Sensors Fault Diagnosis Method Based On Information Entropy

With the rapid development of automation in industrial field,the structure of control system becomes increasingly complex. Therefore, there is an necessary requirement of sensor performance. The method of sensor fault diagnosed which based on the signal characteristics has become one of the key technical of fault diagnosis. This paper aims to solve the problems among signal feature extraction and classification of control system. The method of information entropy and cluster analysis will be used to research in feature extraction and classification of sensor fault signal.Initially, according to the principle of sensor fault establish a sensor model which based on SISO(single input single output) system, the mechanism of sensor fault sets up six kinds of sensor fault models. Used the analysis and reconstructed theory of wavelet transform and the wavelet threshold noise reduction principle to preparatory process the signal.To analysis low-frequency signal fault using generally wavelet analysis, in order to analysis the high-frequency signal fault using its multi-resolution analysis feature.Secondly,considered the advantages of wavelet analysis on non-stationary signal analysis and in order to quantitative analysis the uncertainty level of signal, the author introduce the concept of information entropy and describe numous kinds of entropy in different usages. Defined the concepts of wavelet information entropy and wavelet multi-scaled entropy, research results wavelet entropy can characterize non-stationary signals fault feature and the entropy combined with the wavelet muliti-scaled theory can reflect the information of fault signal in different frequency. In view of the information entropy calculation difficulty, put forward the complexity measure concept to calculating probability distribution. The feature space divided from the angle of complexity and the angle of energy. Through analysis and compare, this method is more sensitive then other methods.Ultimately, in order to make the output fault types simple, put forward the method of clustering analysis, and though it get the clustering centers. Compared the distances of random fault with clustering centers, judged sensor fault type from the minimum distance. Confirmed by experimental data, the method can accurately and direct output the sensor fault type.