| dc.description.abstract | The proposed Winner Take All Experts (WTAE) network is based on 'divide and conquer'. It employs growing fuzzy clustering methods to di vide a complicated problem into a series of simple sub-problems and assign an expert to each of them. It also allocates every new case to one of the experts, and, if the output is incorrect, the weight adaptation is localized to the local expert. As a result, it is a fast learning algorithm without knowing a priori information. After the sensor approximation, the outputs from the estimator and the rea] sensor readings are compared both in time domain and in frequency domain. Three fault indicators are used to detect the sensor failure. In the decision stage, the intersection of three fuzzy sets accomplishes a decision level fusion, which indicates the confidence level of the sensor health. Two data sets, the Spectra Quest (SQ) Machinery Fault Simulator data set and the Westland vibration data set were used in simulation experiments to demonstrate the perfonnance of the WTAE network. Comparisons of tracking performance among the proposed network and MLP, RBF network were performed. The WTAE was found competiti ve with or even superior to the others. Comparisons for decision making processes between WTAE network and traditional time domain indicators were also performed. The WTAE achieved 100% correct detection for both testing data set without knowing a priori information. Using the same testing sets, the traditional indicators only detected Jess than 87.5% failure states depending on knowledge of characteristics of both the sensors and the environments. | |
