Multi-Sensor Heterogeneous Signal Fusion Transformer for Tool Wear Prediction

Area of Science:

• Manufacturing Engineering
• Signal Processing
• Machine Learning

Background:

• Tool wear monitoring is crucial for manufacturing efficiency and safety.
• Heterogeneous multi-source sensor signals present challenges for effective data fusion.
• Existing methods struggle with integrating diverse sensor data for precise wear prediction.

Purpose of the Study:

• To propose a novel Multi-Sensor Multi-Domain feature fusion Transformer (MSMDT) model.
• To achieve precise tool wear prediction by overcoming challenges in multi-source sensor signal fusion.
• To enhance the integration of heterogeneous sensor data for improved wear signature analysis.

Main Methods:

• Developed a physics-aware feature extraction framework for time-domain, frequency-domain, and wavelet packet features.
• Constructed a unified feature matrix to integrate complementary characteristics of heterogeneous signals.
• Employed a position-embedding-free Transformer architecture for adaptive cross-domain feature fusion.

Main Results:

• The MSMDT model demonstrated superior performance in tool wear prediction.
• Experimental results validated the model's effectiveness on the PHM2010 dataset.
• The proposed method outperformed state-of-the-art approaches in prediction accuracy.

Conclusions:

• The MSMDT model offers an effective solution for tool wear prediction using multi-source sensor data.
• Innovative feature engineering and cross-modal attention mechanisms are key to the model's success.
• The approach provides a robust framework for integrating heterogeneous signals in industrial monitoring.