As the cornerstone of modern industrial production, the reliable operation of complex mechanical equipment is crucial for ensuring operational stability, enhancing production efficiency, and preventing safety incidents. However, due to the intricate structures, multi-source coupled fault modes, and prominent nonlinear dynamic evolution characteristics, traditional diagnostic models often fail to simultaneously address high-dimensional feature processing and complex nonlinear relationships, which limits their diagnostic accuracy. To address the high-accuracy diagnostic requirements in industrial applications, this paper proposes a fault-diagnosis method for complex machinery based on Stacking ensemble learning. The method integrates Random Forest (RF), XGBoost, Support Vector Machine (SVM), and Backpropagation (BP) neural networks as heterogeneous base learners, with a BP neural network employed as the meta-learner to construct a layered fusion framework that effectively combines the strengths of each model. In addition, feature selection is performed using time-weighted maximum mutual information coefficients in combination with recursive feature-elimination cross-validation to identify the optimal feature subset. Furthermore, a constraint-based Bayesian optimization approach, tailored for industrial deployment, is used for automatic hyperparameter tuning of the base learners. The proposed method is applied to a lithium battery electrode double-roll pressing machine, where comprehensive data completion and validation experiments are conducted on real-world data containing 11 fault categories. The results show that the proposed method achieves fault classification and severity assessment with a diagnostic accuracy of 96.67%, significantly outperforming individual models and conventional fusion strategies. This study provides an effective technical pathway for the precise and efficient intelligent fault diagnosis of complex mechanical equipment.
Xiao et al. (Fri,) studied this question.