The Research On Fault Diagnosis Of Rotating Rectifier Based On Wavelet And Artificial Neural Network

The exciter system of traditional DC exciter has some weaknesses, whose commutators and carbon brushes often need examining, repairing and maintaining timely. But with the development of electric power industry, the capability of each kind of electric machine increases quickly. Accordingly, the voltage and the power of excitation increase. If the excitation current is too great, adopting DC machine excitation would require more collecting rings. That will bring some inconvenience, such as installation and examination and maintenance of electric machine. So conventional DC excitation system is difficult to satisfy the developmental need of electric power industry.As a result, a new mode of brushless exciter appears. At the same time, brushless exciter system has canceled slip rings and carbon brushes, and improves its operational stability. Because of canceling slip rings and carbon brushes, it is impossible to measure directly the rotating rectifier's current and voltage. So, it is very difficult to monitor and protect the rotating rectifier. This paper theoretically analyses the armature current and magnetomotive force in AC exciter of brushless synchronous machine with rotating rectifier. Under the normal or fault conditions of rotating rectifier, the mathematical expressions of the armature current, the m.m.f and induced electromotive force are approached. According to computer calculating, we can obtain the harmonics amplitude of the stator coil of excitation machine. And we can diagnose the condition of the rotating rectifier with harmonics amplitude.Wavelet theory has the special ability to express the signal local characters inthe time and frequency region. In this paper, the signals of rotating rectifier normal or fault conditions have been analyzed by wavelet transform, and the eigenvector, which includes useful information, has been obtained. Firstly, the voltage sample signals from AC exciter stator coil were de-noised and filtered by means of decomposition of the wavelet function. Secondly, the character frequency region of the rotating rectifier fault was selected clearly with wavelet package decomposition and reconstruction. So the eigenvector, which includes the rotating rectifier fault condition, can be gotten.The artificial neural network deals with nonlinear problems with powerful capacity. In brushless excitation system, rotating rectifier is likely to occur several faults. But there is not clear linearity relation between the cause of fault and corresponding symptom. And it is difficult to describe their relation with mathematics model. Because wavelet transform has forceful ability to pick-up character and artificial neural network has a strong capability to classify information. In this paper, the wavelet network has been formed by the wavelet transform, which is multi-dimension wavelet, and the artificial neural network which is back-propagation network. Taking the eigenvector as an input of the wavelet network, the wavelet network can fulfill diagnosis of faults. As compared with the simple BP network, this diagnosis result is avoiding getting in local-minimum, more accurately and reliably. It also avoids the omitting and mistakes diagnosis of the rotating rectifier faults.Finally, taking the 630kW brushless synchronous motor as the simulation, this paper presents a simulation method which uses the wavelet network. The process of this simulation proves it is practicable.