Study On Fault Diagnosis Of Asynchronous Motor Based On Non-Invasive

Motor which is widely used in all areas of industrial production plays a very important role in the modern industry. The performance of motor will be severely affected and its efficiency will drop, because of the role and influence of various factors such as mechanical, thermal, electricity and the surrounding environment during operation. The finally result is not only a single motor or electrical equipment does not work as usual, but also affect the normal operation of the whole productive process, and lead to huge losses of companies or research departments. So the early fault diagnosis of the motor is particularly important. We can find the fault as soon as possible by diagnosis and analysis for the common fault of motor, and propose some feasible detection methods for motor fault, thereby reduce unnecessary loss.This paper proposes a fault diagnosis method based on non-invasive rotor broken bars after compare the traditional fault diagnosis methods with the new methods which are based on modern analysis. The method includes four steps that stator current signal acquisition, the pretreatment of fault signal, fault feature extraction of rotor broken bars and failure recognition. We can output the signal through the Hall Sensor first, and then use the conditioning circuits to process it. We can convert analog signals into digital signals by A/D conversion circuit. The digital signals are transferred into PC through serial port after DSP. In the second step, we conduct the Park vector modulus processing stator current signal, and make the component of fault feature side frequency far away from the component of fundamental frequency, thus climinating the bad influence of the component of fundamental frequency. Finally, we convert it into DC. In this paper, we use the five layers wavelet packet decomposition to extract the fault feature and recognize the fault of rotor broken bars through the widely used method BPNN. The simulation result shows that after the fault feature extraction used Park vector modulus and wavelet packet decomposition, the correct rate of recognition used BPNN is over 90%.