Industrial environments frequently encounter complex time-series data such as machine vibration patterns, motor thermal imaging, and sensor pressure metrics. Equipment failure prediction grapples with the temporal nature of the data and the challenge posed by minority failure instances. In this paper, we introduce a refined generative mechanism, building on the foundation of the Wasserstein generative adversarial network (WGAN) and the borderline synthetic minority oversampling technique (Borderline-SMOTE). By utilizing time-series features, the proposed method effectively addresses the intricacies of predictive modeling. To demonstrate its efficacy, we used a complex and multisensor hydraulic system dataset for validation. Experimental results indicate that the proposed method outperforms existing strategies, enhancing the F1 score by at least 2.21% and achieving a recall rate of 95.51%. This suggests a promising direction for enhancing fault prediction in complex industrial settings.