Application of the CatBoost Model for Stirred Reactor State Monitoring Based on Vibration Signals

Abstract Stirred reactor is a key equipment in the production process, and will result in large economic losses and safety issues when unpredictable failures occur. Therefore, it is necessary to monitor their health state. With this goal, firstly, this study presets five states of the stirred reactor: normal, shaft bending, blade eccentricity, bearing wear, and bolt looseness. x, y, z axes vibration signals are collected and analyzed in time and frequency domain. Secondly, 93 statistical features are extracted evaluated by Relieff, MIC and XGBoost. The above evaluation results are then fused by D-S evidence theory to obtain the final 16 features that are most relevant to the state of the stirred reactor. Finally, CatBoost algorithm is introduced to establish the health state monitoring model of the stirred reactor.The validation results show that accuracy of the proposed model is 100% for state recognition and 98% for fault diagnosis.