Laser machining process monitoring using stethoscope and neural networks

This paper presents a novel in-situ monitoring technique and deep learning models for laser machining process condition identification and quantification using acoustic wave. Laser machining is a thermal process that removes atoms from a workpiece and generates short pulses induced by the interaction between the laser and target material. Short pulses were collected using a digital stethoscope placed on a 525 μm-thick silicon substrate. The proposed use of stethoscope allows real-time monitoring, analysis, and control of the laser machining process, thereby facilitating accurate pattern parameters (depth, width, and surface microparticulate debris around the heat-affected zone). In this study, the neural network models were implemented for the qualitative and quantitative identification and classification of laser-machined surfaces. Specifically, several signal processing methods (fast Fourier transform, short-time Fourier transform, log scaling, and downsampling) were applied to the time-series stethoscope data. Subsequently, neural network models were developed to predict the pattern parameters (depth, width, and surface debris). Convolutional neural networks, fully connected layers, and a stacking ensemble network were integrated to improve the prediction performance. The developed model successfully predicted the presence of debris with 99.63 % accuracy and estimated the depth and width of the pattern with errors of 43.26 and 4.47 μm, respectively. This neural network model can be used to prevent machining beyond material boundaries by automatically estimating the laser pattern dimensions and minimizing the surface microparticulate debris.