Journal
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
Volume 72, Issue -, Pages -Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TIM.2023.3273683
Keywords
Feature extraction; Predictive models; Mathematical models; Monitoring; Data models; Training; Physics; Online learning; physics-assisted network; smart manufacturing; tool wear prediction
Ask authors/readers for more resources
This article proposes a physics-assisted online learning method for tool wear prediction, which combines knowledge from data and physical domain and adjusts the offline basic model with online data. The method has been found to have the lowest mean square error and time complexity, demonstrating good accuracy and generalization.
Tool wear prediction is critical in machining because worn tools lead to poor precision and quality in production. Conventional tool wear prediction methods include physics-driven methods, which are typically independent of real-time working conditions, and data-driven methods, which are limited by the amount of training data. To address these issues, this article proposes a physics-assisted online learning method for tool wear prediction. The novelties of this method are that it not only utilizes knowledge from data and physical domain but also adjusts the offline basic model with online data. In this method, features extracted by the stacked sparsity autoencoder (SSAE) are combined with measured and physics-generated labels as the training set. A multilayer perceptron (MLP) is trained to get physical model coefficients via working parameters to get pseudo labels under testing conditions. Then, partial and global MLPs are integrated as the basic prediction model. Finally, pseudo labels and real-time signals are treated as the online updating set for basic model calibration. It is found that the proposed method has the lowest mean square error (MSE) 42.0069 and time complexity. The repeated experiment is done to verify this method further. These results show the good accuracy and generalization of this method.
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