(For USM Staff/Student Only)

EngLib USM > Ω School of Electrical & Electronic Engineering >

Fault detection and diagnosis of induction motors using artificial neural network models/Ng Yeong Chuan

Fault detection and diagnosis of induction motors using artificial neural network models_Ng Yeong Chuan_E3_2011_NI
Pengesanan kerosakan dan diagnosis berdasarkan data sampel adalah berkesan dan boleh dipercayai. Data sampel yang dikumpul daripada sesuatu mesin atau peranti seperti torsi, arus dan kelajuan mengandungi maklumat yang penting. Ciri-ciri yang berfaedah boleh dihasilkan daripada maklumat ini dengan menggunakan kaedah yang sesuai. Dalam projek ini, arus tiga fasa iaitu fasa A, fasa B dan fasa C daripada motor induksi dikumpul. Ketumpatan spektra kuasa menukarkan isyarat arus individu kepada spektrum frekuensi. Pasangan pilihan harmonik yang berbeza berdasarkan spektrum frekuensi digunakan sebagai input kepada model rangkaian saraf tiruan untuk tujuan pengesanan kerosakan dan diagnosis. Dua model jaringan saraf tiruan digunakan dalam projek ini iaitu perceptron pelbagai lapisan (MLP) dan min-max kabur (FMM). MLP terdiri daripada lapisan input neuron, satu atau lebih neuron yang tersembunyi dan lapisan output neuron. FMM menggunakan set kabur sebagai kelas. Setiap set kabur dibentuk daripada penyatuan hyperboxes. Empat jenis kerosakan motor induksi disiasat. Kerosakan itu ialah kerosakan bar rotor, masalah esentrisiti, kerosakan belitan stator dan voltan yang tidak seimbang. Prestasi MLP dan FMM dibanding dari segi ketepatan dan masa yang diperlukan. Apabila data sampel daripada semua jenis kerosakan motor induksi 1hp digabungkan, ketepatan untuk FMM ialah 98.90% dan masa yang diperlukan ialah 0.54s. Ketepatan untuk MLP ialah 89.94% dan masa yang diperuntukkan ialah 2.74s. Keputusan ini menunjukan bahawa FMM adalah lebih tepat, lebih stabil dan menggunakan masa yang lebih singkat berbanding dengan MLP dari segi pengesanan kerosakan dan diagnosis motor induksi. ___________________________________________________________________________________ Fault detection and diagnosis based on data samples is effective and reliable. Data samples collected from machines or devices such as the current, torque, and speed may contain important information. Some features can be extracted from this information by using suitable methods. In this project, three phase currents, i.e., phase A, phase B, and phase C currents from induction motors are collected. These data sets are used as the inputs to the Power Spectral Density (PSD). The PSD is used to convert the individual current signals into the frequency spectra. Then, different selection pairs of harmonic magnitudes based on the frequency spectra are used as inputs to the Artificial Neural Network (ANN) models for fault detection and diagnosis purposes. Two ANN models are used in this project, i.e., the Multi-Layer Perceptron (MLP) and Fuzzy Min-Max (FMM) networks. The MLP network consists of an input layer of source neurons, one or more hidden layer of computational neurons, and an output layer of computational neurons. On the other hand, FMM utilizes fuzzy sets as classes. Each fuzzy set is formed from the union of hyperboxes. Four types of fault conditions of induction motors are investigated. The faults conditions are broken rotor bars, eccentricity problem, stator windings fault, and unbalanced voltage. The performances of MLP and FMM are compared in term of test accuracy, and the computational time. When the data samples from all types of faults of 1hp induction motor are combined, the test accuracy for FMM is 98.90% and the computational time is 0.54s. For MLP, the corresponding accuracy and computational time are 89.94% and 2.74s respectively. The results show that FMM is more accurate, more stable, and consumes less computational time as compared with MLP in fault detection and diagnosis tasks.
Contributor(s):
Ng Yeong Chuan - Author
Primary Item Type:
Final Year Project
Language:
English
Subject Keywords:
Fault detection ; torque ; induction motors
First presented to the public:
4/1/2011
Original Publication Date:
2/26/2020
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Electrical & Electronic Engineering
Citation:
Extents:
Number of Pages - 69
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2020-02-26 16:46:36.741
Date Last Updated
2020-06-29 11:46:21.843
Submitter:
Nor Hayati Ismail

All Versions

Thumbnail Name Version Created Date
Fault detection and diagnosis of induction motors using artificial neural network models/Ng Yeong Chuan1 2020-02-26 16:46:36.741