Performance degradation assessment method for linear motor feed systems driven by digital twins

Abstract

Abstract To address the challenges of missing degradation samples and the low accuracy of traditional degradation assessment methods in linear motor feeding systems, and considering the advantages of Digital Twin technology in complex equipment status evaluation and performance prediction, this study introduces Digital Twin into the performance degradation assessment of the linear motor feeding system. To resolve the issue of missing degradation samples, a digital integrated model consisting of the mechanical subsystem, electrical subsystem, and control subsystem is established, based on an analysis of the multi-field coupling mechanisms (mechanical, electrical, and magnetic) of the linear motor feeding system. Normal operating conditions and demagnetization degradation conditions are designed, and the motor output characteristics under multiple conditions are simulated and analyzed. A current simulation dataset for both normal and demagnetization degradation conditions is constructed.In response to the challenges posed by the temporal dependence of operational data and the difficulty of identifying degradation states, and leveraging the superior data transformation capabilities of Gramian Angular Field (GAF), which effectively preserves temporal information, this study proposes a performance degradation assessment model combining GAF encoding with the AlexNet convolutional neural network. The model first converts one-dimensional time-series data into two-dimensional images using GAF encoding, and then utilizes the image recognition capabilities of AlexNet to assess the degradation state of the linear motor feeding system. This approach enables the evaluation of irreversible demagnetization degradation levels, achieving an accuracy of 98.3%.Furthermore, compared to the traditional methods of PNN, GRNN, and LS-SVM, the proposed method shows improvements of 10.3%, 4.9%, and 15.2% in the AUC index, respectively. The average accuracy is improved by 10.5%, 5.3%, and 14.7%, significantly enhancing the evaluation accuracy. This method effectively resolves the issue of insufficient real degradation samples and provides a powerful tool for predictive maintenance and performance monitoring in industrial environments.

Keywords

• Digital twin
• Linear motor feeding system
• Lack of degradation samples
• Deep neural network
• Performance degradation assessment