stage of a future investigation into on-line multiple sensor type condition monitoring of induction machines .... mounted concentrically with the shaft on the rear of.
Australasian Universities Power Engineering Conference (AUPEC 2004) 26-29 September 2004, Brisbane, Australia
A Baseline Study for On-Line Condition Monitoring of Induction Machines I. Ahmed, R. Supangat, J. Grieger, N. Ertugrul and W. L. Soong University of Adelaide, School of Electrical and Electronic Engineering Adelaide, 5005, Australia Abstract Condition monitoring of induction machines relies on being able to detect the difference between healthy and faulty induction machines. An accurate interpretation of a motor’s condition requires knowledge of the effects of different operating conditions. In addition, variability between machines and test repeatability has to be considered. This paper investigates the above effects on the characteristic induction motor fault frequencies using results from an extensive series of tests on a number of identical healthy machines, which accommodates non-invasive measurements of axial flux, vibration, current and voltage of the motors under test. This work represents the baseline comparison stage of a future investigation into on-line multiple sensor type condition monitoring of induction machines, which aims is to set threshold levels for fault detection including level of loading and nonideal supplies. 1. INTRODUCTION Electric motors account for 95% of all prime movers in industrialized nations, and among these, three-phase induction motors consume typically 40 to 50% of all generated electrical energy. Induction motors are inherently reliable and require minimum maintenance. However, like other motors, they eventually deteriorate and fail. This is mainly due to a combination of environmental, duty cycle, installation and manufacturing (primarily in large motors) related factors. There has been a substantial amount of research on the development of various condition monitoring techniques over the past 15 years [1,2,3,4,5,6]. However, very limited studies have been done regarding non-invasive multiple sensor type based monitoring (which is the future aim of this work). Since not all faults can be detected with a single measurement technique, many faults have similar symptoms, and the fault may not be visible while the motor is operating under a light load. In addition, there is a lack of detailed understanding of the effect of the faults on the outputs of different sensor types. For example, it is well established that broken bars produce characteristic fault frequency sidebands, but these sidebands exist even in healthy motors, hence making it difficult to determine whether the measured sideband amplitude corresponds to a faulty motor or not. Moreover, the amplitude of these sidebands varies significantly as a function of the load on the motor. In addition, very little work has been reported on the detection of multiple faults or faults with similar symptoms. Furthermore, some of the external factors are not usually accounted for, such as distorted supply voltages and phase imbalances. The reliability of condition monitoring techniques depends upon the best understanding of electrical and mechanical characteristics of the machines in the healthy state and fault condition. A research towards
different techniques in combination with advanced computerized data acquisition and processing gives the new direction in the field of induction motor online condition monitoring [2,3]. This paper is the first stage to develop a comprehensive database of noninvasive sensor measurements of motors with known faults. In this stage it is aimed to understand the condition of the healthy machines as a function of loading, supply distortion and mounting irregularities. 2. APPROACH USED IN BASE-LINE STUDY The basis of any reliable condition monitoring method hinges on the understanding of the electric, magnetic and mechanical behaviour of the machine in a healthy state and under fault conditions. The induction machine is highly symmetrical, the presence of any kind of fault in it modifies its symmetry, and produces well-understood characteristic fault frequencies in the measured sensor signals. The changes in the magnitude of the fault frequencies can be used for fault detection. However, it is observed that even the healthy machines have a degree of signal magnitude at the fault frequencies. Furthermore it is not well understood how a fault signal magnitude in a given machine and operating condition relates to the presence or absence of specific faults. For example, eccentricity increases the amplitude of the rotor slot passing frequency, which can be observed in flux, current and vibration spectra. However, healthy machines also present measurable amplitude at the relevant fault frequencies. Non-invasive measurement techniques offer a more flexible solution in the condition monitoring, which is also performed in the study. Proposed non-invasive sensor types are vibration, motor current, axial motor flux linkage and voltage. Table 1 classifies the induction motor faults and summarise the detection methods and relevant sensors. The test specifications that can be utilised in a baseline study are also given below, in Table 2.
Table 1 Motor faults and detection methods GROUP OF FAULTS 1) Bearing Faults • 40-50% of failures • damaged races, balls, sleeve 2) Stator Faults • 30-40% of faults • insulation failure related
3) Rotor Bar/End-Ring Faults • 5 to 10% of faults • breakage or cracking of rotor bar or end-ring 4) Eccentricity • Unequal airgap – static and dynamic.
PARAMETER AND DETECTION METHOD • vibration • stator current spectrum • axial flux linkage • stator current components and spectrum • negative sequence currents or impedances detection • partial discharge • stator current spectrum • axial flux linkage • instantaneous power, torque and speed • vibration • current spectrum • vibration • stator voltage/current vector
Table 2. Baseline test specifications Motor setup No coupling, no-load, Coupled to the dc machine with no load Light load ……. ……. Full load 10% overload
Setup diagram
Aims Foundation test: To observe the effects of coupling and misalignment.
measurements. Some of these factors are the foundation of the motor and the alignment. The motors must have a firm and rigid foundation to eliminate soft foot and vibration. In addition, the misalignment reduces the efficiency of the motor setup and also the life-time of the motor. As stated above, the experimental setup has been designed and built in this study as repeatability and accuracy of the test bed is critical for reliable analysis of the results. Therefore, correct shaft alignment in the setup were ensured by using a laser alignment tool (developed for this study) and mounting was carried out by torque spanner (Norbar, 35Nm). The experimental work for this study was conducted using a test rig and a data acquisition system shown in Fig. 1 and Fig.2. The tests were conducted on a set of 6 new identical three-phase induction motors (415V, 4.8 A, 2.2kW, 4 pole, with 32 rotor bars) with dynamometer consisting of a 5kW separately excited DC load.
To investigate the effects of coupling and system inertia. To investigate the effects of loading and overloading.
The fault frequencies that are investigated are: • Slip frequency, [sf] from flux spectrum and its magnitudes • Broken rotor bar (BRB) side bands, [(1±2s) f] and their magnitudes • Eccentricity related fault frequencies, f((R/p)(1s) ± k for n= 1,2,3 …, and f ± fr. • Stator fault frequency, 2f on vibration and shorted turn related frequency bands on current and flux, [f((n/p) (1-s) ±k), n= 1,2,3,.. k = 1,3,5..] and their magnitudes. Here, s is the slip, f is the supply frequency, p is the number of pole pairs. The following issues were also considered in the base line study: • Each motor phase were analysed separately. • Observations carried out between the healthy motors • The tests were repeated for each motor on the same setup but different time. • Two fault baselines were obtained by using a motor with broken bar(s) and the motor is powered by a distorted three phase supply. 3. SPECIFICATIONS OF THE TEST SETUP Due to the nature of the measured parameters (high bandwidth, low magnitude), there are a number of factors, which can affect the accuracy of the
Figure 1. Photograph of data-acquisition hardware (topleft), motor/load setup (top-right), and the front panel of the data acquisition software.
Signal Cond.
Filter
DAQ card
PC
DEV Flux coil NDEH
DEH
Induction motor
DC generator
Figure 2 The block diagram of the test setup including the sensor positions. Where DEH represents Driving End Horizontal, DEV represents Driving End Vertical (DEV) and NDEH represents Non Driving End Horizontal vibration sensors.
Fig. 1 also shows a screen image of the LabVIEW data acquisition software used for data collection. Fig. 2 illustrates the overall block diagram of the test setup. During the tests a total of 8 parameters were measured. Two line voltages, two line currents, machine axial leakage flux (by a search coil that was mounted concentrically with the shaft on the rear of the motor), and three vibrations (mounted on driving end vertical, non- driving end horizontal and driving end horizontal) are measured simultaneously. The sensors used in the measurements and their specifications are given in Table 3. Furthermore, for the reliability and consistency of the measurements a fixed position for the flux coil was defined. The vibration sensors were screw mounted to the motor housing to achieve the highest bandwidth. Table 3. The sensor specifications. PARAMETER
DEVICE
Bandwidth
INPUT RANGE
Voltage Current
Isolation ampl.
30kHz
± 600V
Hall-effect clamp
50 kHz
± 10A
Flux
100 turns, search coil
10kHz
± 1V
Vibration
CTCAC102-1A piezoelectric accelerometers
20kHz
± 2g
The analog signal from the sensors is passed through variable gain amplifiers and low-pass filters to remove high frequency components that may cause aliasing. This is performed by an 8 channel, 8th order Butterworth analog low-pass filter unit. The unit also has individually adjustable channel gains of x1, x10 and x100 to allow amplification of any low level sensor signals. The filtered signal is then sampled by an AD converter to obtain a set of digital data. In order to prevent aliasing, the signal is sampled at more than the twice of the frequency of the highest signal component. The data acquisition hardware used in this study is a plug-in card from National Instruments (NI-PCI-6110, 12-bit, 5MS/sec, simultaneous sampling). All sensors were sampled simultaneously and the two different sampling rates are used as • Low-frequency measurement with a 400Hz sampling frequency (which gives a Nyquist frequency of 200Hz) and up to 100s sampling time, which allows very high-resolution frequency analysis (40000 data points, 0.01Hz resolution), • High-frequency measurement at 8000Hz sampling frequency with a sampling frequency of 5 seconds.
4. EXPERIMENTAL RESULTS AND ANALYSIS In this research, LabView software, sensors and toolset were used to collect and analysing the data to perform baseline study. Since frequency analysis technique using the Fourier transform is a very common signal processing method for the analysis of motors in healthy and faulty
conditions, obtaining the frequency spectra of each signal identified the fault frequencies listed above. In addition, the magnitude of the peaks found is normalized to give a fundamental (50Hz) peak of 0dB for analysis. Fig. 3 shows examples of waveforms obtained from each type of sensor in the data acquisition system with their corresponding power spectrums. The spectrum of the current, voltage and flux measurements were generated using data sampled at 400 Hz to give good resolution in the power spectrums at frequencies below 100 Hz, where most of the critical frequencies lie in these three signals. The vibration spectrum was generated using the data taken with a sampling rate of 8 KHz as most of the critical frequencies in the vibration signal are above 100 Hz but below 2 KHz. The voltage spectrum is used to determine an accurate value of the fundamental frequency of the mains supply, which then used to identify the location of the fault frequencies. In addition, the flux spectrum is used to determine the slip frequency and hence the amount of load placed on the motor. For example, for the sample spectrum given in Fig. 3 while the motor is running at rated current, the slip frequency causes a large peak in the flux spectrum at around 2.9 Hz (circle), indicating that this motor is operating at near full load. 4.1 Fault Frequency Characteristics The current, flux and vibration spectra are used to detect the presence of particular faults. Fig. 4 shows plots of the magnitude of the slip frequency in the flux spectrum on 6 different motors under varying loads. It is evident from these results that there is a significant variation among these motors. However, there does appear to be a general trend, the magnitudes seeming to increase as load is increased to about half of the rated load, then remaining nearly constant as the load is increased further. Figure 5 is given to investigate the repeatability of the tests (in Motor 1). Each test was conducted after the motor had been removed and then replaced, thus testing the consistency of the test setup as well as the testing procedures. The results indicate a good degree of repeatability in the test setup (considering the reduced scale in the figure). Figures 6 and 7 show the variations in the magnitude of the broken rotor bar sidebands of the 6 healthy test motors, in the current and the flux spectrum respectively. Significant amount of variations were observed among the test motors. It was also observed that the magnitude of the sidebands in the current spectrum is relatively high for light loads, which decreases as load is increased to about 25% of full load, then remains approximately constant as the load is increased further.
Figure 3 A set of measured waveforms from the sensors (top row) and their corresponding frequency spectrums (bottom row) when the motor was operating at its rated load. From left to right: stator current in phase A, mains line voltage between phase A and B, voltage induced in the axial flux search coil and acceleration of the drive end horizontal vibration sensor.
this section. The standard deviation plots were also presented in the results.
10
Magnitude (dB)
0 -10 -20 -30 -40 -50 -60 -70 0.00
25.00
50.00 75.00 Load (% of rated)
100.00
125.00
Figure 4 Magnitude of slip frequency in flux spectrum under varying loads for 6 healthy test motors.
Fig. 8 shows the characteristics of the average slip frequency magnitude of the healthy test motors. It can be observed from this figure that at light loads (up to ~45%) the slip magnitude increases at a higher rate than at heavy loads (~above 50%). In addition, as demonstrated by the standard deviation plots that there is a significant degree of variation in slip magnitude among the healthy motors. The standard deviation (s.d) plots in the figures were obtained using standard deviation of each data point of the polynomial fit: Upper s.d. curve = mean +s.d.
-40
-20
Flux Slip Frequency
-30 -60
-80 0.00
(1-2s)f
25.00
50.00
75.00
Current Spectrum BRB Sidebands
100.00
12
(1+2s)f
Magnitude (dB)
Magnitude (dB)
Lower s.d. curve = mean –s.d.
Load (% of rated)
-50 -60 -70 -80 -90 0.00
Figure 5 Repeatability tests on Motor 1.
25.00
50.00
75.00
100.00
125.00
100.00
125.00
Load (% of rated)
a -20 -30 Magnitude (dB)
However, the trend in the flux spectrum (Figures 6 and 7) seems to be relatively steady. It should be reported here that the noise level in the current spectrum was about –90dB, and it was about –70dB in the flux spectrum, which verifies that the detection of BRB sideband using the flux spectrum may not be considered as a reliable method.
-40
-40 -50 -60 -70 -80
4.2 Spectrum Average Analysis Both due to the difficulties in the presentation of a large number of data points and the difficulties of testing the motor under same loading conditions, spectrum averaging is performed. Averaging each motor measurement and then approximating by a 6th order polynomial fit obtained the results presented in
-90 0.00
25.00
50.00
75.00
Load (% of rated)
b Figure 6 BRB sideband test results based on the current. a) (1-2s)f , b) (1+2s)f.
-20
Magnitude (dB)
-30 -40 -50 -60 -70 -80 -90 0.00
a 25.00
50.00
75.00
100.00
125.00
Load (% of rated)
a -20
Magnitude (dB)
-30 -40 -50 -60 -70 -80 -90 0.00
25.00
50.00
75.00
100.00
125.00
Load (% of rated)
b Figure 7 BRB sideband test results based on the flux. a) (1-2s)f , b) (1+2s)f.
Figure 8 Magnitude of the slip frequency in the flux spectrum versus load (solid line) and ± standard deviation (s.d.) from the average (dotted line).
Fig. 9 shows the characteristics of the average BRB sidebands magnitude, which is useful in detecting broken rotor bars. The analysis results show that that the magnitude of the BRB sidebands tend to increase at light loads (up to ~40%) then level off at heavy loads. The BRB lower sideband magnitude is on average higher in the flux spectrum than in the current spectrum by about 20 dB, which is due to the noise level in the flux spectrum. Fig.9 also shows a comparison between healthy and faulty motors (4 broken rotor bars on Motor 1). As it is expected, when there is a broken rotor bar, the BRB sideband magnitude increases significantly, which goes beyond the upper standard deviation curve, from its corresponding average. In addition, it was interesting to note that under light loads, it may be difficult to identify the broken rotor bar fault observing the magnitude of the lower side bands, which is valid both for the current and the flux measurements. In addition, the peak may be close to 50Hz peak and the side lobs of that peak at higher level of noise on the measured signals, which can make the identification of the lower side bands difficult.
b
Figure 9 The characteristics of the average BRB sidebands magnitude of the flux and the current as a function of the load. a) BRB lower, (1-2s)f sidebands magnitude b)BRB upper, (1-2s)f sidebands magnitude
Fig. 10 illustrates the characteristics of the average vibration spectrum magnitude at twice the fundamental frequency (2f) from the three vibration sensors, which is useful in detecting stator faults. It was observed in these results that the two horizontal oriented vibration sensors have similar magnitude on average. The vertical oriented vibration sensor however produces lower magnitude at the 2f frequency. It was also noted that the vibration spectrum magnitude tends to decrease at light loads (up to ~60%) then level off at heavier loads, which may be related to the resonance frequency of the machine setup. In Fig. 10, the effects of unbalance supply voltage (5% unbalance on Phase C) toward the vibration is given. Unbalanced supply voltage can be considered as stator faults, which increases the vibration magnitude significantly at the 2f frequency. However, no noticeable effects have been observed at other fault frequencies.
a
b Figure 10 The characteristics of the average vibration spectrum magnitude at twice the fundamental frequency (2f) from the three vibration sensors: a) DEH and DEV b) NDEH.
The effects of the distorted but balanced three-phase supply voltage were also investigated in this study, but no significant effect has been observed. Fig. 11 show the characteristics of the average shorted turns critical frequencies [f((n/p)(1-s) (+/-) k)]. The figures demonstrate that the magnitude of the shorted turn critical frequency obtained from the flux
spectrum are higher by about 30 dB on average in comparison to the current spectrum. This suggests that detecting shorted turn critical frequency is easier in the flux spectrum than in the current spectrum. It is also observed that the magnitude at the shorted turn frequencies tends to drop slightly as the load increases.
Figure 11 The characteristics of the average shorted turns critical frequencies: a) f1((n/p)(1-s)+k), b) f1((n/p)(1-s)-k). Note: k = 1 and n= 1 were found to be the most detectable peak in this test.
Figure 12 illustrates the characteristics of the average eccentricity critical frequencies [f1((R/p)(1-s) ± k) and f±fr], where fr is the rotor frequency. In Fig. 12, it can be seen that the magnitude of the first eccentricity critical frequency, f((R/p)(1-s)+k, obtained from the flux spectrum is slightly higher (~5dB) on average than the magnitude obtained from the current spectrum. In addition, the magnitude from the flux spectrum increases at light load (up to ~40%) then becomes steady at heavier loads, and not significant variations were observed on the magnitude of the current spectrum.
a
b
c
Figure 12 The characteristics of the average eccentricity critical frequencies: a) f((R/p)(1-s)+k, b) f-fr and c) f+fr . Note: k = 1 was found to be most detectable peak in this test.
Furthermore, it can be concluded that the other eccentricity critical frequency, f±fr, are more detectable in the flux spectrum than in the current spectrum since the difference is higher than ~20 dB. Note also that the magnitude at these two frequencies tends to drop slightly as the load increases.
5. CONCLUSIONS In the paper, extensive tests and analysis were carried out on the healthy induction motors to investigate the
variations of the fault frequencies as a function of motor load. Four different non-invasive sensor based measurements were performed, which include axial flux, vibration, motor current and voltage. The accuracy and the repeatability of the test setup were also demonstrated. A number of results were presented to provide a baseline comparison for the future investigations into on-line multiple-sensor condition monitoring. Significant amount of variations were observed among the test motors, which some has a specific trend with load. In addition, it was verified that the detection of broken rotor bar sideband using the current spectrum is more reliable than the flux spectrum. However, it was observed that under light loads, it may be difficult to identify the broken rotor bar fault observing the magnitude of the lower side bands of the current as well as the flux spectrum. The future direction of this study aims to develop a comprehensive data base of non-invasive sensor measurements of induction machines with different faults and to understand the effect of faults on each sensor output. 6. REFERENCES [1] M. L. Sin, W. L. Soong and N. Ertugrul, “On – Line Condition Monitoring and Fault Diagnosis – A Survey” Australian Universities Power Engineering Conference, New Zealand, 2003. [2] IEEE Motor Reliability Working Group, “Report of Large Motor Reliability Survey of Industrial Commercial Installations” Part 1, IEEE Transaction on Industry Applications, vol. 21, July, pp, 853-872, 1985. [3] M.E.H. Benbouzid, “A Review of Induction Motors Signature Analysis as a Medium for Faults Detection”, IEEE Transaction on Industry Electronics, vol. 47, no. 5, Oct., pp. 984-993, 2000. [4] J.M. Cardoso, S.M.A. Cruz and D.S.B. Fonseca, “Inter-Turn Stator Winding Fault Diagnosis in Three Phase Induction Motors by Park’s Vector Approach”, IEEE Transactions on Energy Conversion, Vol. 14, No. 3, September 1999, pp. 595-598. [5] W.T. Thomson, D. Rankin and D.G. Dorrell, “Onlin Current Monitoring to Diagnose Air Gap Eccentiricity in Large Three-Phae Indiuction Motors-Industrial Case Histories Verify the Predictions” IEEE Transactions on Energy Conversion, Vol. 14, No. 4, Dec. 1999, pp. 13721378. [6] C. Kral, T.G. Habetler, and R.G. Harley, “Detection of Mechanical Imbalances of Induction Machines Without Spectral Analysis of TimeDomain Signals”, IEEE Transaction on Industrial Applications, Vol. 40, No. 4, July?August 2004, pp. 1101-1106.
