Cutting force prediction model based on PSO-VMD and LSTM

The rapid development of artificial intelligence technology has brought opportunities for intelligent transformation in the manufacturing field, especially in the modeling and prediction of high noise and nonlinear industrial signals, demonstrating unique advantages. A cutting force prediction model based on a hybrid multi model neural network is proposed to address the issues of strong non stationarity and mode mixing in cutting force signals. The model first uses particle swarm optimization algorithm and variational mode decomposition to decompose the cutting force signal, and predicts the cutting force through a hybrid multi model neural network. The experimental results show that the loss function of the prediction model rapidly decreases with increasing iteration times and then gradually stabilizes. The training set loss value rapidly decreased from about 0.18 to around 0.02, while the test set loss decreased from about 0.19 to around 0.03. And the model performs well in terms of average absolute error, mean square error, and root mean square error, with values of 0.69N, 0.84N ², and 0.91N, respectively. The determination coefficient is 0.97, and the shortest training time is 92.5 seconds. The research conclusion indicates that the proposed model can effectively capture the complex characteristics of cutting force signals, providing a new method for high-precision prediction of industrial cutting forces.