A hierarchical online fault diagnosis method for sanitation vehicle hydraulic systems based on reliability-refined evidence fusion

To improve diagnostic accuracy and online output stability, this study proposes a hierarchical fault diagnosis method for sanitation-vehicle hydraulic systems based on pressure–time series. The hydraulic circuit of the bucket tipping cylinder in the QDT5120ZYSA6 sanitation vehicle was modeled in AMESim, and the normal condition together with three representative simulated fault scenarios, namely air mixed into the hydraulic oil, abnormal oil temperature, and internal pump leakage, were established. A hierarchical framework consisting of normal/fault discrimination and fault subtype identification was then developed, and offline modeling was performed by combining offline sliding-window segmentation, class-reliability refinement, and evidence fusion. The offline results show that the proposed hierarchical method outperforms the non-hierarchical four-class diagnosis method, as indicated by the confusion matrices, accuracy, and F1-score. In the Stage-2 fault subtype identification task, Reliability-DS achieved 100.00% accuracy and 100.00% macro-F1, outperforming RF, XGBoost, and CNN. Furthermore, online validation was carried out using pressure signals generated from AMESim/Simulink co-simulation, and a label-retention strategy was introduced to improve temporal consistency. The proposed method improves offline diagnostic performance and online output stability under the simulated operating conditions.