Tool wear prediction in ultrasonic vibration-assisted drilling of CFRP: A hybrid data-driven physics model-based framework

Abstract

Accurate tool wear prediction is the key to evaluating tool usability and ensuring drilling quality. Recently, fusion prediction methods have attracted great attentions. However, no similar work has been reported yet in ultrasonic vibration-assisted drilling (UVAD) of carbon fiber reinforced polymer (CFRP). To fill this gap, a hybrid data-driven physics model-based framework for predicting tool wear is proposed. The physics model builds a mathematical description of the tool wear degradation process. The data-driven model predicts tool wear based on multilayer perceptron (MLP). The fusion method fuses the results of both models based on particle filter to obtain the final tool wear. Results indicate that the hybrid model performs better than any of individual models, with the lowest prediction errors.