Reliable Detection of Induction Motor Rotor Faults ... - IEEE Xplore

IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 50, NO. 4, JULY/AUGUST 2014

2493

Reliable Detection of Induction Motor Rotor Faults Under the Rotor Axial Air Duct Influence Chanseung Yang, Student Member, IEEE, Tae-June Kang, Student Member, IEEE, Doosoo Hyun, Student Member, IEEE, Sang Bin Lee, Senior Member, IEEE, Jose A. Antonino-Daviu, Senior Member, IEEE, and Joan Pons-Llinares, Member, IEEE

Abstract—Axial cooling air ducts in the rotor of large induction motors are known to produce magnetic asymmetry and can cause steady-state current or vibration spectrum analysis based fault detection techniques to fail. If the number of axial air ducts and that of poles are identical, frequency components that overlap with that of rotor faults can be produced for healthy motors. False positive rotor fault indication due to axial ducts is a common problem in the field that results in unnecessary maintenance cost. However, there is currently no known test method available for distinguishing rotor faults and false indications due to axial ducts other than offline rotor inspection or testing. Considering that there is no magnetic asymmetry under high slip conditions due to limited flux penetration into the rotor yoke, the detection of broken bars under the start-up transient is investigated in this paper. A wavelet-based detection method is proposed and verified on custom-built lab motors and 6.6-kV motors misdiagnosed with broken bars via steady-state spectrum analysis. It is shown that the proposed method provides the reliable detection of broken bars under the start-up transient independent of axial duct influence. Index Terms—AC machines, fault diagnosis, induction motors, rotor fault, start-up analysis.

I. I NTRODUCTION

M

OST of the rotors of induction motors rated above 100 kW employ axial air ducts, as shown in Fig. 1, for the effective cooling of the rotor, reduction in rotor weight (inertia), and saving of material costs [1]–[3]. The most common axial air duct design is to use shaft spiders, where four, six, or eight radial arms are welded to the shaft to support the rotor core, as shown in Fig. 1(b) and (c). Axial ducts can Manuscript received September 25, 2013; revised November 7, 2013; accepted December 16, 2013. Date of publication January 2, 2014; date of current version July 15, 2014. Paper 2013-EMC-686.R1, presented at the 2013 IEEE Energy Conversion Congress and Exposition, Denver, CO, USA, September 16–20, and approved for publication in the IEEE T RANSACTIONS ON I NDUSTRY A PPLICATIONS by the Electric Machines Committee of the IEEE Industry Applications Society. This work was supported in part by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (NRF-2013R1A1A2010370) and in part by the Human Resources Development Program (20134030200340) of the Korea Institute of Energy Technology Evaluation and Planning Grant funded by the Korea Government Ministry of Trade, Industry, and Energy. C. Yang, T.-J. Kang, D. Hyun, and S. B. Lee are with the Department of Electrical Engineering, Korea University, Seoul 136-701, Korea (e-mail: [email protected]; [email protected]; [email protected]; [email protected]). J. A. Antonino-Daviu and J. Pons-Llinares are with the Department of Electrical Engineering, Universitat Politècnica de València, 46022 València, Spain (e-mail: [email protected]; [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TIA.2013.2297448

also be produced by punching holes in the rotor laminations prior to assembly, as shown in Fig. 1(a). Axial ducts can cause the magnetic reluctance of the flux path to be asymmetric since the spider arm and shaft can provide an alternate path for the flux [4]–[6]. The difference in the flux path depending on the relative position between the magnetic field and axial duct (spider) arms can be clearly seen in the results of the 2-D finite element (FE) analysis in Fig. 2(a) and (b) for a fourpole motor with four axial ducts. The variation in the magnetic reluctance results in the modulation of the magnetizing current, where the frequency of modulation depends on the number of axial air ducts, Nd , and poles, Np . It was shown in [4]–[9] that the magnetic asymmetry can produce components (fduct ) in the steady-state vibration or current spectrum that overlap with motor fault frequency components such as rotor cage damage or air gap eccentricity if Nd is equal to Np , as shown in (1). fe is the input electrical frequency, and k is a positive integer. If the number of air ducts, Nd , and poles, Np , are the same, the air ducts can produce frequency components identical to that of the rotor cage fault frequency (fbrb ), as can be seen in (2). This can be misinterpreted as broken rotor bars when steady-state online FFT-based techniques are applied, as reported in [4]–[9] fduct,vib = 2ksfe , fduct,cur = (1 ± 2ks)fe (if Nd = Np ) (1) (2) fbrb,vib = 2ksfe , fbrb,cur = (1 ± 2ks)fe . Rotors for cases where the axial ducts caused false broken bar indications on Nd = Np = 8 or 4 6.6-kV induction motors operating in power plants are shown in Fig. 1. The results of motor current signature analysis (MCSA) of the three motor samples showed strong broken bar components (see Fig. 3); however, the inspection of the rotor showed that the bars were in good condition for all three motors. The cost of inspection for a motor with a false positive indication is typically tens of thousands of U.S. dollars (USD) without including the loss of production (if standby or spare motors are available). The cost is higher for larger motors due to the difficulty of handling the heavy motors as it involves motor detachment, installation (spare motor), rental of crane/equipment, shipping to motor repair shop, and disassembly/inspection. The inspection cost alone for the 2.4-MW motor of Fig. 1(a) exceeded 100 000 USD. Despite the high cost of inspection, motors are inspected if fault indications are given since the cost of forced outage can be orders of magnitude higher than that of inspection/repair. False positive rotor fault indications due to axial duct influence are a very common ongoing problem in the field, considering that a large portion (34.4%) of high-voltage motors

0093-9994 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

2494


Fig. 1. Rotors of 6.6-kV motors with false positive rotor fault MCSA indications due to axial ducts. (a) 2400-kW eight-pole induced draft fan motor. (b) 280-kW four-pole condensate pump motor. (c) 350-kW four-pole condensate pump motor (test samples A–C).

Fig. 2. FE analysis of flux distribution in four-pole motor with four axial air ducts when magnetic poles and duct/spider arms are electrically (a) 90◦ apart and (b) aligned (0◦ ) under steady-state operation.

are of Nd = Np design, according to the survey in [6]. In most cases, it is unknown whether a motor is of Nd = Np design since the number of axial ducts is unknown by the user after the motor is assembled. Even if it is known that Nd = Np for a particular motor, it is uncertain whether the air ducts would produce the fduct component or not since not all motors with Nd = Np produce false indications. Whether fduct components are produced depends on the rotor design and construction, material characteristics, and operating conditions [6]. An attempt was made to separate the effect of broken bars and axial ducts online in [4], and guidelines for distinguishing the two problems were given in [5]–[8]; however, there are many limitations in the field for applying what has been suggested. In [6], a detailed analysis of the influence of axial ducts on condition monitoring is presented. It is shown that broken bars can be detected independent of axial ducts if tested under high slip (standstill and offline) conditions since there is no magnetic asymmetry due to the limited flux penetration in the rotor (cage eddy current rejection). The downside of offline test methods is the requirement of motor disassembly or manual shaft rotation for testing, which makes frequent, automated, and remote motor testing difficult [10]. The aforementioned arguments show that false indications due to axial ducts are common and the consequence can be serious; however, there currently is no practical online test method available for separating the influence of broken bars with axial ducts. Based on the observation that broken bars

Fig. 3. Is spectrum (MCSA) of “healthy” 6.6-kV motors shown in Fig. 1 with false positive broken rotor bar indications. Samples (a) A1 and A2 (Nd = Np = 8). (b) B1, B2, and B3 (Nd = Np = 4). (c) C1, C2, and C3 (Nd = Np = 4).

can be detected reliably independent of the axial duct at high slip, testing under the start-up transient is investigated in this paper. Although many papers have been published on the startup detection of rotor faults, this is the first time that it is applied to the axial duct problem. A wavelet transform-based method is proposed as an effective solution and verified on a motor with a

YANG et al.: DETECTION OF INDUCTION MOTOR ROTOR FAULTS UNDER ROTOR AXIAL AIR DUCT INFLUENCE

2495

custom-built rotor in the laboratory and on 6.6-kV motors in the field with false broken bar MCSA indications. It is shown that the proposed method is capable of providing reliable detection of broken bars independent of the axial duct influence.

II. I NFLUENCE OF ROTOR A XIAL A IR D UCTS ON MCSA A. False Positive Broken Rotor Bar Indications MCSA was performed for the healthy 6.6-kV motors of Fig. 1 with false positive broken rotor bar indications due to axial ducts (Nd = Np ). The three types of 6.6-kV motors of Fig. 1(a)–(c), which are fan or pump motors used in power plants, are denoted as samples A, B, and C, respectively. The ratings, application, and figure number of the experimental results are summarized in Table I of the Appendix. The MCSA results of samples A, B, and C are shown in Fig. 3(a)–(c), respectively, where the measurements were obtained from two or three units of identical design that share the load. For the 2.4-MW induced draft fan motor, both samples A1 and A2 showed very strong f1−2s sidebands of −35.9 and −36.2 dB, respectively, in the current spectrum [see Fig. 3(a)]. Strong f2s components indicating broken rotor bars were also observed with vibration analysis, but motor inspection did not reveal signs of rotor bar or end ring damage. For the 280- and 350-kW condensate pump motor samples B and C, the MCSA measurements of the f1±2s component for three identical units were not consistent, as can be seen in Fig. 3(b) and (c). The f1−2s component was −52.5 dB, below −70 dB, and −69 dB for samples B1, B2, and B3 and −45.1 dB, −59.5 dB, and −53.4 dB for samples C1, C2, and C3, respectively. Broken bars were strongly suspected for units B1 and C1, considering that the f1±2s component of −55 ∼ −35 dB is the fault threshold used in commercial MCSA products and because the f1±2s components were significantly higher than that of motors of identical design. Since the sample B1–B3 and C1–C3 motors were manufactured and commissioned at the same time and operated under similar load, it was natural to conclude that it is very unlikely for the motors to have larger f1±2s components unless a rotor fault is present. However, the results of offline inspection and testing showed that the rotor cages were in good condition. The single phase rotation test was performed on sample C1 with 380 V applied between two terminals for pulsating field excitation [10], [11]. The results of the current measurements versus rotor position shown in Fig. 4(a) imply that there is no asymmetry in the rotor cage (current varies with rotor position for the faulty case). The FE simulation under pulsating field excitation at motor standstill (s = 1) in Fig. 4(b) shows that flux penetration into the rotor yoke is limited. The inconsistency in the amplitude of the f1±2s component for samples B and C is suspected to be caused by part-to-part variation introduced due to component and manufacturing tolerances such as the variance in the Si–Fe lamination magnetic characteristics, the nonideal anisotropy in the rotor laminations, etc. [6]. It was concluded based on the inspection and test results that the false positive indication for samples A, B, and C is due to the axial air ducts for the Nd = Np motors.

Fig. 4. (a) Single phase rotation test results (current versus rotor position) on sample C1 under 380-V pulsating field excitation at motor standstill. (b) FE simulation of flux distribution under single phase rotation test.

B. Interaction Between Rotor Magnetic Asymmetry and Faults If the f1±2s components exceed a predetermined threshold level with spectrum analysis, the criteria that knowledgeable field maintenance engineers use to confirm broken bars and screen out potential false positive indications are as follows [5]–[8]. 1) Observe the change in the f1±2s component over time: If the amplitude of the f1±2s component increases with time under similar load conditions, it is concluded that there is rotor cage damage; if it is constant or decreases with time, a false positive indication is suspected. 2) Observe the change in the f1±2s component with load variation: If the amplitude of the f1±2s component increases with increase in load, it is concluded that there is rotor cage damage; if it is constant or decreases with increase in load, a false positive indication is suspected. 3) Compare the f1±2s component with motors of identical design: If the amplitude of the f1±2s component is considerably larger compared to that of motors of identical design, rotor damage is suspected. However, it was shown in [6] that these are misconceptions and may lead to incorrect diagnosis for motors of Nd = Np design. This is because the f1±2s components due to the axial duct (fduct ) and the broken bar (fbrb ) interact, and the amplitude can change with operating condition (e.g., voltage level) and due to the characteristics or construction of individual motors. The interaction between the broken bar and axial duct influence has been investigated in [6], and it is shown that the fbrb and fduct components can add or cancel out, depending on the relative physical angle between the duct and broken bar, represented as φe in Fig. 4(b). The analysis is verified in the laboratory on two four-pole 380-V motors with custom-built Nd = Np rotors shown in Fig. 5. The axial ducts of sample D were created by drilling 20 holes in the yoke to produce four axial ducts in the yoke of an Al die-cast rotor, as shown in Fig 5(a). A fabricated Cu bar rotor with four axial ducts, sample E, was designed and built to fit the stator of sample D, as it is representative of large motors [see Fig. 5(b)]. Axial holes were laser cut in the rotor laminations of sample E, as shown in Fig. 5(c). The MCSA measurements performed on samples D–E with zero, one, and two broken bars located at φe = 45◦ and φe = 135◦ are shown in Figs. 6 and 7, respectively.

2496


(counter-example of criterion 2). The MCSA results of samples B1–B3 and C1–C3 are counter-examples of the third criterion listed earlier. The false rotor fault indication (B1 and C1) is not necessarily present for all motors of the same design due to the variance in the characteristics of individual motors, as described previously in Section III-A. The results of Figs. 3, 6, and 7 disprove the validity of the aforementioned criteria 1), 2), and 3) for confirming broken bars and show that rotor faults in motors with Nd = Np design cannot be reliably detected using spectrum analysis. Fig. 5. (a) Al die-cast rotor with four drilled axial holes (sample D). (b) Fabricated Cu rotor with four laser cut axial holes in rotor lamination (sample E). (c) Rotor lamination for sample E.

Fig. 6. MCSA measurements of the f1−2s component as a function of % rated slip with 0–2 broken bars (sample D) located at (dotted line) φe = 45◦ and (real line) φe = 135◦ .

III. D ETECTION OF ROTOR FAULTS I NDEPENDENT OF A XIAL D UCT I NFLUENCE U NDER THE S TART-U P T RANSIENT Since steady-state fast Fourier transform (FFT)-based spectrum analysis cannot provide reliable detection of broken bars for Np = Nd motors with magnetic asymmetry, a new online detection method that is not influenced by axial ducts is needed. Offline standstill tests such as the single phase rotation test provide reliable detection of rotor faults since they are performed under high slip conditions when flux penetration in the rotor is limited (see Fig. 4). Fault detection under the start-up transient, where the slip is high and flux penetration into the rotor yoke is limited, is therefore considered under the expectation that magnetic asymmetry (and false positive indications) due to axial ducts can be avoided. In this paper, the discrete wavelet transform (DWT), which has been successfully applied to motor fault detection under transient conditions [12]– [17], is considered for analyzing the start-up current of Nd = Np motors for the first time. A. Principles of DWT-Based Rotor Fault Detection

Fig. 7. MCSA measurements of the f1−2s component as a function of % rated slip with 0–2 broken bars (sample E) located at (dotted line) φe = 45◦ and (real line) φe = 135◦ .

It can be observed that the f1−2s component increases if the bar is broken at the φe = 45◦ location shown in Fig. 4(b) and decreases (and then increases) if φe = 135◦ . This is observed because the fbrb and fduct components are in phase (add) when φe = 45◦ and out of phase (cancel out) when φe = 135◦ [6]. If the fault is located at φe = 135◦ , the f1−2s component initially decreases but starts to increase if the magnitude of the fbrb component exceeds that of the fduct component. It can be clearly seen in the results of Figs. 6 and 7 that the f1±2s components can remain constant or decrease with increase in broken bars over time, depending on the fault location, the load, and the amplitude of individual fduct and fbrb components (counter-example of criterion 1). The f1±2s components can also decrease with increase in load for a motor with broken bars, depending on the fault location, load, and fault severity

The DWT performs the decomposition of a signal into multiple subsignals that contain different frequency ranges of the original signal according to predetermined frequency bands. This makes DWT a powerful tool for analyzing nonstationary signals with varying fault frequency components as in the case of the f1±2s components of (1) and (2) under the start-up transient. Moreover, the implementation of the DWT based on the Mallat algorithm makes it a tool with low computational requirements in comparison to other time–frequency decomposition tools and FFT [17], [18]. The mathematical expression that characterizes the DWT decomposition of a signal, e.g., i(t), is given by (3), where an and dn . . . d1 represent the approximation and detail signals, respectively. The details of the DWT are not described in this paper, as it can be found in many resources [12]–[18] i(t) = an + dn + dn−1 + · · · + d1 .

(3)

The most important underlying idea in the application of the DWT is that each wavelet signal is associated with a certain frequency band, i.e., the frequency components within the band are extracted from the original signal into each wavelet signal. The range of the band covered by each wavelet signal depends on the sampling frequency fs and on the level n of the wavelet signal, as illustrated in Fig. 8. In the figure, an example of the frequency range for each wavelet signal is shown for


2497

Fig. 8. Decomposition process carried out by DWT for fs = 10.24 kHz and n = 9.

Fig. 9. Evolution of slip and the Λ-shaped broken rotor bar, |fbrb |, component during a direct online start-up transient measured for a 5.5-kW motor. The concept of fault indicator, En , calculation is illustrated.

fs = 10.24 kHz and n = 9 level decomposition (see guidelines for the selection of DWT parameters in [13] and [25]), which are the conditions used throughout this paper. The fault frequency associated with broken bars changes during the direct online start-up transient as the slip s decreases from 1 to a value close to 0 [12]. At the instant that the motor source switch is closed, |fbrb | is equal to fe (s = 1) and decreases to 0(s = 1/2), and it increases back toward fe as s approaches 0. This leads to the characteristic Λ-shaped evolution of |fbrb | illustrated in Fig. 9, which shows the variation in the slip and |fbrb | component measured for a 5.5-kW induction motor start-up. If the DWT is applied to the start-up current of one phase of a machine with broken bars, the evolution of |fbrb | will be reflected in the wavelet signals covering the frequency band of 0 ∼ fe Hz, enabling reliable detection of the fault. This is illustrated in Fig. 10(a) and (b), which shows the start-up current, i, and high-level wavelet signals, a9 , d9 , and d8 , resulting from the DWT of the start-up current of one phase for sample F with healthy (F1) and faulty (F2) rotor cages. Sample F is a 6.6-kV 3800-kW water intake pump motor used at a water supply facility. The wavelet signals, a9 , d9 , and d8 , cover the range of 0–40 Hz (see Fig. 8) for the specific case analyzed. The oscillations in certain wavelet subsignals increase at a particular interval of time, forming the Λ pattern as the |fbrb | component changes with slip at start-up, only if a broken bar is present. The magnitude of the oscillations in the DWT analysis due to the rotor fault |fbrb | component is proportional to the magnitude of the f1±2s component in the MCSA results of samples F1 and F2 shown in Fig. 11. No oscillations are expected for healthy motors for which steady-state FFT gave false broken bar indications (samples A–E) since magnetic asymmetry does not exist under start-up. The oscillation at 6.5 s in Fig. 10 is due to incorrect timing in the closing of the starting reactor bypass switch. Reactor start-up is used in this application for limiting the high start-up current (∼2000 A) by reducing the voltage; however, it can be seen that the timing is set incorrectly. If the switch is closed correctly after reaching steady state, it does not interfere with DWT start-up detection.

Fig. 10. Start-up current and high-level wavelet signals, a9 , d9 , and d8 , resulting from the DWT of the start-up current for sample F. (a) Healthy motor (F1). (b) Motor with rotor fault (F2).

Fig. 11. Is spectrum (MCSA) of sample F (6.6 kV, 3400 kW, Nd = 8, and Np = 12) motor with healthy (F1) and faulty (F2) rotor cages.

B. Quantification of Fault Severity The Λ-shaped pattern in the high-level wavelet signals (a9 , d9 , and d8 ) in the [0,40]-Hz range resulting from the DWT is a reliable qualitative indicator of the presence of rotor faults. The existence of the rotor cage faults in the machine can be identified since no other phenomenon or fault is likely to cause a similar pattern. It is desirable to define a quantitative indicator capable of providing the following: 1) an automated decision on fault existence and 2) a measure of fault severity to complement the qualitative information provided by the Λ pattern. Although there could be numerous wavelet-based fault indicators that can be defined for the quantification of fault severity, an energybased parameter is proposed in this work for simplicity of the algorithm. The energy level of the high-level wavelet signals increases due to the oscillations in the transient evolution of |fbrb | as the severity of the fault increases. Therefore, a quantification

2498


parameter based on the energy level of the wavelet signals influenced by |fbrb | (a9 , d9 , or d8 ) is introduced due to (2). According to an extensive study on the start-up transient current data obtained from motor samples B–F, it was observed that the oscillations in the falling part of Λ of the d8 signal provide the most sensitive indication of the fault existence and severity. The energy in a predetermined time interval of d8 is normalized with respect to the start-up current within the same time interval to make it independent of the operating conditions such as the inertia, load level, etc. The proposed fault indicator En can be calculated using ⎞ ⎛ j=N j=N 83 83 (4) d8 (j)2 i2j ⎠ En (dB) = 10 · log ⎝ j=N67

j=N67

where d8 (j) and ij are the values of the jth sample of the d8 and start-up current signal, respectively. The time interval [N67 , N83 ] used for calculating En corresponds to the interval between 67% and 83% of the startup time, Tstart , shown in Fig. 9. The value of Tstart can be calculated from the pattern of decrease in the start-up current magnitude. In this paper, the end of the start-up transient period Tstart was set to the time when the rate of decrease in current falls below a predetermined limit. If it is assumed that slip linearly decreases from 1 to 0 from the instant of start-up (t = 0 s) to Tstart , 67%–83% of Tstart approximately corresponds to the frequency range of 20–40 Hz, as shown in Fig. 9. This is the range for which d8 shows rotor fault-related oscillations, which is the reason that En is calculated in this time interval. Since slip does not decrease linearly with time, the [N67 , N83 ] interval does not exactly coincide with the oscillations, as can be seen in the highlighted windows of Fig. 10. Although it does not capture the oscillations precisely, it is shown in Section IV that it is sufficient for detecting the fault and providing a measure of fault severity. There are numerous techniques other than DWT such as continuous wavelet transforms, Hilbert–Huang transforms, Wigner–Ville and Choi–Williams distributions, etc., that can be applied to this problem [19]–[24]. The purpose of this paper is to show that start-up current analysis provides reliable detection of rotor faults independent of the axial duct influence. Therefore, the application of time–frequency decomposition tools other than DWT or alternate fault indicators are not within the scope of this paper and are left for future work. IV. E XPERIMENTAL S TUDY An experimental study was performed to verify the effectiveness of the proposed DWT-based method for detecting broken bars reliably independent of the axial duct influence. The startup currents of samples B, C, D, and E (Np = Nd ), which were diagnosed with false broken bars in steady state (MCSA), were obtained and analyzed. The information, rotor photographs, test results for MCSA, and DWT analysis for motor samples B–E are summarized in Table I (see the Appendix). For the custombuilt motor samples D and E of Fig. 5, up to two Al or Cu bars were cut at the bar and end ring joint to simulate broken

bars. To test under the condition where the fbrb and fduct components are in phase (add), bars at the φe = 45◦ location shown in Fig. 4(b) were cut. The case where fbrb and fduct components are out of phase (cancel; φe = 135◦ ) was also tested by rotating the same rotor used for the φe = 45◦ case in the opposite direction. This is equivalent to the broken bar being at φe = 135◦ . For samples B and C, start-up transient data were measured for the motors with the highest fbrb components for each sample (B1 and C1) and for sample B2. It was not possible to damage the rotor bars of these samples since they are HV motors operating in the field. The start-up current of one phase was sampled at 10.24 kHz and analyzed with Daubechies wavelets. This sampling rate constricts the frequency range of d8 below 40 Hz to prevent interference with the mains frequency at 60 Hz [17], [25]. Since the oscillation in d8 in the falling end of Λ is used for calculating En , the influence of the electromagnetic transients at the instant of start-up (see Fig. 10) can also be avoided. To produce the start-up time typical of large motors, a steel disk designed to increase the start-up time to 1.0–1.5 s was attached to the shaft of the unloaded motor for samples D and E. The motor was not loaded, and nothing was connected to the rotor shaft other than the inertia disk, to observe the fault detectability under the worst case condition of no load. A. Experimental Results: Laboratory Test (Samples D and E) The MCSA results for samples D and E with rotor faults located at φe = 45◦ and 135◦ in Figs. 6 and 7 clearly show that MCSA cannot be used for detecting rotor faults for Nd = Np motors since it is influenced by the location of the broken bar with respect to the axial duct. The waveforms of the start-up current and d9 and d8 wavelet signals for sample D are shown for zero, one, and two broken bars at φe = 45◦ and 135◦ in Figs. 12 and 13, respectively. The time interval of d8 used for calculating En described in Section III is highlighted in the figure, and the values of En calculated for each condition are plotted in Fig. 16. For the healthy cases [see Figs. 12(a) and 13(a)], where steady-state MCSA indicated faulty rotors (see Fig. 6), the oscillations are negligible, and the values of En are small. It can be observed in Figs. 12, 13, and 16 that the oscillations in d8 due to broken bars and the values of En increase with the number of broken bars. For the case of φe = 135◦ , in which the fbrb components decrease with broken bars (see Fig. 6), the magnitude of the oscillations and values of En are similar to that of the φe = 45◦ case (see Figs. 13 and 16). The fact that the increase in the d8 oscillations (see Figs. 12 and 13) and En (see Fig. 16) are similar regardless of fault location, φe , verifies that fault detection under the startup transient provides fault indications only if broken bars are present, and that is not influenced by axial ducts. The waveforms of the start-up current and d9 and d8 signals and the time interval used for En calculation for sample E are shown for zero, one, and two broken bars at φe = 45◦ in Fig. 14. The results for the φe = 135◦ case, which is not shown due to space restrictions, are similar to that of Fig. 14. The values of En calculated for the φe = 45◦ and 135◦ cases for sample E in Fig. 16 show that the proposed fault indicator is a good measure


2499

Fig. 12. Stator current and d9 and d8 wavelet signals under start-up transient for sample D with (a) 0, (b) 1, and (c) 2 broken bars at φe = 45◦ . Fig. 14. Stator current and d9 and d8 wavelet signals under start-up transient for sample E with (a) 0, (b) 1, and (c) 2 broken bars at φe = 45◦ .

of broken bars, and that it is not influenced by axial ducts. The test results of samples D–E shown in Figs. 12–14 and 16 clearly show that the proposed method provides reliable detection of broken bars independent of the axial ducts and that En is a good indicator of rotor fault existence and severity. B. Experimental Results: Field Test (Samples B1, B2, and C1)

Fig. 13. Stator current and d9 and d8 wavelet signals under start-up transient for sample D with (a) 0, (b) 1, and (c) 2 broken bars at φe = 135◦ .

The waveforms of the start-up current and wavelet signals d9 and d8 for samples B1, B2, and C1 are shown in Fig. 15. The time interval used for En calculation is highlighted in Fig. 15, and the calculated values of En are shown in Fig. 16 along with that of samples D and E for comparison. The left sideband f1−2s components measured for samples B1 and C1 were −52.5 and −45.1 dB, which were considerably higher than that of motors of identical design, as shown in Fig. 3(b) and (c). The fbrb components were below −70 dB for sample B2, which is a 20-dB difference. For samples B1 and C1, oscillations cannot be clearly observed in the d8 wavelet signal shown in Fig. 15(a) and (c). The values of En for both samples were very low and similar to that of sample B2 (see Fig. 16), which was diagnosed as a healthy motor with MCSA (f1−2s component below −70 dB). Oscillations were not present in the d8 signal of sample B2, as expected [see Fig. 15(b)]. This indicates that the f1−2s components in samples B1 and C1 are due to the magnetic asymmetry caused by the axial ducts. The results of Fig. 15 show that it is possible to determine that the

2500


V. C ONCLUSION

Fig. 15. Stator current and d9 and d8 wavelet signals under start-up transient for samples B1, C1 (“healthy” motors with false positive MCSA indications), and B2.

False positive rotor fault indications with steady-state spectrum analysis (e.g., MCSA) are a common problem in the field for motors with magnetic asymmetry introduced by the rotor axial ducts. However, there currently is no practical solution for diagnosing rotor faults online for motors with such magnetic asymmetry. In this paper, the analysis of the stator current under the start-up transient is proposed as a solution for diagnosing rotor faults independent of the axial duct influence. Start-up transient analysis is considered since there is no magnetic asymmetry due to limited flux penetration in the rotor yoke at high slip. An experimental study was performed on custombuilt rotors and on 6.6-kV motors in the field that were misdiagnosed with broken rotor bars (MCSA) due to axial ducts. The experimental results show that broken bars can be detected reliably independent of the axial duct influence under the startup transient with the proposed wavelet-based algorithm and energy level-based fault indicator. Start-up transient analysis, which has been applied to this problem for the first time in this work, is expected to help prevent unnecessary inspection and/or loss of production due to false broken bar indications that frequently occur in the field. It also allows the reliable diagnosis of motors with axial ductinduced magnetic asymmetry, which was only possible with offline test methods. The proposed algorithm is currently being applied in the field for screening out false positive indications when the broken bar frequency component of online MCSA exceeds the alarm level. A limitation of start-up transient analysis is the difficulty and cost involved with obtaining and analyzing the start-up data; however, this is usually not an issue in the field considering the high maintenance cost associated with false positive indications. A PPENDIX TABLE I ROTOR A XIAL A IR D UCT D ESIGN AND R ATINGS FOR F IELD AND L ABORATORY T EST M OTORS

Fig. 16. Fault indicator: Normalized energy level En for samples B1, B2, C1, D, and E for zero, one, and two broken bars.

two cases of samples B1 and C1 are false positive indications with the start-up transient analysis. The results presented in Figs. 12–16 show that the proposed DWT-based analysis of start-up transient can detect rotor faults independent of magnetic asymmetry caused by axial ducts. This is because flux cannot penetrate deep into the rotor yoke at high slip due to eddy current rejection, as shown in Fig. 4(b). This verifies that start-up transient analysis can provide reliable assessment of rotor condition for Nd = Np motors and help avoid unnecessary maintenance costs due to false indications.

ACKNOWLEDGMENT The authors would like to thank C. Lim of Hansung Electrical Industrial Company, D. Lee of GS EPS Company, E. Johnson of Progress Energy, Y. Park and J. (Charlie) Nam of GS Caltex, and S. Park of MND Technology for providing the photographs, motor current signature analysis (MCSA) field data, and support with start-up transient testing of high-voltage motors and for sharing their experience with online MCSA testing of induction motors.


R EFERENCES [1] H. A. Toliyat and G. B. Kliman, Handbook of Electric Motors, 2nd ed. New York, NY, USA: Marcel Dekker, 2004. [2] S. Williamson and D. Gersh, “Representation of rotor spiders and axial ventilation ducts in reduced finite-element models for cage rotors,” IEEE Proc.—Elect. Power Appl., vol. 143, no. 6, pp. 423–428, Nov. 1996. [3] A. H. Bonnett and T. Albers, “Squirrel-cage options for ac induction motors,” IEEE Trans. Ind. Appl., vol. 37, no. 4, pp. 1197–1209, Jul./Aug. 2001. [4] A. Bellini et al., “On-field experience with on-line diagnosis of large induction motors cage failures using MCSA,” IEEE Trans. Ind. Appl., vol. 38, no. 4, pp. 1045–1053, Jul./Aug. 2002. [5] W. T. Thomson, “On-line current monitoring—The influence of mechanical loads or a unique rotor design on the diagnosis of broken rotor bars in induction motors,” in Proc. ICEM, 1992, pp. 1236–1240. [6] S. Lee et al., “Evaluation of the influence of rotor axial air duct design on condition monitoring of induction motors,” IEEE Trans. Ind. Appl., vol. 49, no. 5, pp. 2024–2033, Sep./Oct. 2013. [7] I. M. Culbert and W. Rhodes, “Using current signature analysis technology to reliably detect cage winding defects in squirrel-cage induction motors,” IEEE Trans. Ind. Appl., vol. 43, no. 2, pp. 422–428, Mar./Apr. 2007. [8] W. Thomson, “Recent case histories on the application of motor current signature analysis to detect broken rotor bars,” in Proc. IRMC, 2006. [9] B. D. Evans, “Induction motor case histories: A focus on electrically related phenomena,” in Proc. Vibration Inst. Annu. Meet., 2009. [10] “Testing of Squirrel Cage Rotors,” Electrical Apparatus Service Assoc. Inc., St. Louis, MO, USA, EASA Tech. Note 23, 2003. [11] D. Hyun, S. Lee, J. Hong, S. B. Lee, and S. Nandi, “Detection of airgap eccentricity for induction motors using the single-phase rotation test,” IEEE Trans. Energy Convers., vol. 27, no. 3, pp. 689–696, Sep. 2012. [12] H. Douglas, P. Pillay, and A. K. Ziarani, “Broken rotor bar detection in induction machines with transient operating speeds,” IEEE Trans. Energy Convers., vol. 20, no. 1, pp. 135–141, Mar. 2005. [13] M. Riera-Guasp, J. Antonino-Daviu, M. Pineda-Sanchez, J. Perez-Cruz, and R. Puche-Panadero, “A general approach for the transient detection of slip-dependant fault components based on the discrete wavelet transform,” IEEE Trans. Ind. Electron., vol. 55, no. 12, pp. 4167–4180, Dec. 2008. [14] S. H. Kia, H. Henao, and G. A. Capolino, “Diagnosis of broken bar fault in induction machines using discrete wavelet transform without slip estimation,” IEEE Trans. Ind. Appl., vol. 45, no. 4, pp. 1405–1415, Jul./Aug. 2009. [15] F. Briz, M. W. Degner, P. Garcia, and D. Bragado, “Broken rotor bar detection in line-fed induction machines using complex wavelet analysis of startup transients,” IEEE Trans. Ind. Appl., vol. 44, no. 3, pp. 760–768, May/Jun. 2008. [16] Y. Gritli et al., “Advanced diagnosis of broken bar fault in induction machines by using discrete wavelet transform under time-varying condition,” in Proc. IEMDC, May 2011, pp. 424–429. [17] J. Antonino-Daviu et al., “Detection of broken outer-cage bars for doublecage induction motors under the startup transient,” IEEE Trans. Ind. Appl., vol. 48, no. 5, pp. 1539–1548, Sep./Oct. 2012. [18] C. S. Burrus, R. A. Gopinath, and H. Guo, Introduction to Wavelets and Wavelet Transforms—A Primer. Englewood Cliffs, NJ, USA: PrenticeHall, 1998. [19] Z. Zhang, Z. Ren, and W. Huang, “A novel detection method of motor broken rotor bars based on wavelet ridge,” IEEE Trans. Energy Convers., vol. 18, no. 3, pp. 417–423, Sep. 2003. [20] J. Cusido, L. Romeral, J. A. Ortega, J. A. Rosero, and A. Garcia Espinosa, “Fault detection in induction machines using power spectral density in wavelet decomposition,” IEEE Trans. Ind. Electron., vol. 55, no. 2, pp. 633–643, Feb. 2008. [21] I. P. Tsoumas, G. Georgoulas, E. D. Mitronikas, and A. N. Safacas, “Asynchronous machine rotor fault diagnosis technique using complex wavelets,” IEEE Trans. Energy Convers., vol. 23, no. 2, pp. 444–459, Jun. 2008. [22] M. Blodt, D. Bonacci, J. Regnier, M. Chabert, and J. Faucher, “On-line monitoring of mechanical faults in variable-speed induction motor drives using the Wigner distribution,” IEEE Trans. Ind. Electron., vol. 55, no. 2, pp. 522–533, Feb. 2008. [23] M. Pineda-Sanchez et al., “Instantaneous frequency of the left sideband harmonic during the start-up transient: A new method for diagnosis of broken bars,” IEEE Trans. Ind. Electron., vol. 56, no. 11, pp. 4557–4570, Nov. 2009. [24] Z. K. Peng, P. W. Tsey, and F. L. Chu, “A comparison study of improved Hilbert-Huang transform and wavelet transform: Application to fault diagnosis for rolling bearing,” Mech. Syst. Signal Process., vol. 19, no. 5, pp. 974–988, Sep. 2005.

2501

[25] J. Antonino-Daviu, M. Riera-Guasp, J. Roger-Folch, F. Martínez-Giménez, and A. Peris, “Application and optimization of the discrete wavelet transform for the detection of broken rotor bars in induction machines,” Appl. Comput. Harmonic Anal., vol. 21, no. 2, pp. 268–279, Sep. 2006.

Chanseung Yang (S’12) received the B.S. degree in electrical engineering from Korea University, Seoul, Korea, in 2012, where he is currently working toward the Ph.D. degree. In 2012, he was an Intern with the Universitat Politècnica de València, Valencia, Spain. His research interests are in condition monitoring and analysis of electric machinery.

Tae-June Kang (S’11) received the B.S. degree in electrical engineering from Korea University, Seoul, Korea, in 2011, where he is currently working toward the Ph.D. degree in electrical engineering. His research interests are in stator winding insulation testing and condition monitoring and diagnostics of electric machinery.

Doosoo Hyun (S’09) received the B.S. and M.S. degrees in electrical engineering from Korea University, Seoul, Korea, in 2009 and 2011, respectively, where he is currently working toward the Ph.D. degree. In 2009, he worked at the SKF Condition Monitoring Center, Fort Collins, CO, USA, on the development of condition monitoring tools for electric machines, as a summer intern. In 2011, he was an intern at the Austrian Institute of Technology, Vienna, Austria, where he worked on condition monitoring of permanent-magnet synchronous machines. His research interests include condition monitoring, diagnostics, and analysis of electric machinery. Mr. Hyun was the recipient of the 2013 SDEMPED Prize Paper Award from the Technical Committee on Diagnostics of the IEEE Power Electronics Society.

Sang Bin Lee (S’95–M’01–SM’07) received the B.S. and M.S. degrees in electrical engineering from Korea University, Seoul, Korea, in 1995 and 1997, respectively, and the Ph.D. degree in electrical engineering from the Georgia Institute of Technology, Atlanta, GA, USA, in 2001. From 2001 to 2004, he was with the General Electric General Electric (GE) Global Research Center (GRC), Schenectady, NY, USA. At GE GRC, he developed an interlaminar core fault detector for generator stator cores and worked on insulation quality assessment for electric machines. From 2010 to 2011, he was with the Austrian Institute of Technology, Vienna, Austria, as a Research Scientist where he worked on condition monitoring of PM synchronous machines. Since 2004, he has been a Professor of Electrical Engineering at Korea University. His current research interests are in protection, monitoring and diagnostics, and analysis of electric machines and drives. Dr. Lee was the recipient of nine prize paper awards from the IEEE Power Engineering Society, the Electric Machines Committee of the IEEE Industry Applications Society, and the Technical Committee on Diagnostics of the IEEE Power Electronics Society. He serves as a Distinguished Lecturer (2014–2015) for IEEE Industry Applications Society (IAS) and an Associate Editor for the IEEE T RANSACTIONS ON I NDUSTRY A PPLICATIONS for the IEEE IAS Electric Machines Committee.

2502


Jose A. Antonino-Daviu (S’04–M’08–SM’12) received the M.Sc. and Ph.D. degrees in electrical engineering from the Universitat Politècnica de València, Valencia, Spain, in 2000 and 2006, respectively. He worked in the private sector, being involved in several international projects. He is currently an Associate Professor in the Faculty of Industrial Engineering, Universitat Politècnica de València, where he develops his docent and research work. He was an Invited Professor at the Helsinki University of Technology, Helsinki, Finland, in 2005 and 2007, and at Michigan State University, East Lansing, MI, USA, in 2010. His primary research interests include condition monitoring of electric machines, wavelet theory and its application to fault diagnosis, and design and optimization of electrical installations and systems.

Joan Pons-Llinares (M’13) received the M.Sc. degree in industrial engineering and the Ph.D. degree in electrical engineering from the Universitat Politècnica de València (UPV), València, Spain, in 2007 and 2013, respectively. He is currently an Assistant Professor in the Electric Engineering Department, UPV. His research interests include time–frequency transforms, condition monitoring, and diagnostics of electrical machines.