In recent years, the incorporation of process monitoring systems has become an integral part across several laser material processing applications. This is due to the feasibility of inline evaluations, which enable the elimination of downstream quality checks that are typically time-consuming and costly. Simultaneous to recent sensor developments, ultrashort-pulsed (USP) laser material processing has become a promising technology for creating complex microstructures in a wide variety of materials due to its high precision and negligible thermal input. So far, process monitoring has yet to be established in USP applications as it is unknown which sensors are suitable, which parameters can be observed, and which process-relevant information can be detected within the data. Within this work, a state-of-the-art photodiode-based process monitoring system was used to collect spectral data from the process zone and extract relevant information from the process. The laser structuring of the two-layered diffusion media used in polymer electrolyte membrane fuel cells was selected as the use case. This structuring process is challenging due to production-related fluctuations in the material thickness and the surface quality. Within this study, information was acquired from a sensor system employing three photodiodes. Through the application of explorative data analysis techniques and machine learning, AI models were created, focusing on determining the focus position of the laser beam and identifying the transition between the two layers. The developed models were capable of determining the initial focus position with varying accuracies. The best performance was achieved by a convolutional neural network using spectrograms as input data, while the worst accuracy was achieved by a ridge regression model using manually selected features. Furthermore, a clustering model was used, demonstrating the capability to detect the layer transitions within a specified tolerance.
