An Innovative Study for Tool Wear Prediction Based on Stacked Sparse Autoencoder and Ensemble Learning Strategy.

Area of Science:

• Manufacturing Engineering
• Machine Learning
• Sensors and Signal Processing

Background:

• Real-time tool wear prediction is vital for effective tool prognostics and health monitoring in Computerized Numerical Control (CNC) machining.
• Current methods may lack the accuracy and reliability needed for advanced manufacturing processes.

Purpose of the Study:

• To develop and validate a novel integrated deep learning model for accurate real-time prediction of milling tool wear.
• To fuse multi-sensor features from vibration and cutting force signals for improved wear prediction.

Main Methods:

• Extracted multi-sensor features from vibration and cutting force signals in time, frequency, and wavelet domains.
• Utilized a stacked sparse autoencoder (SSAE) with a backpropagation neural network (BPNN) as the primary learner.
• Employed a gradient boosting decision tree (GBDT) regression model with optimized hyperparameters as the secondary learner in a stacking strategy.

Main Results:

• The integrated deep learning model demonstrated significantly improved prediction accuracy compared to single SSAE models and shallow machine learning approaches.
• Feature fusion from multiple domains (time, frequency, wavelet) enhanced the model's ability to capture tool wear patterns.
• The stacking learning strategy effectively combined the strengths of different models for robust tool wear prediction.

Conclusions:

• The proposed integrated deep learning model offers a superior approach for real-time tool wear prediction in CNC machining.
• Multi-sensor feature fusion and ensemble learning are key to achieving high accuracy and reliability in tool health monitoring.
• This method provides a promising solution for enhancing the efficiency and safety of automated manufacturing systems.