植基於動態結構類神經網路之馬達旋轉故障診斷系統研製

Design and Implementation of Diagnosis System for Motor Rotary Faults Based on Dynamic Structure Neural Network

蔡明達(Ming-Da Tsai)

指導教授 ： 曾傳蘆

國立臺北科技大學/機電學院/機電整合研究所/碩士(2008年)

摘要

本論文主要是研製一個智慧型馬達旋轉故障診斷系統。此系統的設計是基於常見的馬達旋轉故障，且應用動態結構類神經網路進行鑑別。
系統主體架構可分為：振動訊號量測、濾波處理、故障鑑別。訊號量測部份是將感測設備裝置於待測馬達端，再經由接收器讀取其振動訊號以便後續處理。由於量測的訊號易受外在機械結構的干擾，使得實際量測訊號常附帶有雜訊成份，因此本論文採用小波濾波器建立雜訊消除機制，以降低馬達故障診斷的誤判率。
故障鑑別機制則是採用改良式動態結構類神經網路。以往類神經網路中隱藏層神經元數的選取並沒有標準的理論根據，所以容易造成神經元學習時無法達到最佳網路架構與適時的收斂。因此可調神經元數的網路架構，將可動態調整出最佳鑑別效果的類神經網路。由於馬達故障時，其故障特徵會在特定倍頻中呈現，而且不同故障所擁有的故障特徵也不盡相同。因此本論文利用此種故障特性，擷取特定倍頻做為類神經網路的輸入訊號，而輸出訊號則是馬達故障種類的鑑別結果。
最後，本論文使用MATLAB軟體撰寫整體故障診斷系統，整合訊號量測、濾波、故障特徵擷取以及診斷系統於人機介面中。由實驗結果可知所研製的改良式動態結構類神經網路架構具有良好的鑑別精確度。

關鍵字

馬達旋轉故障 ； 動態結構類神經網路小波 ； 故障特徵

並列摘要

This thesis is mainly devoted to developing an intelligent diagnosis system for motor rotary faults. The design of this system is for the common motor rotary faults, and adopts the dynamic structure neural network to establish the diagnosis functionality.
The main structure of system can be divided into three modules: measurement of vibration signal, filtering process and fault classification. The vibration signal measurement is done by mounting the sensing module on the motor cast, and utilizing the receiver to read vibration signal for follow-up processing. Because the measured signal is apt to be influenced by the mechanical structure or other environmental factors, the signal often contains noises. To solve the problem, this thesis adopts the wavelet mechanism to filter the noises out, which may reduce the error rate of fault diagnosis.
The fault classification method uses the improved dynamic structure neural network. For the conventional neural networks, there is lack of methods to determine the number of hidden neurons. It leads that the network structure is not optimal and the convergence problem arises in the learning process. Using the dynamic structure neural network, the optimal neural network could be obtained by adjusting the number of neurons.
For the motor rotary faults, it is known that the characteristics of motor faults appear in specific harmonic frequencies and different faults cause different frequency patterns. This thesis utilizes the fault characteristics and extracts the special frequency patterns as the input of the neural network. The output of the neural network is the classification result of the corresponding fault.
To implement the intelligent fault diagnose system, this thesis uses MATLAB software. The system includes the signal measurement, filtering, fault characteristics extraction, and diagnosis function. All the functions can be executed by using the friendly graphical user interface. From the experimental results, it is found that the classification results outperform than the previous results.

並列關鍵字

Motor Rotary Faults ； Dynamic Structure Neural Network(DSNN) ； Wavelet ； Fault Characteristic

參考文獻

[38] 趙安明，馬達故障診斷之模糊類神經網路，碩士，中原大學機械工程學系，中壢市，2004。

[2] 王志豪，智慧型馬達旋轉故障診斷系統設計，碩士論文，國立臺北科技大學電機工程系，台北，2007。

[3] G. Betta, C. Liguori, A. Paolillo and A. Pietrosanto, “A DSP-based FFT-analyzer for the fault diagnosis of rotating machine based on vibration analysis,” IEEE Transactions on Instrumentation and Measurement, vol. 51, Dec. 2002, pp. 1316-1322.

[4] S. Wan, Y. Li, H. Li and G. Tang, “The new diagnosis method of rotor winding inter-turn short circuit fault and imbalance fault based on stator and rotor vibration characteristics,” Proceedings of the Eighth International Conference on Electrical Machines and Systems, vol. 3, 27-29 Sept. 2005, pp. 2207-2210.

[5] C. Y. Chang and K. K. Shyu, “Active noise cancellation with a fuzzy adaptive filtered-X algorithm,” IEE Proceedings of Circuits, Devices and Systems, vol. 150, 6 Oct. 2003, pp. 416-422.