Combination of analytical and statistical models for dynamic systems fault diagnosis

Abstract

Complex industrial and aerospatial systems require efficient monitoring and fault detection schemes to ease prognosis and health monitoring tasks. In this work we rely upon the model-based approach to perform robust fault detection and isolation using analytical and statistical models. We have combined Principal Component Analysis (PCA) together with Possible Conflicts (PCs), to improve the overall diagnosis process for complex system. Our proposal uses residuals computed using PCs as the input for the PCA tool. The PCA tool is able to accurately determine significant deviations in the residuals, that will be identified as faults. The integration of both techniques provides more robust results for fault detection, while avoiding false alarms in PCAs due to changes in operation modes. Moreover, it provides the PCA approach with the necessary mechanisms to perform fault isolation. This approach has been tested on a laboratory plant with real data, obtaining promising results.