A MONITORING METHOD OF MILLING CHATTER BASED ON OPTIMIZED HYBRID NEURAL NETWORK WITH ATTENTION MECHANISM

Machining chatter is a self-excited vibration between the cutting tool and the workpiece, which can reduce surface quality and tool life, and even endanger the safety of operators in severe cases. Considering that milling chatter has multi-scale features and the debugging of neural network hyperparameters heavily relies on experience, a milling chatter monitoring method based on an optimized hybrid neural network with an attention mechanism (MISSA-MSCNN-BiLSTM-ATM) is proposed. Firstly, the harmonic of the spindle rotation frequency is filtered out using the spindle rotation frequency removal technique (SFT). Then, an improved sparrow search algorithm (MISSA) is proposed based on multiple strategies including improved circle chaotic mapping, golden sine strategy, and enhanced Lévy flight. Subsequently, MISSA is utilized to optimize the hyperparameters of the milling chatter classification hybrid neural network model, combining multi-scale convolutional neural networks (MSCNN), bidirectional long short-term memory (BiLSTM), and attention mechanism (ATM). In numerical simulations with CEC2005 complex functions, MISSA demonstrates better optimization accuracy, stability, and shorter computation time compared to other intelligent algorithms. Compared with other milling chatter classification models, the proposed method exhibits significant improvements in accuracy and stability.