Machine Learning Advances Precision and Efficiency in Industrial Laser Metal Processing

Researchers at the Swiss Federal Laboratories for Materials Science and Technology (EMPA) have developed machine learning techniques to optimize laser-based metal processing, making it more precise, cost-effective, and accessible. Traditionally, laser processes such as welding and 3D printing (specifically, powder bed fusion or PBF) require extensive preliminary experiments to determine the optimal settings for each batch of material, which is both time-consuming and resource-intensive. The EMPA team, led by Giulio Masinelli and Chang Rajani, used machine learning algorithms trained with data from optical sensors embedded in laser machines. These algorithms can identify the current welding mode and suggest optimal settings for subsequent tests, reducing the number of required experiments by about two-thirds without compromising quality. This innovation could allow non-experts to use PBF devices, broadening industrial access. Furthermore, the researchers extended their approach to real-time optimization of laser welding, employing field-programmable gate arrays (FPGAs) for rapid data processing and control. This system allows for real-time adjustment of laser parameters, which was previously impossible due to the speed and complexity of the process. The team believes that continued development of these AI-driven methods will further enhance the flexibility and reliability of industrial laser processing, benefiting sectors such as automotive, aerospace, and medical technology.