How is artificial intelligence used in laser processing?
Although some see artificial intelligence as the stuff of science fiction movies, others already use it to optimise production. Indeed, it has already proven itself in practice, above all when used with lasers.
The Fraunhofer Institute for Laser Technology ILT in Aachen, Germany, has been working on artificial intelligence (AI) for several years. To develop and adapt the very complex AI processes, the Aachen-based institute has set up its own AI laboratory. Here, users experiment with their own data using prepared modules: Not only can they can test established methods of machine learning, but also practice deep learning with deep neural networks. These users then receive an explanation of the results, which is also made understandable for AI laymen and thus helps them make well-informed decisions for their applications.
Predictive maintenance and advanced weld monitoring
In the AI lab, users learn how to reliably predict how production systems behave in order to better plan their maintenance and servicing. Such AI systems make the dream of 'predictive maintenance' come true in various manufacturing applications – by reducing downtime and detecting errors early long before these issues leave the permissible tolerance range. This reliable prediction helps create production systems that come closer and closer to the ideal of absolute error-free production. The first AI applications in laser welding for automotive engineering or microjoining for battery technology have proven that this is not science fiction. In combination with image-based monitoring systems, it is already possible to analyse weld seams in real time and reliably classify their quality with adapted algorithms.
The Fraunhofer ILT’s Process Control and Systems Technology groups are also active in this field, using machine learning to investigate how so-called unambiguous statements can be made to detect errors in laser welding with absolute reliability. This is a focus of scientist Christian Knaak, who is investigating machine learning algorithms and comparing how errors can be detected when lasers are used in materials processing. According to the mechanical engineer, the challenge for the algorithms is the high speed of laser processing: 'For this reason, the algorithms must run on correspondingly powerful hardware. To implement such systems, you need a great deal of time, easily several weeks.'
And yet the effort has been worth it: The intelligent process sensor technology developed at Fraunhofer ILT, to name just one example, enables users to divide the quality of the weld seam into five categories during the laser welding process. In the laboratory, the accuracy of the classification result agreed with the measured quality of real welding seams to a very high degree. The Aachen-based institute has further developed the methodology and is currently using it for the laser welding of battery cells. Here, the data from a photodiode is evaluated using machine learning. Knaak explained: 'The differences in the data from faultless or defective welds visible at first glance are often very small and difficult to quantify. Thanks to this new methodology, we have made the statement – whether a weld is ok or not – significantly more reliable. With this solution we can now show that the AI application works and is feasible.'
Just as innovative as research institutes, industrial companies are already using AI systems. Not only does the amount of data (big data) play an important role, but so does its quality, according to Stephan Schwarz, team leader of Data Analytics & Artificial Intelligence at automotive firm Daimler in Sindelfingen, Germany. At the 'AI for Laser Technology Conference' at Fraunhofer ILT in November 2019, he emphasised the need to carefully prepare and cleanse data before analysing it in-depth. This is particularly important, he said, when data from different sources is involved – such as laser systems, production machines or camera monitoring systems.
At the first 'AI for Laser Technology Conference' in Aachen, 70 experts from industry and research learnt that the introduction of artificial intelligence requires not only technology, but also tact in dealing with employees. (Image: Fraunhofer ILT)
Preparing and cleaning up big data, however, is not enough. It creates large volumes of data that need to be processed quickly or, ideally, immediately. However, according to Stephan Gillich, director of Artificial Intelligence and Technical Computing GTM at Intel Germany, this requires very fast real-time information systems with accelerated hardware and software architectures, which are being developed at Intel, for example.
Yet this is not just the advertising jingle of a hardware manufacturer, since dealing with big data does indeed require very fast computers and algorithms. A visit to the ultrashort pulse (USP) laboratory of Fraunhofer ILT demonstrates the enormous amounts of data generated, for example, during material processing with ultrashort laser pulses. Laser processes with pulses in nanosecond or even femtosecond cycles generate a great amount of data within a few seconds, data that can only be evaluated and assessed in a high-quality big-data analysis when fast computers and algorithms can interact productively. Visitors to Fraunhofer ILT can see how this works in real time on a digital display directly from the cloud during the demonstration of a USP laser. These examples show how many factors the success of AI in laser use depends on.
Complex real time data acquisition is needed to use AI in fast manufacturing processes – such as here in an ultrashort pulse laser process at Fraunhofer ILT. (Image: Fraunhofer ILT)
Increasing industry interest
Interest from industry was correspondingly high for the first 'AI for Laser Technology Conference' in November 2019, which was fully booked with 70 participants. Here, practitioners learned what can be done with AI and what needs to be taken into account when introducing it into production. At the conference, there were often questions about applications for quality assurance – for example, by optimising processes or analysing the causes of errors.
But, many speakers made it clear that the use of AI does not only require IT technology. What is also in demand, according to expert opinions, is the dialogue between experts in both data and processes. Moreover, the use of new technologies such as AI requires not only communication between experts but also tact and sensitivity in dealing with employees. In this context, Dr Benjamin Kreck, CTO Intelligent Cloud at Microsoft, said that companies need to be able to adapt and transform their culture, which is certainly crucial to the success of introducing new technologies like AI.
Clearly, there is great potential for using AI in laser processes. In Aachen on 13 February, ILT scientist Christian Knaak kicked off the LSE – Laser Symposium Electromobility 2020 with a lecture on 'Comparison of Machine Learning Algorithms for Fault Detection during Laser Material Processing.' These statements and the others above show there is more than enough interesting 'material' for a second edition of the 'AI for Laser Technology Conference,' to which Fraunhofer ILT is inviting members of academia and industry back to on 4 and 5 November.