Prediction of tool wear by improved deep belief networks based on multi-sensor signals

Tool wear online monitoring technology is crucial for improving machining efficiency and product quality. This article proposes a tool wear prediction method using Particle Swarm Optimization (PSO) enhanced Deep Belief Network (DBN) to enhance prediction accuracy. Firstly, features are extracted from sensor signals through time-domain, frequency-domain, and time-frequency domain methods. Subsequently, a correlation analysis method is employed to identify features highly correlated with tool wear. Finally, the effectiveness and performance of PSO-DBN model are validated through cross-validation methods. Research results indicate that the PSO-DBN model can effectively and accurately predict wear values for all tested tools, showcasing robust predictive capabilities. Compared to several popular models and state-of-the-art works in the field, the PSO-DBN model exhibits smaller Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and larger R-squared (R2), highlighting its accuracy and reliability in tool wear prediction.