Fault diagnose based on pattern recognition

• Main Author: Luo, Wen Jie
• Other Authors: Luo Ming
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/63868
• Institution: Nanyang Technological University

The purpose of the final year project is equipping the students the ability to solve the real life problem individually, enabling them to apply what they have learned in university. Students also need to collaborate with professors and other students, through which they have learned the art of effective communication. The purpose of this project is to develop a fault diagnose algorithm to classify the type of fault for electrical equipment. First, the indicator for the electrical equipment, Partial Discharge (PD) is introduced. Partial discharge as an indicator for the health level of the equipment is affected by a series of the variables, the relationship between PD and these variables is discussed in the literature review. Two methods for extracting PD in introduced in this report as well. The modern method, non-intrusive method is emphasis in this project, the raw data collected for the project is come from this method. The artificial neural network as an approach designed by human to simulate the function of the human brains has been growing fast in recent years. The machine learning is a powerful tool in solving the problems that cannot be expressed in steps by steps [1], such as pattern recognition, classification, series prediction, and data mining. The most common neural network model is feedforward backpropagation network which is the one implemented in this project. The more details of this model will be discussed in this report. In this project, I have successfully developed a feedforward backpropagation network that is based on the visual studio for the fault classification. The historical data is used to train the network, new testing data is presented to the network for prediction. The benchmark data and testing data have shown a high accuracy of the prediction, which is up to 90%. The real data give a slightly low prediction, which is around 70%. Considering that more improvement can be done in the future, as better algorithm can be implemented, we can foresee a higher accuracy in the future. In conclusion, artificial neural network is a good model in the equipment diagnosis.