Tool wear prediction based on XGBoost feature selection combined with PSO-BP network

Area of Science:

• Manufacturing Engineering
• Machine Learning
• Signal Processing

Background:

• Accurate tool wear prediction is crucial for optimizing CNC machining processes and reducing operational costs.
• Existing methods often struggle with small sample sizes and selecting optimal input features and parameters for predictive models.

Purpose of the Study:

• To develop a robust tool wear prediction method for CNC machining, particularly effective in scenarios with limited data.
• To enhance the accuracy and efficiency of tool wear state identification by integrating advanced feature selection and parameter optimization techniques.

Main Methods:

• Utilized XGBoost for feature selection and Particle Swarm Optimization (PSO) to optimize a Backpropagation Neural Network (BPNN) within a double-layer programming model.
• Preprocessed CNC machining signals (vibration, cutting force) using time-domain segmentation, Hampel filtering, and wavelet denoising.
• Extracted time-domain, frequency-domain, and time-frequency domain features, followed by screening using Pearson correlation and XGBoost feature importance.

Main Results:

• The proposed XGBoost feature selection reduced model construction time by 57.4% and increased prediction accuracy by 63.57%.
• PSO demonstrated superior performance in optimizing BPNN parameters compared to other algorithms for tool wear prediction.
• The method achieved high accuracy in predicting tool wear states, outperforming traditional methods like Decision Tree, Random Forest, Adaboost, and Extra Trees.

Conclusions:

• The combined XGBoost and PSO-BP network offers an effective solution for tool wear prediction in CNC machining, especially in small sample scenarios.
• This approach contributes to improved production efficiency, reduced tool replacement frequency, and lower maintenance costs in industrial settings.
• The findings provide valuable insights for enhancing predictive maintenance strategies in automated manufacturing environments.