SDR: Sensor data recovery for system condition monitoring

Sensors are widely utilized in many industry scenarios, e.g., aircraft and satellite condition monitoring. However, sensor data may become anomalous due to sensor fault, malfunction of connectors, etc. How to avoid the wrong condition monitoring result caused by the anomalous sensor data is challenge. To deal with this problem, one kind of sensor data recovery algorithm is proposed in this article. Firstly, the correlations among sensors data are analyzed by mutual information. The available sensors data for recovering the anomalous sensor data are determined. Then, the recovered sensor data are achieved by Least Square — Support Vector Machine (LS-SVM). The effectiveness of the proposed algorithm is evaluated by the sensor data set which is adopted as the Prognostics and Health Management 2008 Conference challenge data. Compared with other selected sensors data, the recovered sensor data with the proposed algorithm can reach smaller values of Relative Error and Root Mean Squared Error.