Sensorless fault diagnosis of induction motors - Networked ...

More documents

Recommendations

Info

KIM et al.: SENSORLESS FAULT DIAGNOSIS OF INDUCTION MOTORS 1047Fig. 7. Impact of speed estimation for 2.2-kW motor with broken rotor bars; current residual generated by measured speed and estimated speed (top),Indicator 1 (middle), and Indicator 2 (bottom).is used to load the induction motor. The second dc generator isused to measure the motor speed signal. An eight-channel Lab-VIEW-based data acquisition system is used to record the threeline voltages, the three line currents, and the generator speedsignal. All seven signals are sampled at 3840 Hz and the dataare collected for offline processing.Data from electric motor experiments conducted at thePublic Service Electric and Gas Motor Repair Facility, Sewaren,NJ, under the auspices of the Electric Power ResearchInstitute (EPRI) and the Electric Motor Predictive Maintenance(EMPM) Tailored Collaboration (TC) project are used. This isthe large machine test bed. A six-pole 373-kW and aeight-pole 597-kW induction motor are run directly from thepower supply mains. The motors are connected to dynamometersused to load then. A 13-channel IOTech data-acquisitionsystem is used to record the three line voltages, the three linecurrents, the encoder speed signal and six vibration signals.The currents, voltages and the encoder signal downsampled to3840 Hz are utilized for further processing.B. Impact of Speed Estimation on Fault Detection AccuracyThe developed motor predictors have the mechanical speedas an input. At the motor predictor training stage, the measuredspeed information is used as the speed input to these predictors.But speed measurements are not always available becausespeed sensors are not typically installed in induction motors,especially in small ones. Due to recent developments insensorless speed estimation schemes, accurate state filtering ofthe motor speed using only the electrical measurements is feasible,as discussed in previous sections. To demonstrate the effectivenessof the developed fault diagnosis system when speedestimates are used rather than speed measurements, both thespeed measurements and the speed filter output are utilized forresidual generation. Figs. 6 and 7 show this comparison for ahealthy motor and a motor having broken rotor bars, respectively.The measured speed is used for computing the fault indicatorsduring the first 2 s of the case studies, whereas theestimated speed is used during the next 2 s. All other measurementsare identical. In both figures, the magnitude of theresidual signal is indistinguishable throughout the tests, as is theindicator values. These figures demonstrate the accuracy of thedeveloped speed estimates and the robustness of the fault indicators.Consequently, most of the fault detection and diagnosis resultspreviously reported by the authors using speed sensor measurements[19], [20] are quite similar to the results presented inthis paper.
1048 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 50, NO. 5, OCTOBER 2003Fig. 8.Air-gap eccentricity tests for 597-kW motor; motor current spectra (top) and Indicator 2 (bottom).C. Quasi-Steady-State Machine OperationThe fault diagnosis mode proposed for the quasi-steady-stateoperation is applied to large and small machine staged fault data.The dynamics of large machines are usually slower than thoseof small machines. As a result the transients from disturbances,such as load and power supply variations, are slower and it isrelatively easier to find stationary segments than in small machines.1) Broken Rotor Bars: A mechanical motor fault consideredis that of broken rotor bars, and the experiments are performedto obtain motor measurements with differing number of brokenbars. After collecting the measurements, downsampling is accompaniedwith antialiasing low-pass filtering, and scaling. Thesix measurements, three voltages, and three currents, along withthe speed estimate, are processed through the signal segmentationalgorithm and their stationary segments are obtained. Theresiduals are generated by subtracting the measurements fromthe predictions. The residuals are converted to per unit base. Thebaseline is selected to be the computed healthy motor indicators.The two proposed fault indicators are computed by processingthe residual signal, as discussed before. The results, notshown here due to space limitations, indicate that Indicator 1,, remains unchanged compared to the baseline becausemechanical faults do not result in the unbalanced motor currents.Indicator 2, , reveals the presence of a fault wherethe magnitude of the indicator increases as the severity of thefault is increased by breaking an increasing number of rotorbars.2) Eccentric Air Gap: Another common mechanical motorfault is air-gap eccentricity. Such tests are performed consideringtwo different cases. The first case is a condition of moving25% upward the rotating center at the end of the inboard shaft,and the other case is a condition of moving 20% downwardand 10% right the rotating center at the end of outboard shaft.Fig. 8 shows the air-gap eccentricity test at 100% of rated load.The top segment of Fig. 8 shows the motor current spectra withhealthy condition and air-gap eccentricity. Distinguishing thetwo spectra is very difficult. The bottom segment of Fig. 8 showsIndicator 2, . Air-gap eccentricity makes the air gap distorted,resulting in the modulation of the current. Thus, comparedwith the baseline the harmonics of the residual signal areexpected to change, enabling detection of this fault.D. Transient Machine OperationThe fault diagnosis mode proposed for the transient operationis also applied to large and small machine staged fault data. Afew cases are presented here.1) Motor Bearing Deterioration: According the motorfailure surveys, bearing problems constitute 40% of all motorfaults. Fig. 9 shows the results from experiments with badbearings where deterioration resulted from defect in the balls,and inner and outer race. In Fig. 9, the top segment of the figureshows the motor current spectra with good and bad bearings.
Page 1 and 2: 1038 IEEE TRANSACTIONS ON INDUSTRIA
Page 9: 1046 IEEE TRANSACTIONS ON INDUSTRIA

Sensorless fault diagnosis of induction motors - Networked ...

Create successful ePaper yourself

Delete template?

Save as template?