Please use this identifier to cite or link to this item:
Full metadata record
DC FieldValueLanguage
dc.contributor.authorMeghajibhai, Patel Rajeshkumar-
dc.guideGupta, S. P.-
dc.guideKumar, Vinod-
dc.description.abstractThe subject of condition monitoring of electric motors to diagnose the machinery faults has been of wide interest for several decades. In industry electric motors plays a key role, with demanding performance criteria in some applications. Though the monitoring, diagnosis and incipient fault detection techniques have been widely researched, many uncertainties still remain. To be truly effective, the process needs to be automated to reduce the dependence on manual data interpretation. Different methods for fault identification have been developed and used to detect the machine faults at an early stage using machine parameters such as current, voltage, speed, temperature, and vibrations. Electric motor applications range from the small motors used in intensive care life support units to large motors used in power plants. With proper machine monitoring and incipient fault detection schemes, early warning can be achieved for preventive maintenance resulting in improved safety and reliability of different engineering system operations. The primary goal of the present work is to develop a condition monitoring system that can be used for reliable detection of bearing faults in induction motors. Such a system should be portable to facilitate industrial testing. It is estimated that 40% of the motor faults are attributed to bearing failure. Detection of these defects is important for condition monitoring as well as quality inspection of bearings. It is the aim of this research to assess the bearing condition of machine by automated diagnostic module. The thesis focuses on the design, development, and application of an automatic diagnostic procedure for rolling element bearing faults in rotating electrical machinery. In this thesis, the fundamentals of fault signature production in rolling element bearings are briefly reviewed. The bearing faults are categorized on the basis of inherent fault location, i.e. inner race, outer race, ball defect and multiple locations. Besides, roughness of a bearing component too is also considered as faults. Further, for these faults, the degree of severity is also considered in four levels, level 1 to level 4. This categorization helps in establishing the connection between the faults and the machine parameters which is essential in feature extraction of the diagnostic system. Practical concept of generating faults via bearing current is also implemented. The bearing fault database for all these faulty types of test bearing has been created on varying load condition. Additional data of supply voltage, speed, and stator current is also recorded. The present study is heavily experimental and is aimed at establishing basic relationships between bearing failures and the resulting changes in vibration response in induction motor. The results of laboratory investigations, carried out on a 7.5 kW, 3-phase, cage induction motor are presented to study the mechanical faults related to ball bearing of the motor. First stage of the work deals with the development of a portable monitoring set-up, suitable for laboratory as well as industrial investigations. The set-up has the facility to acquire vibration signals of the motor and process them for analysis in a built-in computer. Provision has been made to acquire and process additional signals of voltage, current and speed. The transducers used in this set-up are as follows: • Piezo-electric accelerometer for vibration acceleration measurements • Hall Sensor type clamp-on current sensors for stator current monitoring • Non-contact type speed sensor for speed monitoring • Potential Transformers for voltage measurements All the signals collected from the test machine are stored, after amplification, in the built-in computer via NI make DAQ card inserted in PCMCIA port. The signals are sampled at high sampling frequencies, 10 kHz, so as to get the desired frequency resolution. MATLAB program script is written using DAQ Toolbox support to pick up the required information from the test machine, as fully controlled manner by the user. User can set sampling frequency, duration and gain of signal as per the specification of transducer before starting the data recording. In this work, three major fault categories of bearing faults are studied, namely, (i) single point defect, (ii) multiple point defects, and (iii) generalized defects due to bearing surface roughness and uneven dents. Total 32 fault combinations have been created and vibration signal is analyzed to understand, what type of signature appears in machine vibration and how to design better diagnostic model to detect these faults. Incipient level defects such as small dent, multiple dents, minor scratches, generalized ii roughness, are introduced in ten bearings with the help of domain experts. Additionally, medium (2mm) size single point and multiple point defects are created in nine bearings. These faults have been introduced by the manufactures namely, NBC (Jaipur, India), TATA (Kharagpur, India) and Premium Bearing (Surendranagar, India). Further to see severe and realistic industrial faults of bearings, thirteen bearings were picked up from the lot of damaged and discarded bearings by user industries. All 32 faulty cases of bearing have been tested under same operating conditions and load on same motorto assure that variation of vibration parameters are true replica of the fault signature In the second stage of the work, signals are recorded, using the portable monitoring set-up, on laboratory test motor with three healthy and 32 faulty bearings on different load. The dataset for all test bearings is created which consists of ten signal records for each load condition for each test bearing. This database is used to evaluate the bearing condition. Vibration data is collected using accelerometer, which is attached to the bearing housing with magnetic base. Accelerometer is placed at the 12 o'clock position at the bearing support housing of shaft. All data files recorded are saved in Matlab (*.mat) format for further use. In the third stage of the work, the detection of failure in rolling element bearing is investigated by vibration analysis. The recorded signals are analyzed in (a) time-, (b) frequency-, and (c) time-frequency domains for diagnostically important features. The aim of vibration signal analysis is to find a simple and effective analytical technique for extracting information contained in the signals and its characteristic features for fault diagnosis. The time domain parameters namely (i) RMS value, (ii) crest factor, (iii) variance, (iv) skewness, and (v) kurtosis are determined from the vibration signal data. The results and significance of the time domain based descriptors for bearing fault condition are presented. It is found that, three time domain parameter namely, (i) RMS Value, (ii) crest factor, and (iii) kurtosis give indication of existence of defects and its severity, while, remaining parameters are not giving keen measures and have been excluded. This data are used for training of the ANN as described later. in For frequency domain analysis, frequency components are extracted from the FFT spectrums of recorded vibration (acceleration) signal for identification of fault characteristic frequencies. Most fault detection algorithm for rolling element bearings are successfully reported to detect defects based on characteristic fault frequencies, corresponding to faults in balls, outer race, inner race and bearing cage. However, the faults which are distributed, such as roughness, do not correspond to characteristic frequencies. It is found that frequency domain analysis becomes uncertain to classify second and third categories of defects. In present work, identification of distributed faults due to bearing surface roughness and uneven dents has been done on the basis of nature of vibration harmonic frequency bands which do not have any relationship to characteristic frequencies that have hitherto been used for bearing fault identification. The time-frequency domain analysis, has the capability of representing signals in both time and frequency domains. Time-frequency analysis techniques based on Discrete Wavelet Packet Transform (DWPT) are conducted to extract defect related sub-band components of signals with better resolution. The features obtained from DWPT are used for training and testing of ANN. If the DWPT coefficients are used as inputs to the ANN, it results in rather large number of inputs posing difficulty for training and testing of ANN with accuracy. Therefore, an approach to compute band energies of DWPT coefficients has been used as inputs to the ANN instead. The concept of Entropy is widely known as a quantification of complexity. Entropy is also used in the context of pattern classification. It can be used to capture the formants or the peakiness of a distribution. The combination of wavelet transform and entropy is investigated in many areas. However, little work has been done to pursue the role of wavelet entropy in bearing condition monitoring, which has been taken up in present work. The work is extended with feature extraction method based on wavelet and energy entropy, i.e. wavelet packet energy entropy (WPEE). The discrete wavelet transform, db5, is exploited to decompose signal into five levels. The signal is decomposed into small sub-bands in order to search those sub-bands which are sensitive to present fault classification and changes. Based on experimental studies and work out, 32 sub-bands are obtained that contribute most to the classification. IV Further, WPEE based feature extraction approach is implemented, for extracting features from each sub-band. The derived 32 features are later fused with three time domain parameters to form a new feature vector of 35 features, which is used as the input to the ANN based classifier. In this work, a three-layer feedforward network architecture, which is most popular one and has been proven to be able to learn arbitrarily complicated continuous functions, is configured. The performance of the ANN with respect to different number of hidden nodes and the different network architectures that are used, have been evaluated. The change in number of input neurons leads to different configurations of ANNs. The starting numbers of hidden neurons are taken as ten and increased till desired results are obtained by testing the neural network. The above all, 35 input nodes and 2 output nodes are primary requirement of present intention to configure the required ANN architecture. Some of the ANN configurations that have been tested are 35-15-2, 35-20-2, 35-25-2, 35-30-2 and 35-35-2. The healthy and faulty patterns have been given to the ANN and it is trained to give desired target respectively. Tangent Sigmoid neurons have been employed in hidden layers of the ANN. The entire training process of the neural network is carried offline on full load data prior to testing online on different load condition. Backpropagation using Levenberg- Marquardt algorithm has been applied for learning process of the neural network. The performance of different neural networks configuration has been evaluated. In addition, the performance of the neural network technique also depends on the training set size. Large training set, which contains 8750 training pattern size and 8750 testing and validation pattern size, for three healthy bearing cases and 32 faulty bearing fault cases, have been employed to improve the diagnostic performance. The effectiveness of the diagnosis module is analyzed by testing the system using data obtained from the machine at various known fault condition. The results of investigations of periodic condition monitoring of two 600 kW induction motors of a steel rolling mill are also presented and analyzed.en_US
dc.typeDoctoral Thesisen_US
Appears in Collections:DOCTORAL THESES (Electrical Engg)

Files in This Item:
File Description SizeFormat 

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.