Physics-informed interpretable machine learning method for DOC monitoring in peripheral milling

Online monitoring of depths of cut (DOC) is an essential way to avoid machining defects, such as over-cutting and machining chatter. Data-driven machine learning method has been widely used to build the monitoring algorithms. However, deficient training data and un-interpretable algorithm make it difficult for application. Therefore, a physics-informed interpretable machine learning algorithm was proposed. Firstly, a physical simulation model incorporated with DOC was established with the rotation speed and feed speed as its inputs and time-domain signals of milling force as its outputs. The output force signals were quantitatively presented by time domain and frequency domain features. Secondly, six dimensionless features, namely the kurtosis, skewness, waveform factor, peak factor, pulse factor, and margin factor of the resultant milling force, were explored through sensitivity analysis method. They were sensitive to DOC but insensitive to milling force coefficient, speed, and feed speed. Then, a quantitative relationship model between features and DOC was established using the least squares linear regression algorithm, which has an intrinsic interpretability. Finally, the model was trained by a labeled 100-group experimental data. The results show that the accuracy of the proposed model for DOC monitoring is higher than 90%.