Design and Implementation of the Bat-Algorithm Based Terminal Attractor Learning Method and Its Application to Motor Fault Diagnosis

• Main Authors: Po-Hung Wang, 王博鴻
• Other Authors: Chwan-Lu Tseng
• Format: Others
• Online Access: http://ndltd.ncl.edu.tw/handle/t2tb47

碩士 === 國立臺北科技大學 === 電機工程系所 === 104 === In order to improve the convergence efficiency of terminal attractors (TA) based neural networks, this thesis integrates the TA method with the bat algorithm and the developed learning method is applied to the motor vibration fault detection. In general, the convergence characteristic of terminal attractor different from the general convergence is that the former violates the Lipschitz condition. Consequently, the system trajectory converges to equilibrium point in finite time and thus the error converges to zero in finite time. On the other hand, the bat algorithm used in this work is to emulate the bat hunt mode. The optimization parameters are set as the positions of bats and the positions to the food source is optimized by the dynamic adjustment mechanism. Hence, this thesis finds the best learning rate by using the bat algorithm in order to reduce the number of terminal attractor learning iterations. This developed method is then applied to train the dynamic structural neural network used in motor fault diagnosis system. Also, the MATLAB R2010b software is used to establish the bat algorithm based terminal attractor learning method. By conducting numerical simulations of training neural networks, the efficiency of the proposed algorithm and other methods are compared and analyzed. Finally, this work uses the method to fulfill the module of motor fault diagnosis. From the results of motor fault simulation and analysis, the proposed method increase the learning speed and reduce the learning error of fault diagnosis.