Please use this identifier to cite or link to this item:
|ANALYSIS FOR FAULT IDENTIFICATION IN ROTATING ELECTRICAL MACHINES
|(NEE AHUJA), Sulochana Wadhwani
|IDENTIFICATION IN ROTATING;ANALYSIS FOR FAULT;ELECTRICAL MACHINES;MECHANICAL FAULT
|This work presents laboratory investigation carried out on a set of induction motors to study the mechanical faults, mainly related to the ball bearing of the motor. The vibration signals are recorded for different bearing faults namely, ball defect, inner and outer race defect. The time records of the vibration signal are then analyzed in time, frequency and time-frequency domains and diagnostically important features are extracted. The extracted features are then used for development of the knowledge-based system for fault diagnosis. The fault diagnosis algorithm developed in this work employs information primarily on the vibration records. The operating speed is also recorded and used in the algorithm. Additionally, signals of supply voltage are also recorded. Though not used in the fault algorithm, they provide an insight into the state of the supply mains. The record of the current drawn from the supply is also obtained to examine if the rotating machine offers asymmetry in the three phases. A record of the motor body temperature is also obtained and checked against the permissible limits. The vibration signals are recorded by an accelerometer mounted on the upper surface of the test-bearing housing, as it is found to be the best position where the largest vibration amplitudes are obtained. The supplyvoltages and currents are recorded continuously to check that there is no dip in the supply and supply voltages are within the tolerance band of ±5% during the vibration recording. The signals are first recorded for no load condition. Next the motor is loaded by means of directly coupled dc generator feeding power to a lamp load. The recording of the signals is done on a set of five identical motors under same operating conditions, to ensure consistency in recorded data. In this manner a dataset for healthy bearing is created which consists of fifty signal records for each load condition for each motor. Having obtained the documented set of data for new bearing, bearing of each motor is in turn removed from the housing and replaced with a defective bearing in which known fault is introduced. The defects are introduced on a selected component of that bearing such as balls, inner race or outerrace one at a time, and the signals are recorded for varying load conditions. Onceagain, care is taken to ensure consistency in recorded data for different fault conditions. Rawvibration signal is analyzed in three domains: time domain, frequency domain and time-frequency to extract diagnostically important features. Initially, the signal is analyzed in time domain and five number of time features namely RMS level, Peak level, Crest factor, Skewness and Kurtosis are obtained from number of signal for same bearing condition. Though these feature are sufficient to know the health status of bearing however it was found that none of these feature is able to classify the component of bearing in which fault is developing. Next, raw vibration signal is analyzed in frequency domain by using Fast Fourier Transform and ARspectrum. The two methods are compared and it is concluded that though AR spectrum provides smooth spectrum, selection of proper model order increases complexity and therefore AR model can be used in parallel to FFT to support the diagnosis. In order to select important frequency components from vibration spectrum, feature selection criteria is applied. It is found that vibration spectrum of 0-250 Hz gives maximum fault information. In this manner, diagnostically important fault frequencies are identified and their amplitude levels are obtained. To obtain improved information on fault, raw signal is now analyzed in timefrequency domain by using Wavelet transform. Information on both time and frequency domain is extracted from the Wavelet Packet Decompositions ofraw vibration signal. In the final stage of work, a Knowledge Base (KB) is created by use of fuzzy membership function. The fault index of agiven vibration signal is obtained by matching the measured value ofthe frequency feature ofgiven vibration signal with that stored in the KB by using Fuzzy distance measure. With triangular membership, the fault index varies linearly. The Normalized Fault Index (NFI) is then obtained such that it gives '0' value when the measured value coincides with that value of KB for healthy condition, which is having maximum membership value and T when the measured value coincides with that value of the KB for faulty case, which is having maximum membership value. The strategy for fault detection and identification is as follows: Each time the signal processing procedure is completed, the chosen fault characteristics in time domain and/or in frequency domain and/or time-frequency domain of the measured vibration signal is compared with the with that of reference characteristic stored in the knowledge base. A weighted sum of difference between the actual and healthy feature is evaluated and compared with the set threshold, if it exceeds the threshold, checks are applied for fault detection. For example, a defect in inner race gives rise to frequency Ball Pass Frequency Inner Race (BPFI) with increased level. If the corresponding fault index exceeds the set threshold, the diagnosis process stops and the fault in inner race will be highlighted along with its fault index. The research work is summarized as • A health monitoring system is developed in laboratory for on-line monitoring of electrical rotating machine and has been successfully tested, both in laboratory and industrial site. • A database is created for induction motor for different bearing fault conditions. • A rigorous analysis of the obtained vibration signals has been carried out in time domain, frequency domain and time-frequencydomain. • The diagnostically important features are extracted and re-analyzed to establish suitability of the system to detect and identify the fault efficiently. • The vibration features are fuzzified to consider the variation in their values. • The 'Normalized Fault Index' based on fuzzy distance measure is introduced for machine fault detection. • The developed system is flexible in the sense that other types of fault can be accommodated. A Normalized Fault Index (NFI) for each fault is obtained with respect to healthy case. • The main contribution of this work has been development of Normalized Fault Index based diagnosis algorithm, which identifies the fault accurately on-line. • The approach can now be extended to development of fault diagnosis algorithm for other mechanical faults like air gap eccentricity, looseness of rotor bars or broken rotor bars, misalignment etc.
|Research Supervisor/ Guide:
Gupta, S. P.
|Appears in Collections:
|DOCTORAL THESES (Electrical Engg)
Files in This Item:
|ANALYSIS FOR FAULT IDENTIFICATION IN ROTATING ELECTRICAL MACHINES.pdf
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.