Research On Experiment Table For Rotating Machinery Fault Diagnosis

Research on rotating machinery fault diagnosis is of great significance for avoiding catastrophic accidents and huge economic losses. In order to put scientific research achievements into experimental teaching, a small experiment system for rotating machinery fault diagnosis was designed in this paper, which aimed to make students understand the general faults and their harms during rotating machinery vibration as well as corresponding diagnosis methods.In this paper, the experiment system was composed of rotating motor, eddy current sensor, photoelectric sensor, signal conditioning module, data acquisition card and operation software on PC and it mainly simulated rotor imbalance fault, measured and collected vibration data and rotation velocity data of the rotor. Fault features were extracted from vibration data by time-domain analysis and energy analysis in frequency domain using wavelet package. Then the fault features were evaluated in order to select proper features for building fault diagnosis model. The fault diagnosis model was realized respectively with back propagation neutral network (BP), self-organizing competitive neutral network (SOC) and supporting vector machine (SVM). The influence of feature number and training sample number to the fault diagnosis performance of BP was analyzed. The structure and fault diagnosis performance were compared between BP, SOC and SVM. The comparison result showed that it was easier to determine the structures of SOC and SVM and their fault diagnosis performances were better than BP. Furthermore, a new method for fault feature extraction was presented in order to reduce the discriminability degradation of fault features when the rotor was in a very tiny imbalance status, which combined wavelet package decomposition and time-domain fault features. Finally, operation software on PC was developed by LabVIEW and four experiments were designed according to the requirements of experimental teaching.