Journal
JOURNAL OF INTELLIGENT MANUFACTURING
Volume 34, Issue 8, Pages 3365-3387Publisher
SPRINGER
DOI: 10.1007/s10845-022-02005-z
Keywords
Tool wear prediction; Cross-domain adaptation; Gaussian mixture model; Attention mechanism; Joint distributed adaptation
Ask authors/readers for more resources
On-machine monitoring of tool wear is crucial for improving machining efficiency and reducing cost. However, the complex data distribution and feature representation pose challenges. This paper proposes a novel model called CDATT, which utilizes data filtering, attention mechanism, and joint distribution adaptation regularization to enhance cross-domain tool wear prediction performance.
On-machine monitoring of tool wear is of great significance in machining processes to enhance machining efficiencies and reduce cost. For the data-driven tool wear prediction, the consistent feature distribution is critical to the machining learning model. In CNC machining of complex parts, the data distribution is complex due to the time-varying operational parameters. Even the considered as perform well of transfer learning methods may not mine consistent feature distribution representations and may not explain the transferable components to break through the bottleneck of prediction performance. The attention mechanism has shown its unique advantages with the adaptive weighted structure to capture the specific feature in tool wear state prediction. In order to further capture the specific characteristic frequencies and determine the specific transferable components for different working conditions and enhance the performance of cross-domain tool wear prediction, we propose a new cross-domain adaptation network based on attention mechanism(CDATT). Firstly, a data filtering method based on the singular value decomposition and mixed Gaussian model is proposed for removing the idling signals to avoid the effects of invalid samples adaptively. Then, the attention mechanism is introduced to build a prediction network for highlighting the specific tool wear frequency and making the wear characteristic more significant. Next, the joint distribution adaptation regularization term is utilized to construct a network loss function to build the CDATT model for different working conditions. Finally, experiments are carried out on a machine tool to verify the effectiveness of the proposed method. The result shows that the proposed CDATT model can capture the specific tool wear frequency and identify the tool wear state accurately under different machining parameters.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available