Performance comparison of machine learning models for kerf width prediction in pulsed laser cutting

This study aimed to compare the performance of three machine learning (ML) models, including support vector regression (SVR), random forest (RF), and extreme learning machine (ELM) for kerf width prediction of pulsed laser cutting. Selected features from the optimal base wavelet transformation of vibration signals from the optimal base wavelet selection were adopted as the inputs to the ML models. Averaged kerf width of a straight cut of a 0.1 mm thickness silicon steel sheet was chosen as the output. The performance comparison of three ML models was divided into two stages. In the first stage, the effects of varying the validation data size and data randomness analyses were investigated using training data. In the second stage, the prediction accuracy of these machine learning models on testing data was compared. The results from the first stage revealed that the RF model emerged to be the best model in the validation data size and random state analyses with averaged mean average percentage error (MAPE) scores being of 5.32% and 7.61%, respectively. Compared with the SVR and ELM models, the RF model had the least discrepancy between the MAPE scores, training (2.83%) and testing (1.69%), in the second stage of analysis. This indicates that the selected vibration features from the optimal base wavelet selection combined with the RF model are efficient for forecasting the straight kerf width of the workpiece by pulsed laser cutting.