Scaling procedures for starting current measurements IEEE INDUSTRY APPLICATIONS MAGAZINE NOV j DEC 2010 WWW.IEEE.ORG/IAS BY WAQAS M ARSHAD, SAMI KANERVA, SILVIA BONO, MASSIMO MENESCARDI, & HOLGER PERSSON © CREATAS N THIS ARTICLE, THE ACCURACY machines This is performed to ascertain the confidence level levels of different scaling procedures for of the predicted full-voltage starting currents from the starting current measurements are addressed reduced-voltage factory measurements, a necessity for motors through the statistical analysis of locked- with a strict tolerance to the starting current level It is rotor tests involving hundreds of medium–large induction shown that, for traditional scaling methods employing only I measurements, this accuracy level is 88–90% for scaling step 28 Digital Object Identifier 10.1109/MIAS.2010.938385 from 0.6 to 1.0 p.u voltage This figure could be raised to 1077-2618/10/$26.00©2010 IEEE 95–97% through a novel method that also uses finite element method (FEM) simulations for each individual test voltage The differences between FEM and voltage measurements are documented and used for correcting the full-voltage FEM simulation Further, the use of FEM is shown to help in the understanding of the causes and locations of measured voltage-dependent saturation uncertainties The region (end-core, overhang, or main core) that contributes the most to saturation uncertainties is shown to be identifiable through origin/parametric dependencies, leading to a better product understanding and more reliable scaling methods in future Starting Currents in Induction Machines THE DIFFERENCES BETWEEN FEM AND VOLTAGE MEASUREMENTS ARE DOCUMENTED AND USED FOR CORRECTING THE FULL-VOLTAGE FEM SIMULATION The Trend for Low-Inrush Current Motors Recently, the demand for low-inrush current motors has been slowly increasing because the starting current is typically less than p.u., and no tolerance margin is accepted between the specification and delivered motor (referred below as zerotolerance designs) Marine, offshore petrochemical, and, in general, all industry connected to a weak grid are increasingly demanding such direct online solutions Costs, weight/ space requirements, and complexity are some of the factors that rule out the use of alternative solutions, such as softstarters, transformer-starting, and converter-start [6] The zero tolerances impose stringent demands upon exact knowledge of all steps of the motor manufacturing process Factors Affecting Starting Currents The inductance of a machine and its variation during startup roughly define the starting current for a given applied voltage The inductance is generally divided into three distinct regions: subtransient, transient, and steady state The inductances are influenced by a number of motor geometrical details as well as materials The starting currents Design Parameters Accuracy Since all the manufactured motors need to be tested before delivery, in addition to the design effort, the emphasis is on (especially for zero-tolerance designs): n design parameters accuracy n measurements accuracy n scaling accuracy The accuracy level of the design parameters primarily depends upon the production site history that has been achieved through years of experience of machine modeling (reluctance networks, equivalent circuits, and FEM analysis), statistically calibrated against measurements on prototypes and delivered products Skin effect, iron saturation, and inductances modeling are some of the key parameters that define the true starting performance of an induction motor Locked-Rotor and Run-Up Measurements The starting current measurement is carried out by either locked-rotor or run-up methods The accuracy of the measurement is influenced by the test engineers’ experience and care, rotor position, and temperatures in the stator and rotor In the locked-rotor measurement, the current is measured at different supply voltage levels, and, based on that, the rated starting current is scaled to the rated voltage IEC 60034-1 [2] defines the locked-rotor current as the greatest steadystate, root-mean-square current taken from the supply with the motor held at rest, over all angular positions of its rotor, at rated voltage and frequency According to IEEE 112-2004 [15], for this test, when possible, the measurements shall be taken at rated voltage and frequency since the current is not directly proportional to the voltage because of changes in reactance caused by saturation of the leakage paths In the run-up measurement, the motor is started by reduced voltage, and the measured starting current is scaled to the rated voltage Usually, the motor is rotating slowly in a backward direction before startup (reverse-rotation start) to obtain the current at zero speed by the time the supplied electromagnetic transients have decayed IEEE INDUSTRY APPLICATIONS MAGAZINE NOV j DEC 2010 WWW.IEEE.ORG/IAS Starting current levels for a direct-online, single-cage induction motor are typically specified to be four to seven times the rated motor currents (4–7 p.u.) [1] These figures are for medium–large induction motors, whereas these values can be as high as 10–12 p.u for smaller motors Starting currents of the delivered motors must usually follow one of the following: n the International Electrotechnical Commission (IEC) 60034-1 [2], allowing a 20% tolerance margin n the National Electrical Manufacturers Association (NEMA) MG1 standard [3], specifying tolerances according to kilovoltampere per horsepower n the customer specifications, defining individual tolerance or absolute figures, e.g., in chemical oil and gas sector: n the Shell Design and Engineering Practice (DEP) 33.66.05.31 [4] or American Petroleum Institute (API) 541 standard [5] that requires a starting current without tolerances, e.g., below 6.5 p.u for medium-voltage motors during the design stage can thus be influenced [6]–[12] by the following geometrical and physical parameters: n Stator design: number of turns, end-windings layout, core length, and slot geometry n Rotor design: slot dimensions, bar profile, and material n Frame design: end-ring shielding, etc The trade-offs between other performance parameters such as starting and breakdown torque, rated efficiency, and power factor also need to be considered when limiting the inrush current during the design stage The mechanical and thermal limits of the structure also need to be taken into account [10]–[14] Scaling When the full-voltage test is not possible on the factory test floors (e.g., for the largest induction machines) or when 29 the maximum allowable stall-time limits full-voltage tests, the reduced-voltage test is applied In this case, the starting current is extrapolated from the reduced-voltage measurements, a method commonly known as voltage scaling In such cases, not only the measurements themselves but also the accuracy of scaling methods (full-voltage currents prediction from reduced-voltage measurements) attains vital importance For run-up tests, the same scaling methods may be used as in the locked-rotor test Accuracy of Scaled Measurement Results As will be discussed later, the traditional scaling methods, including those presented in [15] and [16], have a degree of uncertainty Sometimes, factory acceptance tests (FATs) provide scaled starting current levels that are not a true representation of the motors’ actual starting behavior Documented experience from many years reveals that even though there may be occasional complaints for scaled results predicted by FATs, seldom have such complaints been raised during commissioning or the normal operation of the 30 Scaling (1–15%) Measurements (1–3%) Current IEEE INDUSTRY APPLICATIONS MAGAZINE NOV j DEC 2010 WWW.IEEE.ORG/IAS Starting Current Uncertainty Band Starting Currents Uncertainty Design (1–2)% Slip Different accuracy bands for the full-voltage starting currents The arrows show distinct regions corresponding to the design and measurements uncertainty motors This indicates that scaling would only cause the differences between measured results and design values Figure shows the authors’ perception regarding uncertainties associated with the starting currents A careful benchmarking procedure by the authors on a few hundred machines has singled out scaling as the biggest factor associated with the full-voltage starting currents accurate prediction Outline and Scope of the Study This section presents the details of the work conducted by the authors regarding scaling The study covers a vast range of direct-online induction motors (taken from the list in Table 1) Five to eight locked-rotor measurement results for each individual machine, all at different voltage levels, are used The following are the topics addressed: n accuracy evaluation of the traditional scaling n presentation of the novel combined FEM-measurements scaling method n confidence-level evaluation of the aforementioned method based on statistical analysis n insight into starting currents saturation physics by studying parametric dependencies Issues such as repeatability, reliability, and test-conditions influence (e.g., ambient temperature and hot or cold motor) are addressed This is done through new measurements on 50 motors and the utilization of earlier documented measurement results of a few hundreds of randomly selected motors This vast amount of measured data allows the separation of measurement uncertainties from those of scaling procedures In this study it is revealed that no significant differences exist between the measured values of locked current and run-up starting currents Hence, the analysis in the following sections is restricted to locked-rotor tests only Since the present study deals with voltages at machine terminals, the role the network impedance plays is not significant and is omitted from the study The FEM approach is introduced as an alternative to traditional scaling methods that employ only measurements Thus, two new scaling methods that, in addition to measurements, employ FEM simulations are studied Since the FEM models use the true hysteresis (BH) iron curve for TABLE MACHINE DATA Output power 140–18,000 kW at 50 Hz, 200–24,100 HP at 60 Hz Frame size IEC 400 to 1120, NEMA 17 to 44.1 Number of poles 2–24 Voltages 380–15,000 V Frequency 50 or 60 Hz Environment IP23, IP54, IP55, IPW24, IC01, IC611, IC81W, IC 411, IC416, IC511, and IC516 Enclosure material Fabricated steel or cast iron Motor type AMA, AMI, AMC, AMD, and HXR Mounting type Horizontal or vertical Standards IEC, NEMA, country, and customer-specific standards Protection For zone or 2, temperature class T3/T4 Data might differ according to machine type modeling saturation, at least theoretically, it was believed that the FEM should be able to provide a better picture of the saturation at the full-voltage conditions in the machine than through one of the traditional empirical extrapolation methods that are based only on measurements Unfortunately, experience showed that even FEM simulations required some sort of calibration/tuning to achieve the acceptable confidence level in absolute terms Two such calibration approaches are presented and discussed Traditional Scaling Methods Inom ¼ Imeas ksat (Unom =Umeas ), (5) ksat ¼ (Imeas2 =Umeas2 )=(Imeas1 =Umeas1 ): (6) 6) An improved saturation factor approach uses a voltage dependence derived from the simple magnetic circuit approach NI ¼ (Rairgap þ Riron )U, Description of the Methods Unom Umeasured k , (1) where Inom is the nominal current, Unom is the nominal (rated) voltage, Imeasured is the measured current, and Umeasured is the measured voltage 2) A more exact method (log–log) requires two measurement points, and the exponent k in (1) is calculated according to (2), IEEE 112-2004 [15] k ¼ log(Imeas1 =Imeas2 )= log(Emeas1 =Emeas2 ), (2) where E is the electromotive force that is sometimes approximated with the supply voltage U 3) IEEE 112-2004 [15] also states that, for better accuracy, a least-squares fit should be used when determining the scaling exponent k (2) This requires multiple reduced-voltage test points 4) IEEE 115-1995 [16] instead suggests a semilog fit Inom (Unom ) ¼ e f2 (Unom ) , (3) where e is the base of natural logarithms (log) and Unom À Umeas1 Imeas2 log f2 (Unom ) ¼ Umeas2 À Umeas1 Imeas1 þ log(Imeas1 ): (4) (7) k U / U; Riron / 1=liron ; liron / 1=U : (8) 7) A second-degree polynomial fit of measured data is also considered A complete onset of saturation is only possible at full voltages, which can be rather difficult to achieve in the tests rooms for many machines The accuracy of the traditional scaling methods, i.e., those based only on measurements, is dependent upon the maximum test voltage that is possible during the measurements (test room supply and motor stall time limitation), as well as on the number of measurements (test room cost and test window limitation) In summary, the extrapolation accuracy of the employed curve-fitting technique is benchmarked in this study Accuracy Evaluation Through Statistical Analysis The accuracy of the different scaling methods is evaluated by comparing the differences between the scaled results obtained from n À reduced-voltage measurements and the n:th measurement 1) The studied scaling method is applied on the n À measurements (lowest test voltages), estimating the starting current value on the n:th measurement point (highest voltage) 2) The estimated value in n:th measurement point is compared with the actual measured value This is followed by a detailed statistical analysis of these differences for certain sets of studied motors The analysis serves as the basis for defining the confidence level (error margin in terms of mean difference and standard deviation) of the various studied scaling methods Classification of the motors based on different criteria such as power range, voltage level, and speed is also carried out Comparative Analysis The following methods are primarily investigated for benchmarking and also for improving their accuracy with the better curve-fitting techniques Log–Log, Semilog, and Linear Scaling The linear fit [the old practice, see 1) mentioned under the “Description of the Methods” section], the semilog fit [see 4) above] and the log–log fit [see 2) and 3) above] are compared first It is found that, of these, the last is the most promising for predicting the full-voltage currents from the reduced-voltage tests Some of these results for synchronous motors have been provided in [17] IEEE INDUSTRY APPLICATIONS MAGAZINE NOV j DEC 2010 WWW.IEEE.ORG/IAS Traditional scaling methods rely only on measurements These include the methods presented in literature [17] as well as those recommended by standards such as IEEE 1122004 [15] and IEEE 115-1995 [16] The locked-rotor tests are made for maximum available test room voltages The onset of saturation and its dependency upon the test voltage is captured through various curve-fitting techniques of the three to six available measurement results The full-voltage current is obtained through an extrapolation procedure The accuracy of the result is based on the employed curve-fitting technique, the maximum test voltage, and the number of tests Usually, reference to any physical dependencies and design factors is not explicitly made Some of these methods that have been studied by the authors are listed below: 1) The current is calculated as varying directly with voltage (1), with the exponent k ¼ (IEEE 112-2004 [15]) and then the full-voltage current is corrected based on a certain rule of thumb (experience-based) correction factor Inom ¼ Imeasured 5) A saturation factor approach is based on estimating the saturation factor (ksat ) from tests and, consequently, predicting the full-voltage current from 31 values, whereas the saturation The log–log curve fitted extrapfactor method always predicts olation provides the lowest error THE FEM higher than the measured values to the actual current at maximum tests voltage This error is APPROACH IS roughly two to ten times better Accuracy Levels than the other two approaches To provide generalized accuracy figures INTRODUCED AS In this comparison, the highest for the investigated scaling methods, a AN ALTERNATIVE test-room voltage is treated as more detailed study was carried out for the rated voltage the log–log, the saturation factor, and TO THE the second-degree fit scaling methods n A difference of up to 30% in Results are provided in Table for a scaled values is observed when TRADITIONAL scaling step of 0.4 p.u., i.e., when the comparing the log–log approach maximum test voltage used for scaling [see 3) above] against the linear SCALING of the results is 0.4 p.u below the approach [see 1) above] maximum available test voltage The n For the log–log approach, a METHODS THAT scaled results in Table are compared spread in scaled results of up to EMPLOY ONLY against the maximum available test 20% is observed when k (2) is voltage, which may be lower than derived from only two test MEASUREMENTS p.u For example, when the maximum points [see 2) above] correspondavailable test voltage is 0.6 p.u., the ing to different test voltages estimation at this voltage is made from n There is also spread in log–log estimations when successive highest voltages are tests that use a maximum test voltage of 0.2 p.u The results in Table show that this set of scaling methremoved from the curve-fitting algorithm The spread among scaled results is as high as around 5% ods have an uncertainty margin of roughly 10% The simpler in certain cases when at least three test points are methods always provide an underestimation, whereas the used for the estimation Results with only two test more advanced ones can provide an overestimation as well It is found that the error margin of the saturation factor points have a much greater spread approach (2) decreases more than that of the other approaches, as the scaling voltage step is decreased Still, all Log–Log, Saturation Factor, and Second-Degree Fit Scaling The best of the earlier discussed scaling methods, the log–log these curve-fitting methods depend primarily on the quality method [see 3) discussed earlier] is consequently compared for of information embedded in the test data (saturation), the accuracy against a suggested second-degree polynomial fit full-voltage scaled results becoming poorer with a decreasing [see 7) above] as well as the traditional saturation factor tech- highest test voltage Hence, it can be concluded that the scalniques [see 5) and 6) stated earlier] For the studied cases, it is ing methods that are based only on measurements are unacceptable for zero-tolerance inrush current design motors found that: n All methods are equally good if the maximum test FEM Approach 1: Tuning of FEM Models voltage is above 0.95 p.u In the first variant, a two-dimensional (2-D) FEM simulation n For scaled results using 0.6 p.u maximum test voltage, the differences to actual measured values is tuned against a reduced-voltage measurement with an are roughly of the same order for all the three expectation that the full-voltage simulation could provide methods The saturation factor method exhibits correct results A common understanding is that the uncerthe lowest differences band, i.e., 5.5% The sec- tainties/errors in 2-D FEM simulations are due to the wrong ond-degree fit has a difference band of 6%, input data fed, such as the end-windings inductances and which is slightly below that of the log–log material properties, to 2-D-FEM models It was believed that such differences could be neutralized by a single calibration method of 6.5% against the measurements, achieved through tuning of the n The differences for the second-degree fit are random, whereas for the other two methods, they are input data, such as the end-windings inductances One such example is provided in Figure Here, the always in the same direction Thus, the log–log fit is always predicting lower than the measured FEM model of a certain machine is tuned against a reduced-voltage locked-rotor measurement through variation of end-winding inductances and TABLE SCALING THROUGH MEASUREMENTS: material properties However, ACCURACY LEVELS FOR A SCALING STEP OF AT LEAST 0.4 PER UNIT this approach did not yield Error to the correct full-voltage starting Scaling Method Trend Measurements current For a scaling step of Saturation factor (1) À10 to À2% Underestimation 0.45 p.u., an underestimation by 7% can be observed This Saturation factor (2) À3 to þ6% Under-/overestimation error is believed to be due to the IEEE log–log À10 to À5% Underestimation influence of three-dimensional (3-D) effects upon saturation Second-degree polynomial À5 to þ5% Under-/overestimation (frame region and end-winding IEEE INDUSTRY APPLICATIONS MAGAZINE NOV j DEC 2010 WWW.IEEE.ORG/IAS n 32 inductances), which are missed in a reduced-voltage 2-D tuning endeavor By considering the starting current (saturation) levels for other machines, an error figure of 10% was established for this method The error margin in this case is still comparable with the traditional scaling methods Further, the amount of effort required in the iterative tuning procedure for each machine in this method makes it unsuitable for daily factory scaling usage and is not discussed further n n n n n The Analysis Method The analysis method is investigated for accuracy on 205 different machines Table provides a listing of the machine types and frame sizes that have been used in the analysis (full details are provided in Table 1) In the presented analysis: Current Results for the differences are summarized in Figure It can be seen that the mean difference is 1.4% and the Difference: FEM Measured (%) Starting Current: FEM to Measured (%) Motor 0 –15 –20 Voltage (kV) 10 12 Motor Motor and O, Solid Line: Harmonic [], Dashed Line: Time Stepping Test Voltage (p.u.) TABLE DESCRIPTION OF MACHINES USED IN THE TRENDS-BASED FEM STUDY Frame (mm) No Poles No Voltage (kV) No 355–400 80 81 ... direct-online, single-cage induction motor are typically specified to be four to seven times the rated motor currents (4–7 p.u.) [1] These figures are for medium–large induction motors, whereas... saturation is due to 2-D effects, which would be dominant for machines having a high aspect ratio (long machines with a low shaft height) and/or for machines havInterpretation of Different Slopes Figure... starting currents have been studied through a detailed statistical analysis of hundreds of medium–large induction machines The various traditional scaling methods using only measurements are evaluated