Enabling in-situ monitoring of ultrashort pulse Laser Surface Texturing

This work is focused on laser texturing using femtosecond lasers. Such a process enables surface treatment for functionalization purposes like coloring or modification of tribological properties or wettability. However, the laser texturing process is sensitive to changes in laser parameters, changes in the nature and material of the textured surface, and also to environmental conditions. In this work, the objective is hence to develop a monitoring strategy to help the development of the process. An instrumentation is first proposed and discussed. The process emits light signals that we propose to track with a spectrometer. The light intensity is characterized across the entire spectrum between 200 and 1100 nm. Our proposal is to use these spectral data acquired over time to train a deep learning model and then to get a prediction of process output and process conditions. Two Resnet models are trained and characterized on experimental datasets of spectro-temporal data. Datasets and models are released as open-source resources. The models are built considering two scenarios of supervision : i) monitoring texturing conditions, specifically fluence and laser spot overlap, and ii) monitoring the gray level of a generated texture. We present two deep learning models. The first achieves R² coefficients of 0.98 for fluence and 0.96 for overlap supervision. The second model estimates the gray level of the produced texture with an average accuracy of 11% of the full grayscale range. The performance of both models is also assessed using data collected with an intentionally introduced laser focus defect. The models successfully detect changes in texturing conditions, demonstrating their robustness relative to focus defects. Finally, a study is performed to identify the more relevant wavelengths in the spectral data. It shows the use of photodiodes at specific wavelengths could constitute a relevant instrumentation approach for supervision in industrial applications.