Monitoring of ultrashort pulse laser surface texturing using spectroscopy and deep learning model

Laser surface texturing is a process enabling surfaces to be functionalized with a wide variety of functions such as coloring, tribological manipulation, or hydrophobicity. Process monitoring can enable the detection of variations in texturing conditions, which are sensitive and vary with changes in laser parameters, the nature and material of the textured surface, or environmental conditions. The laser texturing process produces a set of luminous signals that we propose to use to monitor the texturing performed using a femtosecond laser, here to generate a variable grayscale value. We instrument the texturing process with a spectrometer to track the intensity of the different signals emitted by the process. We then propose a monitoring method based on the use of a deep learning model that we assess for two monitoring approaches. Datasets and models have been developed and made publicly available as open-source resources. In one approach, focus is given on the monitoring of texturing conditions, i.e. the fluence and overlap. In the other, monitoring of the process result, namely the gray level of the texture, is achieved. We propose a deep learning model capable of monitoring fluence and overlap with R 2 coefficients of 0.98 and 0.96 respectively. The second model estimates the gray level of the texture produced with an average accuracy of 11% of the full gray scale. We also quantify the performance of these two models on data acquired with a deliberately introduced laser focus defect. The proposed models can detect changes in texturing conditions, demonstrating their robustness to focus defects. Finally, the sensitivity of models to the spectrum content is assessed. It outlines a simple instrumentation based on a set of photodiodes could be considered, to only measure the backscattered light and the main rays of the plasma spectrum.