Fault Diagnosis Based on PCA and D-S Evidence Theory

A fault diagnostic method combined with principal components analysis and D-S evidence theory is presented. Firstly a set of principal components models (PCM) are established by the type faults sample data, and high dimensions data could be characterized by the low dimensions data as new sample data under the condition of information lose least, and then the new data are to train the radial basis function neural network as recognition network to construct basic probability assignment function; Measurement data are analyzed with the PCM to get the low dimension characteristic vector, that are identified by recognition network; Finally the recognition result is fused by combining rule of D-S evidence theory as decisionmaking. The simulation study on boiler feed water control system shows that the fault of sensor can be isolated correctly and effectively by this method.