A Study Of Motor Monitoring And Intelligent Fault Diagnosis Based On Soft Starter

Motor is the main energy power equipment of industrial production, in order to ensure the safe operation of the motor, and to avoid the huge economic loss caused by the production disruptions, motor fault diagnosis technology has become an important part of the security maintenance. Most of the current motor fault diagnoses are off- line mechanical failures. The algorithm shortcomings are resource- intensive, slow speed and low accuracy. In recent years, the requirement of on- line electrical fault diagnosis in the process of motor starting, running and parking has become more and more urgent. This paper deeply studies the condition monitoring and fault diagnosis technology of asynchronous motor, and puts forwa rd a kind of motor on- line intelligent fault diagnosis algorithm used for embedded system based on soft starter. The paper has two innovations. One is the new algorithm of the neural network weights directly determined on harmonic detection in the process of motor condition monitoring. The other is the motor fault types determined by intelligent diagnosis algorithm presented in this paper more comprehensive and reliable than domestic existing algorithm, and realizing the innovation of soft starter driving motor system on intelligent fault diagnosis. The author’s major contributions are outlined as follows:1. The mechanical and electrical failures in the process of motor starting, running and parking have been studied. According to O n-site maintenance records and the phone records in nearly two years, the paper summarizes nearly ten kinds of fault types of the induction motor such as over voltage, overload, short of equal and so on. The generating mechanisms of failures are analyzed in detail and the result shows that the motor stator voltage, current, temperature and speed information contained effectively characterizes of these failures. At last, an information comprehensive, fully and no redundant motor fault diagnosis knowledge bases have set up.2. The motor condition monitoring has been studied. Firstly, according to the requirements of motor soft start system, a monitoring method in the operation of the motor stator voltage, current and rotating speed is analyzed. Using the two algorithms of Fast Fourier Transform(FFT) and weights of the neural network directly determined(WDD) for the voltage and current signal spectrum analysis, has extracted the voltageand current features in frequency domain. According to the principle of harmonic admittance, the motor speed has been further got. The algorithm performance comparison proves that WDD based on triangle basis functions more quickly and better performance. At last, the paper compares the motor speed with the inherent motor model in MATLAB.The simulatio n result is in the error range and shows that the rotational speed measurement algorithm can be applied to the motor without speed sensor and realizes the rotating speed monitoring. The fault characteristic values have had a detailed study. This is also an innovation.3. The application algorithms of several kinds of neural network in fault diagnosis have been studied and analyzed. The neural network algorithm in fault diagnosis, BP, RPROP and BAYES theorem, are discussed and analyze the application charac teristics and advantages and disadvantages of three kinds of algorithm. The study found that the BAYES neural network has powerful nonlinear approximation, simple algorithm, faster convergence rate, especially suitable for the diagnosis of soft starter sys tem which requires faster speed and less resource- intensive. Make a detail of the BAYES in aspects of recognition principle and the network structure and improve the BAYES neural network.4. The motor intelligent fault diagnosis algorithm has been studied and designed. Based on the soft starter embedded system, the development process and fault diagnosis model of BAYES neural network(PNN) are proposed. In order to achieve a high speed and accuracy, the fast algorithm is optimized from network model select ion, structure design, learning algorithm design and other aspects. Finally, the precision and real-time of the algorithm are verified though simulation programming, and compared with other commonly used algorithms. The result shows that this algorithm has a higher real-time and accuracy, and achieves its intended goal.