The Research On Methods Of Condition Monitoring And Fault Diagnosis Based On SVD

With the increase of system complexity, usually it is very difficult to forecast and diagnose its fault only by manipulator's experience. So system monitoring and diagnostic techniques become indispensable to provide more efficient system maintenance, avoid catastrophic failures, and improve the efficiency of production. The fault diagnosis of machinery is a kind of technology, which utilizes the technology of signal analysis and processing to the sampled signals with fault information. From the data with noise, it can find out the characteristic parameter related with fault and then use them to distinguish real-time status of the equipment.Periodic signal exists greatly in modern mechanical equipments and usually can include lots of fault information, so it's of great research value. It should be pointed out that some systems' characteristic signal is periodic. While they are faulty, new signal compositions will come up. At the same time, there are some faults accompanying with some periodic compositions. In this research, I will propose new methods based on singular value decomposition (SVD) to solve these two classes of problems. These methods may be more effective than others while SNR (signal noise ratio) is lower.This paper analyzes the characters of strange attractors and presents a new method based on SVD to monitor the change of periodic signal's energy. To improve the efficiency of this method, the definition of singular entropy is introduced , analyzed and improved. A new concept, PCTE (percent of contribution to total energy) of signals is proposed and its advantage to singular entropy is proved.SVR (singular value ratio) spectrum is one of the most effective methods to detect and attract periodic compositions of signals. This paper deeply studies its advantages and shortcoming. A new method using SVD is proposed to detect and attract periodic compositions of signals. Numerical simulation examples for both sine and non-sine signals with noisy backgrounds demonstrate its validation. The accuracy of period and the recovered signal given by our method are quite satisfied. And it fully employs the numerical stability associated with SVD.