See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/303248506 Transformer failure diagnosis based on parameter analysis Conference Paper · November 2012 CITATION READS 1 378 6 authors, including: D.A.K. Pham H. Borsi Ho Chi Minh City University of Technology (HCMUT), Vietnam National University … Leibniz Universität Hannover 19 PUBLICATIONS 96 CITATIONS 208 PUBLICATIONS 2,737 CITATIONS SEE PROFILE Ernst Gockenbach Leibniz Universität Hannover 269 PUBLICATIONS 3,309 CITATIONS SEE PROFILE Some of the authors of this publication are also working on these related projects: power transformer condition evaluation View project Experimental Project (Wind Energy) View project All content following this page was uploaded by Ernst Gockenbach on 29 May 2016. The user has requested enhancement of the downloaded file. SEE PROFILE Diagnostik Elektrischer Betriebsmittel ∙ 15. – 16.11.2012 in Fulda Transformer failure diagnosis based on parameter analysis D. A. K. Pham1, 2 (*), T. M. T. Pham1, 2, M. H. Safari1, A. Kazemi1, H. Borsi1, E. Gockenbach1 1 Leibniz Universität Hannover, Schering-Institut, Hannover, Germany 2 Ho Chi Minh City University of Technology, Vietnam E-mail (*): pham@si.uni-hannover.de Abstract There are currently several methods applied to diagnose failures on the active part of power transformers, which can be categorized into two main groups: traditional- and advanced methods. Traditional methods include measurements of transformer ratio, exciting current, magnetic balance, winding resistance, short-circuit impedance etc. Advanced methods consist of tests of Frequency Response Analysis (FRA), Frequency Response of Stray Losses (FRSL), Frequency Response of Leakage Inductance (FRLI) etc. Results derived from these methods reflect transformer parameters in open- and short-circuit configurations. In general, these methods should be performed and combined to enhance the accuracy of failure diagnosis for power transformers. In order to reduce the number of tests while keeping the diagnostic quality, this paper introduces a study case in which a new method based on only the driving-point impedance test analysis is applied to determine transformer parameters from which results of almost above-mentioned tests are recovered appropriately. In addition, the tendency of parameter change in regard with different failures, which may be valuable for diagnosis quality improvement, is also researched. In the study case, experimental measurements are performed on a 200 kVA 10.4/0.4 kV YNyn6 opened distribution transformer which is simulated under healthy- and faulty conditions. 1 Introduction To improve the diagnosis quality for transformer failures, the combination of traditional- and advanced diagnostic techniques is required. That is understandable since the transformer is a complex object and each technique reflects only certain response from all transformer components at fixed frequencies. However, there is always a demand to develop more techniques for better diagnosis as well as to improve the interpretation of all measured results. In transformer structure, the most important part is the active part which includes mainly windings and the core. Failure modes on the transformer active part consist of electrical, thermal, mechanical and degradation/aging [1]. Among these failure modes, electrical failure is of importance since it reveals first signals of abnormal conditions of transformers; and in case there is no appropriate “treatment”, other failures can appear. In order to fulfill the demand in developing a new technique for better diagnosis and comprehensive interpretation of electrical failures on the active part of power transformers, this paper proposes a new method based on only driving-point impedance test analysis to determine physical transformer parameters. Once these parameters are reasonable derived, the interpretation of results of other diagnostic methods such as magnetic balance tests, FRLI, FRA etc can be reached. Furthermore, several physical parameters might be helpful for failure diagnosis since they are good indicators for detecting failure modes. The test object of this paper is a 200 kVA 10.4/0.4 kV YNyn6 opened distribution transformer which is changed ISBN 978-3-8007-3465-8 from Yzn5 vector group. The transformer has no tank and also no insulating liquid. Electrical failures simulated on the transformer active part consist of open-circuit, shortcircuit between discs, short-to-ground (HV windings) and loss of core ground. Driving-point impedance tests are performed by mean of the vector-network analyzer (VNA) named “FRAnalyzer” of Omicron. Different equivalent impedances of the test transformer are measured under different combinations of terminal conditions such as being excited by the instrument, opened (left floating), shorted together or grounded. For all connections, coaxial cables are used to avoid interferences and additional external coupling. Frequency range for all tests is from 20 Hz to 2 MHz. Open-circuit voltage of the instrument is 1 V. Receiver bandwidth is chosen as 30 Hz for better denoising effect. 2 Transformer parameter determination based on driving-point impedance analysis As mentioned in the introduction section, there are two kinds of conditions investigated on the test transformer: healthy and faulty. For healthy condition, reference [2] presents an equivalent transformer circuit and a simple procedure for determining circuit parameters for the purpose of frequency response analysis and hence this procedure is briefly recalled and improved in section 2.1. The parameters in the equivalent circuit are then called as physical transformer parameters since they represent physical structure parts of the transformer. Afterwards, © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach Diagnostik Elektrischer Betriebsmittel ∙ 15. – 16.11.2012 in Fulda section 2.2 illustrates verifications of these parameters through comparisons of frequency responses which are measured through other diagnostic methods and recovered from the physical equivalent circuit. Finally, in section 2.3, attempts in detecting simulated failures based on physical parameter analysis are presented. 2.1 Parameter analysis in healthy condition 2.1.1 Equivalent transformer circuit An appropriate equivalent circuit in which most physical parameters of transformers appears logically is the first important thing to deal with. In [2] the duality-based equivalent circuit adapted from the one developed for the purpose of transient analysis seems to be the most appropriate circuit. Figure 1 presents this circuit for the test transformer in a context of a driving-point impedance test performed on the winding of phase A at HV side. In this test configuration, the VNA source is connected to the HV terminal of phase A while the HV neutral is grounded. Then the measured impedance is the equivalent impedance of the transformer observed at the excited terminal. vant LV winding. The inter-winding capacitances between HV windings such as CAB and CBC have insignificant influences but make the calculation more complicated; thus they can be neglected. 2.1.2 Parameter determination from short-circuit tests at HV side In a normal short-circuit test at HV side, each singlephase winding at LV side, e.g. an, is shorted whereas the relevant HV winding, i.e. AN, is excited by the VNA instrument (HV neutral is grounded). The measured impedance consists of equivalent resistance from stray losses (Rsl) in real part and total leakage reactance (L3total from L3H and L3X) in imaginary part. With appropriate factors, resistance and equivalent leakage inductance of HV- and LV windings (RH/RX and L3H/L3x) are reasonably derived [3]. Afterwards, frequency-dependent functions RH(f), RX(f) and L3H(f), L3X(f) for each phase are developed based on measured values. They are valid from 20 Hz to at least 30 kHz since the inductive feature of the measured impedance can be observed within this frequency range. Figure 3 shows a verification of measured- and recovered impedances from the HV winding of phase A in whole frequency range, i.e. from 20 Hz to 2 MHz. L4 VNA A RA 50 RH L3H R4 R1 CA B N RB RH L3H R4 CB L4 RX a L3X RX b L1 Ly NH:NH L3X NH:NX Ry R1 L1 n Ly NH:NH C RC RH L3H R4 L4 Ry NH:NH L3X C1 R1 CC NH:NX RX c L1 N1 NH:NX Figure 1 Simple duality-based equivalent transformer circuit of a YNyn transformer seen from HV side in lowand mid frequency ranges [2] In Figure 1: - Z1 = R1//L1 : non-linear core leg impedance - Zy = Ry//Ly : non-linear core yoke impedance - RH and RX : resistances of HV- and LV windings - L3H and L3X : leakage inductances at HV- and LV sides - Z4 = R4//L4 : zero-sequence impedance - ZA = RA//CA : total capacitive impedance of phase A - ZB = RB//CB : total capacitive impedance of phase B - ZC = RC//CC : total capacitive impedance of phase C Due to the fact that the HV neutral is grounded in most measurement configurations, total capacitive impedance of a HV winding consists of different capacitances such as series-, ground capacitances of this winding and the combination of inter-winding capacitance between HV and LV sides, series- and ground capacitances of the rele- ISBN 978-3-8007-3465-8 Figure 3 Verification of equivalent impedances in the short-circuit test on phase A between measurement and recovery at low- and mid frequency ranges. 2.1.3 Parameter determination from zero-sequence test at HV side In the zero-sequence test at HV side, three HV terminals are all excited in the same time by the instrument while the HV neutral is grounded (LV windings are opened). At low frequencies, the total zero-sequence current from the instrument is divided into three parts each of which flows through RH and L3H of each phase winding and Z4 (if there are three equal Z4 in the core circuit) [4]. There is © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach Diagnostik Elektrischer Betriebsmittel ∙ 15. – 16.11.2012 in Fulda almost no current flowing through core impedances Z1 and Zy since magnitude of Z4 is much lower than those of Z1 and Zy (see Figure 1). From that, Z4 of each phase is determined thanks to available RH and L3H from shortcircuit tests. Afterwards, like in the short-circuit test, frequency-dependent functions R4(f) and L4(f) for HV side are developed appropriately based on calculated values. Figure 4 presents a comparison of measured- and recovered equivalent impedances in the zero-sequence test. It can be concluded that all frequency-dependent functions developed from short-circuit and zero-sequence tests are reasonable derived within frequency range from 20 Hz to at least 3 kHz. C and HV neutral are grounded and other winding terminals are left opened. If one assigns conditions of winding terminals under order “HV_LV” as “ABCN_abcn” for a short and convenient configuration name, one would have the name of this configuration as “EOGG_OOOO” (E– excited, O–opened, G–grounded). From Figure 1, when the HV terminal of phase C is shorted to ground, core legand yoke impedances in this phase are bypassed. Hence, the equivalent impedance observed from the excited phase winding includes only core leg- and yoke impedances of phases A and B. Therefore, if one performs three measurement configurations mentioned in Table 1 [2], one would have a mathematic problem with two variables and three equations. The best solution for this problem is a solver of nonlinear least squares problem in which objective functions are differences of real- and imaginary parts between measured- and calculated equivalent impedances at each frequency point [5]. Index 1 2 3 Configuration EOOG_OOOO EGOG_OOOO EOGG_OOOO Equivalent impedance (Z1 // Zy + Z1) // (Z1 // Zy) (Z1 // Zy) // (Z1 // Zy) Z1 // (Z1 // Zy) Table 1 Three measurement configurations for phase A at low frequencies [2] Figure 4 Verification of equivalent impedances in the open-circuited zero-sequence test on HV side between measurement and recovery at low- and mid frequency ranges. 2.1.4 Parameter determination from open-circuit tests at HV side 2.1.4.1 In low frequency range (20 Hz to 400 Hz) In this frequency range, core effect is dominant. Thus, core impedances (Z1 and Zy) are main components in the equivalent circuit in Figure 1. Capacitive impedances can be neglected since they are much larger than core impedances at low frequencies. According to [2], in case only Z1 and Zy appear in the circuit for an approximation calculation (RH, L3H and Z4 are bypassed since they have only little influence), there are three different configurations required to determine core impedances when a HV winding is excited. The three configurations are necessary since they provide different core circuits in determining variables (Z1 and Zy). Let illustrates a measurement configuration in which the HV terminal of phase A is excited, HV terminal of phase ISBN 978-3-8007-3465-8 Core impedances (Z1 and Zy) as results after the solver are relatively reasonably determined. However, an improvement can be reached if other parameters (RH, L3H, Z4) and more configurations from other phases are taken into account. Our experience shows that when all measurement configurations from three phases are supplied as inputs to the solver, core impedances are best derived and proportionate to each other. Afterwards, components from Z1 and Zy, i.e. R1(f), L1(f), Ry(f) and Ly(f), are developed in whole frequency range in following procedure. At low frequencies, due to the fact that calculated values vary in an exponential tendency, they are best fitted by normal exponential functions. At higher frequencies, since core impedances are no longer dominant, experimental formulae can be exploited instead. From literatures, there are three main effects contribute to the core losses: eddy current, hysteresis and anomalous [6]. However, when a transformer is excited under very low applied voltage, eddy current (skin) effect in the core is dominant [7] and therefore, skin-effectbased core-section resistance and inductance can be determined thanks to estimated dimensions and material of core section of the test transformer [8]. 2.1.4.2 In mid frequency range (5 kHz to 6.5 kHz) This frequency range is chosen so as all measured impedances in configurations mentioned in sub-section 2.1.4.1 are capacitive. Therefore, equivalent capacitive impedances (ZA, ZB and ZC) have most influence and can be calculated by mean of a solver of nonlinear least squares problem [5]. The procedure of calculation is similar like © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach Diagnostik Elektrischer Betriebsmittel ∙ 15. – 16.11.2012 in Fulda that for low frequency range, except equivalent impedance equations are different since all components are involved in the circuit [2]. For illustration of an example of the difference, let have a look on the equivalent impedance between C1 and N1 in Figure 1. In low frequency range it is (ZC+RH+jωL3H)//Z1~Z1 because ZC>>Z1; but in mid frequency range it will be: (ZC+RH+jωL3H)//Z1~ZC since ZC>>RH+jωL3H and ZC<<Z1 One important thing should be noted that these capacitive impedances (ZA, ZB and ZC) do not belong to a certain kind of capacitance such as series- or ground capacitances, but include different kinds of capacitances as abovementioned. Therefore, in order to separate them, capacitive impedance measurements are required. ISBN 978-3-8007-3465-8 2.2 Diagnostic test recovery via calculated physical parameters in healthy condition Once physical parameters are available, several diagnostic tests such as magnetic balance, FRLI, “standard” FRA etc can be recovered appropriately. Following figures depict verifications of these tests whose results are measured by means of different measurement testing devices and from the physical parameter recovery. Figure 6 compares test results of magnetic balance tests between a terminal measurement and from the equivalent circuit recovery at 50 Hz. The terminal test is conducted by mean of a testing set including a sinusoidal generator and a 12-bit oscilloscope to measure voltages on winding terminals at HV side. The applied voltage in the terminal test is the same with that of the impedance test (1 V). Results confirm that core impedances are well determined since this test reflects the core behaviour at low frequencies. 100 80 VAN 60 40 VBN VCN Phase A excited Phase B excited Recovered Measured Recovered Measured Recovered 20 0 Measured 2.1.6 Parameter verification A simple way to verify the parameters determined in previous sub-sections is to compare measured- and recovered equivalent impedances. To recover equivalent impedances for whole range of frequency, frequency-dependent functions of inductive components as well as values of capacitances in Figure 1 are applied in a circuit simulation software. For a better recovery of frequency responses of equivalent impedances, CA, CB and CC should be replaced by series-, ground and inter-winding capacitances thanks to CHG, CLG and CHL. Figure 5 depicts a verification of a frequency response of equivalent impedance in the open-circuit test on phase A. Observed good agreement in low- and mid frequency ranges proves that parameters are reasonable derived. Besides, deviations of magnitude and phase angle at high frequencies appear due to the fact that the equivalent circuit is the lumped circuit which does not reflect fully the distributed behaviour of the real transformer. Figure 5 Verification of the open-circuit test on phase A Voltage in percentage (%) 2.1.5 Parameter determination from capacitive impedance tests There exists a diagnostic test for measuring capacitances of power transformers performed by mean of an universal testing device such as the CPC100 of Omicron under high-voltage applied at low frequencies [9]. Measured results for a two winding transformer consist of HV-toground, LV-to-ground and HV-LV capacitances (CHG, CLG and CHL respectively). Under assumption that there is a balanced condition between each phase, corresponding capacitances of each phase, e.g. CgA, Cga and CAa, can be calculated as one third of the measured ones. On the other hand, our experience confirms that those capacitances can be independently determined from drivingpoint impedance tests by mean of the VNA under 1 V excited. Of course those impedance tests should have the same configurations with the ones from diagnostic test and thus they are called capacitive impedance tests. Based on the capacitive tendency of measured impedances at high frequencies, capacitances CHG, CLG and CHL are reasonably determined. Phase C excited Figure 6 Verification of magnetic balance tests on HV side © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach Diagnostik Elektrischer Betriebsmittel ∙ 15. – 16.11.2012 in Fulda Figure 7 presents a good verification of FRLI from a short-circuit test performed by mean of the CPC100 under 1 A current source supplied and the corresponding one recovered from the calculated physical parameters in frequency range from 20 Hz to 400 Hz. and reliable information. At high frequency range, interpretation of diagnostic tests requires a more complicated equivalent circuit with more parameters. 2.3 Parameter analysis in failure condition There are in total ten failure cases in each of which a single failure is simulated on the active part of the test transformer. Failure modes include open-circuit, short-circuit between discs, short-to-ground (each failure is simulated on each HV winding of three phase in succession) and loss of core ground. Positions of failures on HV windings are at the middle of the windings for first investigations. The parameter determination procedure in section 2.1 is applied for each of all cases appropriately. However, there is a fact that only a few parameters could be recognized and determined. Following sub-sections explain more details. Figure 7 Verification of FRLI from short-circuit test of phase A at HV side Figure 8 shows good agreement at low- and mid frequency ranges (until 20 kHz) for the end-to-end opencircuit (EEOC) FRA test between the “standard” FRA measurement performed by mean of the FRAnalyzer instrument and the recovery result from the physical parameters. 2.3.1 Open-circuit failure on HV windings From the first step of the parameter determination procedure, i.e. short-circuit test analysis, negative leakage inductances are derived for this failure condition. It is logical since capacitive currents flow in the faulty winding through the open-circuit position. Therefore, other next steps are not possible because of negative leakage inductance value and this value is useful to detect an opencircuit failure on a winding. 2.3.2 Short-circuit-to-ground and short-circuitbetween-discs failures on HV windings From the calculation procedure, it is concluded that one of physical parameters which are helpful to detect the failures, i.e. being different from those in healthy condition and having clear tendency, is leakage inductance. It can be explained that the failures reduce energy in the leakage channel and hence the leakage inductances decrease [10]. Figure 9 confirms a decrease of energy from 2D-FEM simulations for the short-circuit tests in different conditions. Figure 8 Verification of the EEOC-FRA test on phase A A short conclusion can be drawn: The physical parameters and the equivalent circuit are reasonably determined for interpretation of several diagnostic test results at lowand mid frequency ranges. These physical parameters can be therefore investigated under different failure conditions for a better diagnosis since they may supply important ISBN 978-3-8007-3465-8 Figure 9 Energy distributions on the gap between HVand LV windings of phase A in healthy-, shorted-disc and short-to-ground conditions at 50 Hz (from left to right respectively) © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach Diagnostik Elektrischer Betriebsmittel ∙ 15. – 16.11.2012 in Fulda Table 2 introduces a comparison of leakage inductances between the equivalent circuit recovery and from FEM simulations at 50 Hz. A clear tendency of value change between all conditions is observed from FEM-based calculation whereas from the recovery, the shorted-disc- and short-to-ground failures can not be recognized exactly; in fact they are only distinguished from the healthy condition and in such case, ground capacitance can be a good indicator to discriminate them. However, a clear tendency of leakage inductance change from the recovery like that from the FEM calculation appear at higher frequencies, for example at 10 kHz. It recommends that the impedance measurement performance of the VNA instrument at low frequencies should be further investigated for better analysis. L3total (mH) Condition healthy shorted-disc short-to-ground based on FEM calculation 49 41.6 34.4 2.3.3 Ungrounded core A parameter which is reliable for detecting this failure condition is the ground capacitance of the LV windings since the LV windings are inner windings and core ground is lost. Table 3 shows a comparison of measured capacitances between healthy- and failure conditions. As predicted, total ground capacitance of the LV windings (CLG) changes significantly whereas there is little change with inter-winding capacitance (CHL) and ground capacitance of the HV windings (CHG). As a result, it can be the best indicator to detect this failure mode. Condition healthy ungrounded core CHL CHG CLG 1257 1386 343 264 2843 264 Table 3 Approximate capacitances from capacitive impedance tests 3 4 recovered from physical parameters 54.6 45.6 46.5 Table 2 Approximate total leakage inductances at 50 Hz Capacitance (pF) Conclusions The interpretation of several diagnostic tests until mid frequency range is now possible thanks to the physical equivalent circuit with physical parameters. Number of diagnostic tests may be reduced since they can be recovered through impedance test analysis. On the other hand, the impedance test analysis can be considered as a new diagnostic technique since it supplies more information to the diagnosis for more reliable interpretation. A few physical parameters can be good indicators to detect electrical failures on the active part of power transformers. Besides, they are also useful in diagnosis of mechanical failures which are not investigated in this paper. References [1] Velasquez, J. L.: Intelligent monitoring and diagnosis of power transformers in the context of an assesst management model. PhD thesis, Polytechnic university of Catalonia: Spain, 2011 [2] Pham, D. A. K. et. al.: Duality-based lumped transformer equivalent circuit at low frequencies under single-phase excitation. International conference on high-voltage engineering and application (accepted), 2012 [3] Martinez, J. A.; Walling, R.; Mork, B. A.; MartinArnedo, J. and Durbak, D.: Parameter Determination for Modeling System Transients—Part III: Transformers. IEEE Transactions on Power Delivery. Vol. 20, No. 3, July 2005, pp. 2051-2062 [4] Mork, B. A. et. al.: Hybrid transformer model for transient simulation – Part II: Laboratory measurements and benchmarking. IEEE Transaction on Power Delivery. Vol. 22, No. 1, January 2007, pp. 256-262 [5] Mathwork Inc.: Curve Fitting Toolbox [6] ABB: Transformer Handbook. 3th Edition, 2007 [7] Abeywickrama, N.: Effect of dielectric and magnetic material characteristics on frequency response of power transformers. Doctoral thesis, Chalmers university of technology: Sweden, 2007 [8] Lammeraner, J.; Stafl, M.: Eddy current. Iliffe Books, London: U.K. 1996 [9] Krüger, M.: Capacitance and dissipation factor measurement with CPC100 + CP TD1. Omicron Application Guide, 2004 [10] Kulkarni, S. V.; Khaparde S. A.: Transformer engineering design and practice. Marcel Dekker, 2004 The paper introduces a new and comprehensive procedure in determining physical parameters for power transformers based on only the driving-point impedance tests for the purposes of test result interpretation and electrical failure diagnosis. Several main contributions can be drawn as follows: ISBN 978-3-8007-3465-8 View publication stats © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach