Undergraduate Thesis

The original thesis is in Chinese, and its English abstract is below. You can also download the PDF here.

Research on Motor Fault Diagnosis Method Based on the Fusion of Vibration and Current Signals

Author: Fan Chen

Thesis Advisor: Gang Niu

Abstract

Motors play an important role in modern society and are widely used in transportation, mining, chemical industry and other fields. Motor fault monitoring and diagnosis can ensure safe, reliable and efficient operation of motors, while reducing the cost of operation and maintenance. The single-signal-based motor fault diagnosis method is deficient in terms of accuracy. The amount of information in a single signal is relatively limited, and it is likely to produce misclassification as a result. At the same time, although the single signal has better diagnosis for some fault types, it is difficult to make more accurate diagnosis for all fault types. Therefore, the single signal cannot meet the needs of people, so the method of motor fault diagnosis using data fusion technology has been born. The data fusion technology can avoid the misclassification due to insufficient information to a certain extent. This paper focuses on induction motor fault characteristic frequency and the use of decision-level data fusion for motor fault diagnosis. It also completes the corresponding experiments. The main content of this paper is as follows.

Firstly, the research background of the motor fault diagnosis method based on the fusion of vibration and current signals and its significance are introduced. The current status of domestic and international research is summarized, including the motor fault diagnosis based on vibration signals, motor fault diagnosis based on current signals and motor fault diagnosis based on current and vibration signals.

Then, the induction motor fault diagnosis system based on decision-layer fusion is introduced, including the methods and principle involved in each system's aspect. Three types of multi-sensor information fusion with different structures, six commonly used classifiers, and five decision-level data fusion methods are listed. The classifier selection method based on accuracy and correlation coefficients is introduced to achieve better fault diagnosis results.

Afterwards, the induction motor model is established and the corresponding motor states are set, including healthy states and six fault states. A set of stator three-phase current signals and three-dimensional electromagnetic vibration acceleration spectra in different states are obtained through finite element analysis. The current data is used for feature extraction and analysis of different faults. The feasibility of fault diagnosis of induction motor by collecting stator three-phase current signals is verified.

Finally, a set of stator three-phase current signals and three-dimensional vibration signals is collected by experimental equipment. The current signals and vibration signals of this data are processed and feature extracted separately. Then, use a variety of classifiers to classify test samples and select the classifiers. The fault diagnosis of induction motor is realized by five decision-level data fusion methods. The results show that the multi-agent method is significantly better than the other four methods.

Fig. 3. ANSYS Maxwell and Workbench simulation flow chart.

Fig. 4. Majority voting accuracy using multi-source data versus single-source data.
Compared to the method based on current data, multi-source data fusion enhances the motor fault diagnosis accuracy by 20 percentage points.

Fig. 5. Fault diagnosis accuracy of five decision-level fusion methods.

Key words: induction motor, fault diagnosis, multiple-sensors data fusion, feature extraction, finite element method

Page updated

Google Sites

Report abuse