|
Real-time monitoring of the laser surface texturing is very important to maintain the process in control and identify any geometrical deviations of the surface features from their referenced/predefined values. In this research, a novel compact in-line/in-axis monitoring system is reported. The system employs light diffractometry principle to extract geometrical information about the laser generated surface topographies and thus to judge if the process is in control. A collimated white light is focused through the laser beam delivery sub-system that integrates beam deflectors and a telecentric lens onto the workpiece. Then, the reflected light from the textured surface is sent back through the same optical path to a spectrometer. As the collimated white light is diffracted by the periodic surface structures, only the 0-order diffracted light is reflected to the spectrometer. The reflected spectra are dependent on the periodicity, depth, and amplitude of the surface features as they diffract the focused white light beam. Thus, by capturing the spectra from fields processed under different conditions, especially one with zero focal offset distance (FoD) and others with a varying FoDs, it can be determined if such processing disturbance is present. Then, the collected data is used to train machine learning (ML) classifiers, which can automatically detect the presence of any focal offset during the laser texturing operations. In this regard, decision tree (DT) ML classifier is trained. The obtained results show that significant dimensionality reduction and high levels of classification accuracy can be achieved using DT, with up to 99% accuracy based on the full reflection spectrum and 93% with reduced spectra. Based on its in-built dimensional reduction capabilities, inherent interpretability, and reduced prediction latencies, DT approach offers unique advantages and can be considered for further inline/in-axis monitoring tasks depending on the specific surface features that should be monitored.
|
