Fault sample augmentation method: Integrating adaptive signal reconstruction and multi-scale stochastic quantization

To address the challenges of insufficient abnormal samples, data imbalance, and poor generalization in industrial fault diagnosis, this study proposes a universal fault sample augmentation framework integrating adaptive signal reconstruction and multi-scale stochastic quantization. The framework comprises five core modules: adaptive signal decomposition, optimized cost function construction, sliding window segmentation, multi-scale stochastic quantization, and clustering threshold constraints. Multi-objective optimized variational mode decomposition (VMD) via Limited-memory Broyden–Fletcher–Goldfarb–Shanno with Bound constraints is employed to extract stable and physically interpretable components, reducing manual parameter dependence. A multi-scale sliding window mechanism combined with Gaussian-based stochastic quantization enhances sample diversity while preserving intrinsic signal characteristics, and a clustering-based constraint is introduced to filter pseudo-anomalies. Experimental results on multiple benchmark datasets demonstrate that the proposed method significantly improves diagnostic performance and achieves robust generalization under imbalanced conditions.