Title: Intelligent condition monitoring of PMSM-based drive systems using a Swin-CNN fusion network for manufacturing applications

Authors: Hao Yu; Hao Zhang; Xiaojing Liu

Addresses: School of Mechanical and Electrical Engineering, Beijing Institute of Graphic Communication, Beijing, 102600, China ' School of Mechanical and Electrical Engineering, Beijing Institute of Graphic Communication, Beijing, 102600, China ' School of Mechanical and Electrical Engineering, Beijing Institute of Graphic Communication, Beijing, 102600, China

Abstract: Permanent magnet synchronous motors (PMSMs) are widely employed in advanced manufacturing and mechatronic systems, where early detection of mechanical faults is vital for system reliability and productivity. This paper presents a novel current-based radial misalignment diagnosis framework for PMSM-driven systems using a dual-branch Swin-CNN fusion network. The proposed method integrates time-frequency and spectral domain analysis through continuous wavelet transform (CWT) and fast Fourier transform (FFT), respectively. A two-stream deep learning architecture is introduced, combining a 2D Swin Transformer for global time-frequency feature modelling and a 1D CNN enhanced with convolutional block attention module (CBAM) for frequency-domain emphasis. Experimental results on a lab-scale PMSM setup demonstrate an average classification accuracy of 98.125% under varying fault levels. The proposed approach offers a cost-efficient, non-invasive solution for condition monitoring in intelligent manufacturing systems, supporting predictive maintenance and enhanced equipment availability in Industry 4.0 environments.

Keywords: PMSM; permanent magnet synchronous motor; sliding window attention network; CNN; convolutional neural network; attention mechanism; fault diagnosis.

DOI: 10.1504/IJMMS.2025.150065

International Journal of Mechatronics and Manufacturing Systems, 2025 Vol.18 No.2, pp.160 - 181

Received: 07 Jun 2025
Accepted: 04 Aug 2025
Published online: 28 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article