565 IEEE Transactions on Energy Conversion, Vol. 5,No. 3, September 1990 MEASUREMENTOF THE TORQUJiXSPEEDCHARACIERImCSOF INDUClTON MOTORS USING AN IMPROVED NEW DIGITAL APPROACH B. Szabados, SM,IEEE, J.H. Dableh, SM, IEEE, R.D. Finday, SM,IEEE G.M. Obermeyer, R.E. Draper McMaster University, Power Research Laboratory, Hamilton, Ontario, Canada, U S 4K1 Westinghouse Canada Inc., Motors Division, Hamilton, Ontario, Canada, LAN 3K2 A new measurement technique for the determination of the torque-speed characteristics of induction motors is presented. "his technique is in compliance with the IEEE standard test procedure for polyphase induction motors and generators. It is based on the acceleration method which is performed under no load conditions. A fast data acquisition system is used to record the speed and other signals of interest. The data is then processed digitally with the objective of removing undesired extraneous signals,while preserving the accuracy of the machine characteristics over the complete range of speed. To achieve this objective, new and unique algorithms to perform adaptive window-size average filtering and numerical differentiation have been developed. Experimental tests performed on 1-hp,20-hp and 600-hp induction motors confirmed that this approach has several advantages over other currently available methods. It produces very accurate results over the full dynamic speed range of the machine and permits precise correlation between the switchingtransients on the line with the variation of the speed signal. line drawn along the middle of the oscillations would indicate tha the minimum torque of this motor is at least 160% of the rated torque. This is in accordance with the specificationsfor this motor. However, the user experienced difficulty bringing the machine to full load speed. The motor stalled before it reached 600 rpm. The user was led to believe that the lower envelope of the oscillation should be considered the representative characteristic of the motor. Clearly, a new measurement technique was required. It must be capable of producing a smooth torque-speed profile without distorting the actual characteristics of the machine. The problem was to eliminate the need for interpolationof the digitally differentiated analog signal. The analog signal is itself very noisy due to coupling noise, interference, etc. Hence differentiation is highly suspect. 300 1 Torque (in % of rated torque) A KEYWORDS Induction Motors, Torque-Speed Characteristics, Acceleration Method, Data Acquisition, Numerical Filtering, Digital differentiation. INTRODUCIION A new measurement technique for the determination of the torque-speed characteristics of induction motors was developed recently in the Power Research Laboratory at McMaster University in response to a request from Westinghouse Canada Inc., Motors Division [l]. The need for this development became necessary when the accuracy of the torque-speed profiles for two, three phases, 2@hp, induction motors was in question. Specifically,the value of the minimum torque at an approximate speed of one third rated speed, obtained from the graph shown in figure 1 was doubted. The torque-speed charactesisticsillustratedin figure 1for one of the motors was obtained using conventional equipment and an analog differentiator. The oscillationsin the curve, particularly in the region of minimum torque, obfuscated the actual torque. A 90 WM 142-0 EC A paper recommended and approved by t h e IEEE Rotating Machinery Committee d t h e IEEE Power Engineering Society for presentation a t t h e IEEE/PES 1 9 0 Winter Meeting, Atlanta, Georgia, 8, 1990. Manuscript submitted February 4 August 30, 1989; made a v a i l a b l e for p r i n t i n g December 6, 1989. - 0 Figure 1: 600 1200 1800 Torque-Sped characteristic for a three phase B h p squirrel cage induction motor obtained using conventional analog di€ferentiator. The new measurement method was designed to comply with the IEEE standard test procedure for polyphase induction motors and generators[2]. The IEEE standard 112-1978outlines four methods to obtain data for a torque-speed curve. The selection of which method to apply depends upon the characteristics of both the machine and the testing facilities. The main criterion is that sufficient number of test points must be recorded to ensure that reliable and accuratecurves, includingirregularities, can be produced from the test data. The technique presented in this paper is based on the acceleration method described in section 4.9.2.2 of the IEEE standard mentioned above. According to this method the motor is startedwith no load after it has been rotated manually in the reverse direction to that expected when the motor is energized. The acceleration at 0885-8%9/90/0900-0565$01.00 0 1990 IEEE 566 each instant of time is determined by differentiating the speed signal. The torque at the respective speed is given by the product of the acceleration by the moment of inertia of the rotating part. If the moment of inertia is not known the relative torque versus time is obtained. The absolutevaluesof the torque need to be scaledfrom the locked rotor or pull out torque measurements. Accurate measurement of the speed and determination of its first derivative, the acceleration, are crucial. Over the past two decades several methods to measure the velocity of rotating shafts have been reported in the literature [3-81. Older methods are based on analog principles; they generally suffer from low resolution and severe noise contamination. Early versions of digital methods were reported in the late sixties and early seventies [3-51. In these versions the sampling period varied with the speed. This had the disadvantage of having very slow readings at low speeds and the necessity of processing the time information. Corrective methods to increase the number of pulses per revolution and obtain fast readout were suggested [6,7]. However, these methods required theuse of either a fairly complexsensing method [6] or an accurate servomotor [7]. A recent digital method to obtain angular velocitywas given by Christiansen [8]. None of these digital methods can be used to obtain speed by differentiationbecause of the coarseness of the quantization. Even if a fast clock is used, one runs into either clock overrun or too many counts per intervals. An intelligent device which switches between modes might conceivably be constructed, but the complexity outweighs the use as a conventional tool on an industrial test platform, and its proper functioning has yet to be proven. The method presented in this paper uses a fast data acquisition system to sample the output of a dc tachometer as well as other parameters of interest, such as the line currents and voltages. The collected data is then processed digitally to remove the noise, perform dynamic average filtering to eliminate extraneous coupling vibrations and numerically differentiatethe resultant clean speed signal to obtain the relative torque profile. The various algorithms used to achieve this objective and typical results obtained from different induction motors are also presented in this paperDEscRipIlON OF THE NEW DIGITAL,APPROACH The data acquisition system used to perform the dynamic speed measurements consists mainly of a conventional dc tachometer,an analog-to-digital (A/D) converter and a personal computer. The A/D converter has 8 differential channels and can be operated at avariable sampling rate up to 40 kHz. An external pulse generator is used to trigger the A/D converter and set its samplingrate. If the time required for the tested machine to reach its maximum speed is larger than 0.2 second, a sampling rate of 5 kHz per channel is quite sufficient for the acceleration tests. All of the tests performed during the development of this method were conducted at full rated voltage of the respective machine and without any loading, except for the small tachometer. The machine was rotated, by hand for small machinesbelow 25-hp, in a direction opposite to that of the expected operatingrotation,prior to energizing the machine. In the case of large machines, plugging was used to achieve starting from a reverse rotation. This was done in accordance with IEEE standard test procedure [2]. It had the advantages of deliminating the zero-crossing of the speed curve representing the sampled data. It also allowed the switching transient on the supply voltage waveform to disappear while the machine was deccelerated from the imposed reversed rotation. Therfore, the collected data in the speed range of zero to full speed was not affected by the switchingtransients due to the supply bus. Figure 2 illustrates typical raw speed data obtained during an acceleration test of a 600-hp induction motor. This raw data was collected at a sampling rate of 5000 samples per second. As can be clearly seen, this data is contaminated with several undesired signals. Such signals include tachometer commutator spikes and modes of oscillationsexternal to the motor itself, specificallylarge coupling oscillations. I Figure 2: . < F -. Typical unprocessed speed data collected during acceleration of a 6Whp induction motor at a sampling rate of 5 kHz The first task of the data processingphase adopted in this measurement method involves the cleaning of the collected speed data to eliminate all of the extraneous signalswithout distorting the actual profile of the speedcurve and the oscillationsthat are caused by the electromagnetic field in the machine operation. To achieve this objective special adaptive average filtering algorithms have been developed and used to preserve the accuracy of the data over the full speed range between zero and synchronousspeed. This is done in contrast to the normal averaging technique with a fixed bandwidth. Since the output waveform of the tachogenerator is proportional to speed, the amplitude of unwanted oscillations due to uneven windingsare proportional to speed, while their frequency is inversely proportional to the speed. Hence a fixed bandwidth filtering would have to be tuned for a low speed, and thereby removes the higher frequencies that are actually caused by the electromagnetic fields of the machine in the upper range of speed. In the proposed method an adaptive bandwidth is used. The bandwidth is inverselyproportionalto speed,with a clampingvalue at very low speed. The next data processing task consists of performing numerical differentiation of the cleaned speed curve. The success of this task and the smoothness of the derivative curve depends on the cleanliness of the speed curve and the differentiating algorithm used. This is outlined later in this paper. The final task is plotting and presenting the results in curves as functions of time or as torque versus speed characteristics as may be required by the user. DATA PROCESSING AND TYPICAL NUMERICALRESULTS A number of software packages were developed to perform the various data processing tasks mentioned in the previous section. The preliminary preparation of the data consists of separatingthe 567 samples of the various channels of the A/D converter and identifying the monitored signals. A digital scope program is then used to view each signal and to select the interval of interest to be stored for further processing. Note that the A/D converter is activated prior to energizing the motor and is kept runing for a shortwhile after the motor is de-energizedto ensure complete coverageof the acceleration period. Thus the digital scope is very useful to view the data and cut out any redundant segment. A typical speed signal extracted from the collected data by using the spliting and digital scope routines has been displayed in figure 2 above. Filtering of the speed signal is performed very carefullyin a number of steps to remove only the extraneous signals and preserve the main signalover the complete speed range of the machine. The first step is aimed at removing the noise spikes caused by the commutator brushes of the tachometer used and other sources of random spike noise. This is achieved by applyingawindowingpeak removal algorithm. This algorithm simply examines three data points at a time (window size is 3 ) and checks if they are in a monotonic ascending or descending order. If so, the central point is left intact. Otherwise, the point in the middle is replaced by the average of the two neighbouring points to prevent it from appearing as a spike. Note that the median filter algorithm is also included in the “tool kit”. In case several sample points cover the spike, a slidingwindow of 10 samples using the median filter is applied as a more robust spike removal algorithm. This algorithm is very slow though, and it was found that the 3 point window (much faster) used above is sufficient. If needed, it can be applied more than once, to lead to a better result. Applying this algorithm to the speed data shown in figure 2, with a window around the top knee of the curve (figure 3 (a)) provides the results illustrated in figure 3(b). The ripples remaining in this graph are thought to be due to the uneven windings of the tachometer. A special dynamic window averaging algorithm was used to filter out the small ripples in the above signal. Since these ripples are caused by the uneven windings of the tachometer, their frequency depends on how fast the tachometer is rotated. Therefore, the size of the averaging window ought to be adjusted dynamically as a function of the speed. Note that the size of the window is inversely proportional to the speed. However, to establish a referencefor the size of the window one should “zoom-in”to the synchronous speed portion of the curve, using the digital scope program. The width of the repetitive pattern of the high frequency ripples can then be found.The number of samples within this pattern establishes the window size at synchronousspeed, see figure 3 (b). At half synchronous speed the window sizewould be double the original measured size. Naturally, at low speeds the size of the window has to be clamped to a maximum window size. Figure 3 (c) shows the results of applying this algorithm to the data of figure 3 (b). At this stage , the oscillation modes due to the flexible coupling between the shafts of the motor and tachogenerator are still present in the speed signal. These oscillations can be grouped into two broad categories: a low frequency mode and a high frequency mode. Each of these modes is filtered out seperately using a piecewise fiied window averaging algorithm.To ensurethat these modes are identified properly and no other oscillations that are inherent to the motor itself are removed, the speed data from a run down test is examined. The data collected after the motor is de-energized, as it deccelerates from synchronous-speedto zero, displays only the modes of oscillation due to the mechanical parameters of the machine itself. Oscillationsdue to electromagnetic fields are elimi- Figure 3 Processhg of speed data (“zoom-in“ on upper knee) a) Raw data b) commutator spikes and random noise removed c) Tachogeneratoruneven winding noise removed d) couplingnoise removed nated by this process leaving flexible coupling noise and other mechanical modes. Therefore, the run down test data is used to establish the appropriate size of the averaging window at various subranges of the synchronous speed. These window sizes are then used to filter the data of figure 3 (c). Figure 3 (d) show the results of applying this algorithm after the high and low frequency modes of coupling ocillations have been filtered. The amount of random noise and extraneous signalsthat have been removed from the raw speed data can be appreciatedby superimposing the raw data of figure 2 and the cleaned up signal, as shown in figure 4. At the completion of the above filtering, the speed signal can be fed to the numerical differentiationalgorithm to derive the acceleration or relative torque profile of the machine. In the early stage of development of this method, standard differentiation algorithms were used. Namely, these were various polynomial fittings and spline fittingswhich all introduced large parasitic oscillationsin the derivative masking the real curve. Unless the speed data is sufficiently smoothed, the resultant derivative curve would display a significant amount of distortion which is not representative of a physically realizable system. This indicated that some additional smoothing of the speed signal was needed. A mean square window averaging algorithm was used to achieve this objective. This new smoothed curve was then differentiated numerically. The algo- 568 NEW DIFFERENTIATlON AEORlTHM speed (rpm) The need to perform mean square window averaging in some cases and the smoothing of the derived torque curve were considered to be disadvantageous because of three main reasons: - 4000 - 3000 a) The final degree of filtering is not determined until the differentiation is executed. b) The smoothing of the derivative curve results in small attenuations of the curve peaks: the amount of smoothing versus the reduction of the peak value is left to the judgement of the operator. c) The additional mean square averaging is very time consuming even on a powerful personal computer. - 2000 - 1000 To circumvent these disadvantages and the inconvenience of having to execute these tasks, a novel differentiation algorithm that treats sudden changes in the signal was developed. Basically, the algorithm looks at thresholds of variations within a certain sensitivity region and adapts a window size dynamically. Within this window the derivative is assumed to be continuous. The algorithm starts at a sample point So. It searches subsequent samples until the variation AS = I Sn - So I is larger than a preset tolerance t.It is assumed that the slope I‘betweeen So and Sn, the new “breakpoint”, is linear and is given at sample (i) by: r i = i(Sn-So)/n + ro. This assures continuity of the slopes at breakpoints, and does not introduce high frequencycomponents due to quantization. This algorithm was implemented in the data processing of this method with a high degree of success.The success of this new differentiation algorithm was demonstrated when the resulting derivative curve showed that the amplitude of the lower frequencieswere not affected by the tolerance parameter of the algorithm. Furthermore, the derivative curve stabilized quickly above a certain tolerance level. Figure 6 shows the results of applying this algorithm to the speed data of figure 4 displayingthe final torqueversus speed curve. Figure 4 Speed data before and after digital filtering Relative Torque IA - 100% Time Torque (in % of pull out torque) Figure 5: Relative torque versus time curve for a m h p motor a) First method of differentiation b) Smoothing of curve (a) rithm was to take a certain window size and find a linear best fit to the points within the window. The slope of this line is taken as the best value for the slope at the center point of the window. The resultant derivative curve still showed high frequency oscillations introduced by the algorithm, as shown in figure 5 (a). These were easily removed by conventional digital filtering, but care had to be taken because the amplitude of the lower frequencies were also affected slightly. Thus, the resulting smoothed relative torque versus time curve, shown in figure 5 (b), may be questionable in spite of the fact that it was repetitive and agreed well with design data. A new differentiation algorithm was developed to overcome the problems. Figure 6: Torque versus speed curve produced by the new differentiation algorithm. The oscillations remaining in the final torque profile are attributable to the actual electromagnetic pulsation torques. They form an integral part of the machine characteristics. 706Apeak 1 (a) Line Current DI!XUSIONS The appoach presented in this paper to determine the torquespeed characteristic of induction motors has been demonstratedto produce excellent results. The capability of this method to produce smooth curves by eliminating extemal disturbance signals, while preserving the characteristic pulsation torques of the machine, has undisputed advantages over other available methods. A typical indicator confirming that electromagneticallycaused oscillations of the machine are preserved can easily be examined by focusing on the speed or torque traces at synchronous speed. Figure 5 (b), for example, shows the small variation in the torque as the machine speed oscillates about synchronous speed under no load. Also the rigour of this argument can be validated by examining the steady state and run down segments of the speed data as they are subjected to the numerical filtering algorithms.To illustrate this point typical results and intermediateoutput of the various filtering algorithms are presented in figure 7 (a) to (d) for a portion of the run down test data. ! Tie Figure 8 Correlationbetweenthevariationsof the Line Current (a) and the Torque @) of a W h p inductionmotor. the rotor would be pulled into alignment with the stator by the air gap field. This results in torque oscillationsdue to air gap reactance variations. This phenomenon had not been detected with analog methods. The torque-speed characteristic of the 20-hp motor is shown in firgure 9 to facilitate direct comparison and illustrate the merits of the new digital approach. Figure 7: Filteringof run down test data a) Rawdata b) Commutatornoise removed c) Uneven winding noise removed d) High-frequencycoupling noise removed Another advantage of the method is its ability to collect data pertinent to other relevant variables in the system. For example, in the tests performed on a 600-hpmotor the line current of one phase was monitored. Figure 8 (a) shows the variation of this current over the complete acceleration period. It is interesting to compare the variations in the torque profile to those in the starting current profile, as depicted in figure 8. Note that the small oscillations in the torque near the starting of the machine coincide very well with the oscillations seen on the envelope of the currentwaveform. Also,when the motor reaches about two third of its synchronous speed, an unexpected torque oscillation occurs, without any signs of line current oscillation. This has been explained bv the fact that the machine under test had no axial stabili&ion, and as much as 10cmaxial travel was noticed. Hence One possible limitation of the proposed method may arise in the case of certain machines that may have electromagnetic torques in the same range as the noise introduced by the tachogenerator or the coupling transients. In this case, these actual torque pulsations are eliminated in the filtering process. Although it is reasonable to Torque (in % of rated torque) 600 1200 1800 Analog method 660 1~00 Digital method 1 Figure9 Comparison between d o g and proposed digital method torque-speed characteristicsof a 2OHp motor I ., 570 assume that the noise bandwidth of the tachogenerator and flexible couplingare outside the frequencyrange of the machine torque, the authors are investigatingaspecial measure to circumvent this issue. The new measure involves using two completely different tachogenerators with clearly different non overlapping noise bandwidth and coupling characteristics.Applying the new measurement technique to the signalsof both tachogenerators clearlyreveals whether this limitation applies or not for a specific test run. CONCLUSION A new digital technique to determine the torque-speed characteristics of induction motors, in accordance with the acceleration method of the IEEE standard test procedure for polyphase induction motors and generators, has been presented. In contrast to other currently available methods, this technique has the advantage of eliminating all extraneous noise and disturbance signals while preserving the actual characteristics of the machine. The digital data processing and algorithms used to realise the advantages of this technique have also been discussed. The use of these algorithms is not restricted to this application only. The digital scope, the dynamicwindow averaging and the new differentiation algorithms provide unique and powferful tools to handle a wide range of contaminated signals, such as those observed during transient conditions. The validity of these algorithms have been verified by investigating the data produced during the steady state and run down operation of the machine. This investigationwas also crucial in establishing the size of the various windows used, and to systematically identify the disturbance signals. The data acquisition system used to collect and store the desired information offers a high degree of flexibility as far as the amount of collected data and sampling rate are concerned. The availability of accurate measurements of other variables in the system is very beneficial in assessing the torque profile of the machine and for trouble shooting tasks that may be required. The approach described in this paper can be automated to a large extent to form a standard test method in the manufacturing of induction motors. REFERENCES [l] B. Szabados, J.H. Dableh, R.D. Findlay and D. Stafford, “A New Approach For Measurement Of The Torque-Speed Characteristics Of Induction Motors”,Paper Accepted For Presentation At The Fourth International Conference On Electrical Machines And Drives, September 13-15,1989, London, England. [2] IEEE Standard 112-1978, “IEEE Standard Test Procedure For Polyphase Induction Motors And Generators”,pp 7-30. [3] G. Hoffman de Visme, “Digital Processing Unit For Evaluating Angular Acceleration”, Electron. Eng., 40,April 1968, pp 183-188. [4] A. Dunworth, “Digital Instrumentation For Angular Velocity And Acceleration”, IEEE Trans. Instrum. Meas., IM-18, June 1969, pp 132-138. r ----r [ S ] N.K. Sinha, B. Szabados and C.D. DiCenzo, “New High Precision Digital Tachometer”, Electron. Lett., 7, April 1971, pp 174-176. [6] B. Habibullah, H. Singh, K.L. So0 and L.C. Ong, “A New Digital Speed Transducer”, IEEE Trans. Ind. Electron. Contr. Instr., IECI-25, No. 4, November 1978. [7] C.D. DiCenzo, B. Szabados and N.K. Sinha, “Digital Measurment of Angular Velocity For Instrumentation And Control”, IEEE Trans. Ind. Electron. Contr. Instr., IECI-23, No. 1, February 1976. [8] C.F. Christiansen,R. Battaiotto, D. Fernandes and E. Tacconi, “Digital Measurment of Angular Velocity For Speed Control’’, IEEE Trans. on Ind. Electron., Vo1.36, No. 1, February 1989, pp 79-83. Barna !jzabadq was born in Hungary and received the DiplSme d‘IngCnieur ENS1 from the Universitt de Grenoble in 1967, and his Master’s and Ph.D. degrees in Electrical Engineering from McMaster University, Hamilton, Ontario, Canada, in 1969 and 1971, respectively. He is presently Professor of Electrical and Computer Engineering at McMaster University in the Power Research Laboratory. He is working in the area of solid state converters, field representations in electrical apparatus and electromagnetic interference, and local area networking in factory environment. All these projects are done in a tight cooperation with industrial partners, Westinghouse Canada and General Motors. Dr. Szabadosis a Senior Member of the IEEE, and holds positions in several committees. He is also a member of several other international professional societies. Joseph H. Dableh was born in north Lebanon. He received his B.Sc.E. and M.Sc.E. degrees in Electrical Engineering from the University of New Brunswick in 1976 and 1978, respectively, He obtained his Ph.D. degree in Electrical Engineering from McMaster University in 1986. He worked as a Research Engineer at the Ontario Hydro Research Division from 1978 to 1987. He has been at McMaster University since 1987,where he is a part-time Assistant Professor of Electrical and Computer Engineering, and a full-time Senior Research Engineer in the Power Research Laboratory since 1988. His research interests include pulse power and electromagnetic metal forming for assembly and rehabilitation of CANDU nuclear reactors, electromagnetic field computations in electric machines and power apparatus, control and power systems. He holds several patents related to nuclear reactor repair and pipe-type cable systems. He is a registered professional engineer in the Province of Ontario and a Senior Member of IEEE. RaM o lndD F .n id a lv was born in Toronto, Canada, obtaining degrees of B.kSc., M.kSc., and Ph.D. in 1963, 1965 and 1968 respectively, in electrical engineering from the University of Toronto. Dr. Findlay is a registered professional engineer in the Province of Ontario,a Senior Member of IEEE, and a member of ASEE. His interestsinclude electromagnetic fieldsand losses in power apparatus, an area inwhich he holds one US patent.From 1967to 1981 he was affiliated with the University of New Brunswick, achieving the rank of Professor in 1978. During 1972-3he was a project leader at Canadian General Electric Co. He has been a Visiting Fellow at the University of Southampton (U.K.) in 1979-80,at the Katholieke Universiteit Leuven (Belgium) in 1988, and at the Commonwealth Scientificand Industrial Research Organization (Australia), also in 1988. Dr. Findlay is a Professor and a member of the Power Research Laboratory of McMaster University, where he has held an appointment since 1981, and during 1984-7 as Assistant Dean of Engineering. Robet E. Draper was born in Welland, Ontario, Canada. He received a BSc of Applied Science in Electrical Engineering from the University of Waterloo in 1985. From 1985 to 1987 he worked in the field of application engineering for Union Gas Limited in Chatham, Ontario.From 1987to present he is engaged in the design of large induction motors for Westinghouse Canada in their Motor Division in Hamilton, Ontario. He is a member of the Association of Professional Engineers of Ontario. F.M.Obermever graduatedfrom McMaster UniversitywithaBSc in Electrical Engineering in 1977. From 1977to 82 he worked in the steel and paper industries and joined Westinghouse in 1982. He is presently engineering supervisor in the Motor Division and his technical interests are rotating equipment. He is a member of the Association of Professional Engineers of Ontario.