Tool Wear State Identification Method with Variable Cutting Parameters Based on Multi-Source Unsupervised Domain...

Area of Science:

• Manufacturing Engineering
• Machine Learning
• Signal Processing

Background:

• Accurate tool wear state identification is crucial for machining quality and efficiency.
• Current unsupervised domain adaptation methods often rely on single-source transfer learning, limiting their ability to leverage diverse cutting parameter data.
• This limitation hinders the improvement of wear state recognition performance.

Purpose of the Study:

• To propose a novel wear-state recognition method for variable cutting parameters using multi-source unsupervised domain adaptation.
• To address the limitations of single-source domain adaptation in utilizing multi-parameter sensor data.
• To enhance the accuracy and efficiency of tool wear state identification in machining processes.

Main Methods:

• Utilized non-stationary Transformer encoders for extracting non-stationary common features from sensor data.
• Employed sliced Wasserstein distance for domain-specific feature distribution alignment.
• Implemented classifier output alignment to reduce domain shift and simplify multi-domain synchronous alignment.

Main Results:

• The proposed method effectively extracts relevant features from sensor data under variable cutting parameters.
• Domain-specific feature and classifier output alignment successfully reduced domain shift.
• Milling experiments demonstrated the superior recognition performance of the developed method.

Conclusions:

• The multi-source unsupervised domain adaptation approach significantly improves tool wear state recognition accuracy.
• The method effectively handles variations in cutting parameters by leveraging information from multiple sources.
• This research offers a promising solution for intelligent manufacturing and predictive maintenance.