applied AI is also making the invisible visible. The old dream of autonomous production processes based on closed loops is becoming a reality.
Dr. Markus Kogel-Hollacher: I’d be glad to. Since its foundation in 1971, Precitec GmbH & Co. KG has focused on sensor technology and processing tools for laser material processing. Our modern sensor technology for laser cutting heads goes back to a 1968 patent that describes a capacitive distance sensor for cutting tools. Today, Precitec is one of the leading global suppliers in this market and has set up a second site in Neu-Isenburg, Precitec Optronik GmbH, which develops, manufactures, and sells sensor technology for the semiconductor and glass industry, medical technology, consumer electronics, and the automotive industry with the growing e-mobility sector. From these origins, Precitec has matured into a globally active group of companies with more than 750 employees. As a family-run company, we are independent and agile. This is the key to us still being an innovation and market leader today in our core areas of laser and 3D measurement technology. It is an advantage that decisions to enter into new technologies and business fields are made quickly if we can convince our owners to do so.
Kogel-Hollacher: We have been working intensively on industrial digitalization and networking for around a decade, and are consistently expanding our development approaches in this area. We have, for example, set up a Precitec Incubator at the Karlsruhe Institute of Technology (KIT), where a team of data scientists is developing digital concepts for the laser processes and measurement tasks of tomorrow and beyond in a startup-like atmosphere. They use the entire arsenal of modern IT: the Industrial Internet of Things (IIoT), machine learning (ML), artificial intelligence (AI), and cloud computing. That helps us, and is at the same time an excellent opportunity for talents to test their skills against real industry requirements and challenges. We do not require them to have any prior knowledge of photonics or laser material processing. Instead, the “Precitec Incubator” is fully focused on data analysis. We send the team process data and let them know what information we would like to obtain from it. We then give them free rein since they can better assess which methods and algorithms will lead to the goal. In the meantime, we can say that the digitalization of the Precitec Group has been a success.
Kogel-Hollacher: We have always developed the hardware for recording and processing measurement data at both of our sites. Our products for laser material processing consist of the actual production tool and the sensor technology, which monitors both the process and the tool itself, whether for laser cutting, laser beam welding, additive processes, or other laser-based processes. To that end, we use photodiodes, camera systems, or optical coherence tomography (OCT). It’s about quality, production that is as error-free as possible, and ensuring that defective products do not reach end customers. Over the years, dedicated sensor systems have emerged for the various processes. Our Laser Welding Monitor LWM records the main emissions from the processing zone during laser beam welding. That allows a large number of industrial users to make clear quality statements about their production processes. We have decades of experience in processing sensor data efficiently, quickly, and effectively for the respective application...
Kogel-Hollacher: Exactly. We assumed that we could use modern Al models to extract information from these emissions that would provide our customers with an even more accurate picture of their laser processes and the causes of any quality fluctuations. During application, the question has gradually arisen as to whether we can ultimately make accurate physical statements, for instance, about welded joints. In other words, whether, for example, we can draw conclusions about the tensile strength or even the electrical contact resistance of a welded joint from the optical data. If so, it would be a big step toward robust and error-free processes. That is exactly the path we are currently on, and we are cooperating with selected customers.
Kogel-Hollacher: You raise an important issue. An AI must be trained. While a few dozen welding processes were sufficient for quality assurance based on our LWM to define process limits, we now need to evaluate several hundred data records to train the AI models. In this case, “evaluation” means destroying every weld, making cross-sections, and labeling data. Creating a regression with the aim of modeling the relationships between a dependent and one or more independent variables, requires the complete, meaningful parameter field to be recorded. That sometimes involves a lot of manpower and, depending on the product, can also be extremely cost-intensive. But the effort is worth it: Because the data comes from the real process environment, AI models are extremely robust and reliable. And over the years, I have found that there is a great willingness among AI enthusiasts to go that extra mile to discover more hidden information.
Kogel-Hollacher: There are application scenarios in which extremely fast object recognition in near real time is essential. Think of self-driving cars, whose AI has to detect several objects – traffic lights, pedestrians and other road users, or any obstacles – at once, and classify them in real time to be able to make the right decisions at lightning speed. In view of that, it is probably only a matter of time before AI-supported real-time process controls will also be available in the manufacturing environment. That does, however, raise the fundamental question of whether the tools and processes will also be real-time capable. Years ago, Precitec was a pioneer in implementing OCT as an industrial sensor method. We thus created the possibility of measuring the welding depth in laser processes, controlling it in real time, and keeping it constant. That was pioneering metrological work. Nevertheless, it proved more difficult to convince customers. Unfortunately, the process has not yet managed to gain widespread acceptance. However, AI-based real-time process control will be possible very soon.
Kogel-Hollacher: Closed loop solutions already exist for certain tasks in the context of laser material processing. They are available wherever it is possible to measure real physical quantities, for example, using OCT. In addition, numerous processes in the context of additive manufacturing or laser surface hardening are controlled on the basis of pyrometric temperature measurements. The providers are marketing it as autonomous process management, and not without good reason. However, whether there will ever be a universal laser machine that autonomously produces fault-free components right from the first pass is currently more of a research issue.
Kogel-Hollacher: Every manufacturing process based on photons that penetrate a workpiece and develop their specific effect there harbors uncertainties. In order to achieve the desired result in the required quality, all key process input variables must be meticulously adhered to. Deviations inevitably result in quality fluctuations, which, however, remain invisible to the customer without sensors. This is precisely where we at Precitec come in with our AI models. It is no longer just a question of “good/bad”. Instead, we use AI to elicit precise statements about physical properties from optical data from the process, whether it’s the strength of a seam in kilonewtons, or the contact resistance of a weld seam in microohms. The latter represents real added value, especially in the production of batteries or fuel cells. However, the use of AI also makes sense in the inline monitoring of fast laser welding processes in consumer electronics with outputs of one component per second. When nanosecond lasers weld seams there, AI supports the data analysis of the high-frequency sensor data collected by the LWM in order to identify any errors, such as deviating gap dimensions, a shift in the laser focus, or excess welding filler material, from the emissions at the welding point.
Kogel-Hollacher: For more than 25 years now, my area of responsibility at the Precitec Group has been the acquisition of joint public sector projects. We actively get involved in funded research at national and European level, and help shape it. The aim is to achieve short paths from basic to applied research and on to industry so that innovations can be launched quickly on the market. We alone at the Precitec Group are currently involved in funded projects that deal with the use of AI in the areas of laser cutting, direct energy deposition (DED) with powder, and battery development. The field of application and potential photonic applications of AI are so broad – and the development is so dynamic – that any attempt to name even a few here is doomed to fail from the start. AI is currently conquering markets at breakneck speed. Anyone who still isn’t looking into it should do so as soon as possible!
Kogel-Hollacher: Our Working Group currently includes around 170 people, almost all of whom work in the photonics industry. Many of us have a past at the Fraunhofer Institute for Laser Technology ILT in Aachen. And we often cross paths at specialist conferences such as the recent “AI for Laser Technologies,” the World for Photonics Congress in Munich, or the AKL – International Laser Technology Congress (April 17 to 19, 2024 in Aachen), at which our working group presents the renowned Innovation Award Laser Technology. A fairly clear picture now emerges: AI methods will change everyday life in the photonics industry. The range of applications extends from the automated design of optical systems and components, and the optimization of conventional, additive and subtractive laser processes in manufacturing, to process control and the aforementioned real-time control of autonomous manufacturing processes. At Precitec Group, we are currently rolling out the first industrial solutions and are convinced of their success. AI represents an opportunity for photonics. It will vastly expand the possibilities of what can be achieved with light.