基于 DCCA⁃DAE 模型的传感器故障检测

As a key component of complex equipment monitoring system,sensor fault can cause false alarms, significantly compromising the reliability of complex mechanical system condition monitoring. To address this problem, a sensor fault detection method (DCCA-DAE) based on detrended cross-correlation analysis (DCCA)and a dual auto encoder(DAE)is proposed from the system perspective. First,a coupling network is constructed using the DCCA method to extend the monitoring data from Euclidean space to topological space, thereby enabling a comprehensive representation of the information embedded in multi-source,multi-state system monitoring data. Then,an anomaly detection model is developed based on DAE to eliminate the influence of changes in operating conditions on the sensor monitoring sequence,achieving the accurate detection of sensor fault in systems with complex changes in conditions. Finally,the comprehensive performance of the proposed method is verified using historical data of a turbine unit from a power plant. The results show that the DCCA-DAE method has a strong feature extraction capability and the detection accuracy is significantly better than that of models,such as the traditional support vector data description and auto encoder. DCCA-DAE method shows good application prospects in sensor fault detection in industrial scenarios.