An Adaptive Passive Fault Tolerant Control System for a Steam Turbine Using a PCA Based Inverse Neural Network Control Strategy

Fault tolerant control (FTC) becomes an effective way to defectively control a plant and ensure reliability and safety in the system. This paper presents a new adaptive passive fault tolerant control (FTC) methodology based on inverse control strategy. An adaptive principal component analysis (PCA) algorithm is incorporated as a pretreatment data processing to recursively capture inherent time-varying information embedded in the plant time-series measurements. A multi-layered perceptron (MLP) neural network is then trained online with the reduced PCA extracted features to emulate an adaptive inverse controller based on actual post-fault plant dynamic model. The adaptive MLP-based controller will be able to minimize induced tracking error using an error back propagation (BP) learning algorithm without a priori knowledge of the occurred faults on the basis of the PCA-uncorrelated measurement data. This enhances the generalization capability of the realized controller due to distinctiveness of the PCA-based data representation. An extensive set of test scenarios has been considered to explore effectiveness of the proposed FTC scheme against three major faults in an industrial steam turbine benchmark. The results demonstrate promising capability of the proposed FTC to automatically maintain the steam turbine availability with efficient fault accommodation.