Date Available

3-25-2023

Year of Publication

2022

Document Type

Master's Thesis

Degree Name

Master of Electrical Engineering (MEE)

College

Engineering

Department/School/Program

Electrical and Computer Engineering

Advisor

Peng Wang

Abstract

Industrial motors are widely used in various fields such as power generation, mining, and manufacturing. Motor faults and time-consuming maintenance process will lead to serious economic losses in this context. To monitor motor faults and detect motor conditions, different types of sensors that can test vibration and current signals are mounted on motors. However, the main challenge was how to use information gained by sensors to analyze or diagnose motor conditions.

Machine learning is a popular technology in recent years, and it's very suitable for crunching and analyzing data. As an important subset of machine learning, deep learning is suitable for classifying objects through a large amount of data. However, since the development of deep learning for a long time, scientists can only get results through data and cannot know the process of model training. Shapley value was developed to solve this problem. To put it simply, the calculation of the Shapley value can know the contribution of each feature, which enables us to have a good understanding of the data and model.

This thesis focuses on using an artificial neural network that is the backbone of deep learning algorithm to analyze the data obtained by a set of vibration sensors of an industrial motor. Another focus of this thesis is around feature extraction, which refers to how vibration signal can be processed into useful machine learning dataset. This process would significantly improve the efficiency of machine learning model. Shapley value is introduced into the post-training model to explain the model. Meanwhile, other machine learning but non-deep learning models such as support vector machine and random forest are used to analyze the same data and obtain corresponding results. These machine learning models will act as a control group for the neural network.

Digital Object Identifier (DOI)

https://doi.org/10.13023/etd.2022.359

Share

COinS