Real-time control parameter update and stochastic tool wear monitoring framework for nonlinear micro-milling process

Abstract

In modern manufacturing, micro-milling technology encounters challenges such as unpredictable tool wear and dynamic variations in cutting parameters, which adversely affect machining accuracy and safety. This study presents a nonlinear micro-milling mechanical model that combines tool runout, chip separation, stochastic tool wear, and tool-tip trajectory change to accurately predict cutting forces. The Hippopotamus optimization algorithm is introduced to address the particle impoverishment problem in coefficient recognition and improve the real-time update efficiency of the cutting model. Additionally, a DASAT network model combining Recurrent Neural Networks and Convolutional Neural Networks with an attention mechanism is proposed for more precise tool wear prediction, achieving lower prediction error rates compared to LSTM/TCN-based methods. By correlating the predicted tool state with the wear threshold, the system can perform active maintenance interventions to reduce tool failures. The experiment demonstrates that the machining based on the proposed framework can improve surface accuracy while maintaining a stable cutting state, ensure the safety and reliability of the micro-milling process, and provide strong support for process optimization and equipment maintenance.