Yunlin Ma

To address the critical challenge of low diagnostic accuracy in multistate bearing fault diagnosis caused by inefficient discriminative feature extraction under varying operating conditions, this paper proposes a novel Parameter-weighted Viridis Image Encoding (PVIE) method. Unlike conventional image encoding techniques (e.g., GADF, GASF, MTF, RP) that often suffer from high computational complexity and limited feature separability in complex scenarios, PVIE integrates Variational Mode Decomposition (VMD) with a newly designed Parameter-weighted Euler Difference Feature Extraction (PWEDFE) module. This module explicitly enhances the nonlinearity and periodicity of fault signatures, mapping them into lightweight 2D feature images via Viridis Feature Value Mapping (VFVM). Extensive experiments on two benchmark datasets demonstrate that PVIE achieves exceptional diagnostic accuracies of 99.92 % and 99.74 %, respectively. Compared to state-of-the-art encoding methods, PVIE improves average accuracy by 21.06 % to 39.78 % while reducing diagnostic time by 53.3 %, significantly outperforming existing approaches in both accuracy and efficiency. Furthermore, the method exhibits robust performance under strong noise interference and small-sample scenarios. These results confirm that PVIE offers a substantial advancement over current research by providing a more discriminative, lightweight, and robust solution for real-time industrial fault diagnosis.