Method for state diagnosis of asynchronous motor

Abstract

The invention relates to a method for diagnosing the state of an asynchronous motor (M) to which three-phase voltages UR, S, T (t) having a supply frequency (fU) are applied, comprising the following steps: a) detecting a voltage time sequence (Ut) of the voltages UR, S, T (t) and a current time sequence (It) of three-phase motor currents IR, S, T (t) flowing through the asynchronous motor (M) as a result of the voltages UR, S, T (t) applied to the asynchronous motor (M); b) calculating a current spectrum I (f, t) and a voltage spectrum U (f, t) on the basis of the detected time series (It, Ut) by means of FFT; c) calculating one or more of the following time-dependent spectral variables 1-6: 1) the current components Ia (f, t), Ib (f, t), where Ia (f, t) and Ib (f, t) are formed from IR, S, T by Park transformation; 2) total current amplitude AI (f, t) = square root of (Ia (f, t)) 2 + (Ib (f, t)) 2); 3) the phase angle phi ab (f, t) between the current components is equal to arctan (Ia / Ib); 4) the phase angle phi UI (f, t) between the current and voltage components = arctan (U (f, t) / I (f, t)); 5) instantaneous spectrum power P (f, t) = U (f, t) * I (f, t); 6) admittance X (f, t) = I (f, t) / U (f, t), d) is a spectral variable calculated for anomaly analysis; and e) reporting the exception when the exception is confirmed.

Description

[0001] The invention relates to a method for diagnosing the state of an asynchronous motor.

[0002] Asynchronous motors (ASMs) are very common in industry due to their cost-effective and robust design. They are used in numerous applications, such as fans, saws, and pumps, ranging from relatively small systems in the multi-watt range to systems in the megawatt range. As a result, ASMs are among the largest consumers of electricity worldwide. The widespread use of ASMs offers enormous potential for increased availability and reliability, saving both time and money. This is the context for the concepts of "Condition Monitoring" (CM) and "Predictive Maintenance" (PM).

[0003] One known method for identifying faults in electric motors is Motor Current Signature Analysis (MCSA), see, for example, EP 3961230 A1 (Siemens), March 2, 2022, and Kumar, K. Vinoth et al.: A Review of Voltage and Current Signature Diagnosis in Industrial Drives. International Journal of Power Electronics and Drive Systems (IJPEDS), Volume 1, Number 1, September 2011, pp. 75–82, ISSN 2088-8694. The principle of MCSA is to detect the electrical phases of the motor, typically three (L1, L2, L3). In addition to MCSA, the spectral components of the voltage are also considered for CM operation (MVSA; Motor Voltage Signature Analysis).

[0004] Measuring devices, such as the current / voltage detection modules in Siemens' SIMOCODE motor management system, measure the voltage U of the electrical phases in addition to the current I. These measurements are performed at a high scanning frequency and are time-synchronized as much as possible. These measured values ​​are then analyzed. For example, Fast Fourier Transformation (FFT) is used to detect and quantify motor faults and operating conditions in the frequency domain. Traditional MCSA is applied to quasi-steady-state conditions, i.e., nominally constant speeds. An extension of MCSA is the use of short-term FFTs or wavelets to analyze processes in a time-resolved manner.

[0005] Asynchronous motors (ASMs) operate based on the principle of electromagnetic induction, where the interaction between current and magnetic fields in the stator and rotor generates torque, thereby driving the rotor shaft. Motor faults such as eccentricity, particularly air-gap eccentricity (when the rotating axes of the motor's stator and rotor do not completely coincide), broken bars, bearing damage, or misalignment, can alter the amplitude of the fault-signaling frequencies under constant slip. The fault frequencies themselves are slip-dependent, meaning that changes in slip shift the frequency location of the fault frequencies. Therefore, motor faults can be identified using MCSA based on the amplitude changes of these characteristic frequencies in the frequency spectrum.

[0006] Therefore, the task of the MCSA is to determine the slip s of the asynchronous motor as accurately as possible. One alternative for accurate slip determination is the determination of the frequencies of the so-called main slot harmonics (PSH), i.e. those originating from the rotor design, according to the following equation:

[0007] f PSH =[k•(R / p)•(1-s)±ν]•f supply Equation (1)

[0008] in:

[0009] f PSH is the frequency of the main slot harmonic

[0010] f supply is the frequency of the supply voltage, for example 50Hz

[0011] k is a natural number

[0012] p is the number of pole pairs = number of pole pairs

[0013] R is the number of rotor slots = number of rotor slots

[0014] s is the slip

[0015] ν is an odd number 1, 3, 5, 7, ... = f supply Harmonic order

[0016] For this purpose the PSH frequency f is evaluated PSH , these frequencies are functions of the number of rotor bars R (e.g. 28, 44, 150, etc.), the number of pole pairs p, and the slip s (usually between 0 and 0.05 (0-5%)): f PSH =f(R,p,s), based on the corresponding current signals of one or more electrical phases (L1, L2, L3). The evaluation can also be performed within the spectrum of the reduced current amplitudes D and Q, or within the spectrum of the total amplitudes of the phase currents. The reduced current amplitudes D and Q are the current amplitudes along the d and q axes of the two-axis coordinate system, into which the three-phase variables along the U, V, and W axes are transformed using the Park transformation. The problem here is the network supply frequency f supply PSH frequencies other than fPSH The amplitude A(f PSH ) and the amplitude of the supply current A(f supply ) is very low compared to: A(f PSH )<10 -2 A(f supply ).

[0017] Faults such as broken rotor bars (BB), eccentricity (ECC), bearing faults and the effects of coupled machines, as well as the load state of the machine, can be detected in the motor current spectrum by amplitude changes at characteristic frequencies. In MCSA, the amplitudes for various fault conditions are evaluated:

[0018] i) Broken rotor bars (BB=BrokenBars):

[0019] f BB1 =(1±2k•s)•f supply Equation (2)

[0020] ii) Static / dynamic rotor eccentricity (ECC, see ISO 20958:2013)

[0021] f ecc1 =[1±(k / p)•(1-s)]•f supply Equation (3)

[0022] f ecc2 =[(R±n d )•(k / p)•(1-s)±ν]•f supply Equation (4)

[0023] iii) Bearing damage of rolling bearings

[0024] f o =[(N b / 2)•(1–(D b / D p )•cos(β)]•f rot Equation (5)

[0025] f i =[(N b / 2)•(1+(D b / D p )•cos(β) / D p )]•f rot Equation (6)

[0026] in:

[0027] f BB1 is the frequency when the bar breaks

[0028] f supplyis the frequency of the supply voltage, for example 50Hz

[0029] f ecc1 is the eccentric frequency (first family)

[0030] f ecc2 is the eccentric frequency (second family)

[0031] f PSH is the frequency of the main slot harmonic (=PSH)

[0032] f o is the frequency of the outer ring of the bearing (o=outer)

[0033] f i is the bearing inner ring frequency (i=inner)

[0034] f rot is the rotation frequency of the rolling bearing

[0035] k is a natural number

[0036] p is the number of pole pairs = number of pole pairs

[0037] R is the number of rotor slots = number of rotor slots

[0038] s is the slip

[0039] ν is an odd number 1, 3, 5, 7, ... = f supply Harmonic order

[0040] n d is an index that depends on the type of eccentricity (static: n d =0; dynamic: n d =1,2,3,4,...)

[0041] N b is the number of rolling elements

[0042] D b / D p is the characteristic diameter on the rolling bearing

[0043] ß is the contact angle

[0044] For some fault categories, it is necessary to know the number of rotor bars R. This can also be known from the structural design or estimated by known methods.

[0045] In practice, however, not only do the aforementioned theoretical fault frequencies occur in many cases, but drive-train-specific oscillations can also generate new, often unknown, frequencies in the current spectrum. This applies particularly to imbalances, misalignment, and similar "mechanical" faults, which can lead to air gap asymmetries that appear as harmonics of the rotational frequency and, in theory, could result in modulation of the supply frequency. Real drives can exhibit even more complex oscillation patterns due to self-oscillations and resonant oscillations. In known fault scenarios, theoretical frequencies are insufficient to assess the fault and its severity in the current spectrum.

[0046] Therefore, the technical problem to be solved by the present invention is to improve fault identification based on MCSA.

[0047] According to the invention, this object is achieved by a method having the features of claim 1. The method according to the invention is used for state diagnosis of an asynchronous motor to which a three-phase voltage U R,S,T (t). One step of the method comprises detecting the voltage U R,S,T (t) Voltage time series U t and due to the voltage U applied to the asynchronous motor R,S,T (t) and the three-phase motor current I flowing through the asynchronous motor R,S,T (t) Current time series I t Another step has to be to detect the time series (I t ,U t ) Calculate the current spectrum I(f, t) and the voltage spectrum U(f, t). Another step comprises calculating one or more of the following time-dependent spectral variables 1–6: 1) Current component I a (f,t),I b (f,t), where I a (f,t),I b (f,t) is obtained by applying Park transform from I R,S,T 2) Total current amplitude AI(f,t)=√((I a (f,t)) 2 +(I b (f,t)) 2 ); 3) Phase angle φ between current components ab :φ ab (f,t)=arctan(I a / I b ); 4) Phase angle φ between current and voltage components UI :φ UI(f,t)=arctan(U(f,t) / I(f,t)); 5) instantaneous spectral power P(f,t)=U(f,t)*I(f,t); and 6) admittance X(f,t)=I(f,t) / U(f,t). Another step comprises analyzing the calculated spectral variables for anomalies. The step of "analyzing the calculated spectral variables" may comprise the following steps: forming a feature vector from one or more time-dependent spectral variables; and using the formed feature vector as an input value of the ML algorithm, i.e., as training data or test data. Another step of the method according to the present invention comprises reporting the anomaly if an anomaly is determined.

[0048] The present invention utilizes known machine learning algorithms to detect fault conditions of the motor based on the frequencies according to the invention.

[0049] One approach is to use an "unsupervised" anomaly detector. Here, the motor is trained for a specific period of time in a "good" state. During the training phase, the algorithm is only exposed to "good" states and is unaware of "bad" states. After the training phase, anomaly detection is activated. From this point on, anomalies are identified and reported to the user.

[0050] This AI requires input values, so-called "features." These can be derived from the current scan values. While the current scan values ​​could, in principle, be considered "features," this is unsuccessful for the reasons described in points a) through c) below: a) Very large "feature" vectors: At a typical scan rate of 2 to 20 kSps (Sps = samples per second), with a recording time of approximately 1 second, a feature vector with 2,000 to 20,000 values ​​is obtained. The "information content" of each individual value is low. "Compression" of the data, or increasing its entropy, is appropriate here. b) Phase: The "phase" of the data within the timeframe is a problem. Depending on the time at which the current value recording begins, a random phase occurs. While it is possible to start the recording at a specific value, this would result in a phase "out of focus." A PLL could also be used, but this is technically very complex. c) Quantity: The current values ​​are quantitatively dependent on the load; that is, higher motor currents flow at higher loads. However, anomaly detection should not classify load changes as anomalies, but only as motor anomalies. Therefore, the current values ​​must be "normalized."

[0051] Instead, according to the invention, features of the distribution of spectral amplitudes in one or more spectral types are used as features, which features:

[0052] i) as load-independent as possible, but its correlation with the fault class has been established in laboratory tests - e.g. the phase angle φ between the current components ab (f) The third winding harmonic.

[0053] In addition, alternative features include, for example:

[0054] ii) Fault frequencies and their spectral amplitudes / phases that can be calculated by one or more spectral categories traditionally using the MCSA formula.

[0055] iii) The spectral magnitude / phase of a spectral class can be determined from the spectrum at a selected frequency resolution.

[0056] The present invention circumvents the problems of "traditional" MCSA, which are described in points a) to d) below:

[0057] a) Resolution and data length / recording time: For MCSA, a good frequency resolution is required, but this leads to excessively long recording times during which the system must remain "stationary", resulting in excessively large eigenvectors, which can quickly lead to the "Curse of Dimensionality".

[0058] b) The fault frequency of traditional MCSA is insufficient to detect existing faults in general applications.

[0059] c) The current amplitude may be disturbed by voltage artifacts.

[0060] d) The spectrum types of current spectrum and MCSA spectrum are not optimally suitable for detecting all faults; in some cases, faults can be more clearly detected in other spectrum types.

[0061] The invention utilizes spectrum types, in particular phase relations, and a combination of slip-independent WH-based features with features calculated according to the MCSA formula (WH=winding harmonics).

[0062] The method according to the invention: a) avoids the need for too high frequency resolution for the feature (WH is slip independent), and b) minimizes amplitude effects by using phase relationships, e.g. the phase angle φ between the current components ab (f, t): If the voltage artifact is mapped on the current component I a and I b , then it should be separated out.

[0063] The typical time resolution according to the present invention is 1 second; the characteristic spectrum therefore has a frequency resolution of 1 Hz. A constant speed must exist within this time interval; sufficiently static operation exists if at least one of the following three conditions (i) to (iii) is met, which can be determined from the acquired data:

[0064] (i) This condition is considered to be satisfied if the power factor cosφ of the motor does not vary too strongly during this time; for example, if the power factor cosφ of the motor varies by less than three times the standard deviation of the power factor.

[0065] (ii) If the slip line, i.e., the spectral line affected by slip, such as PSH, ECC, BB, has a small width; for example, if the width of the slip line is less than 3 times the frequency resolution, then the condition is considered to be met.

[0066] (iii) If the variance of the slip curve is within df = 1 Hz. Variance is a measure of the dispersion of the values ​​about the mean. Variance is a quadratic function of the standard deviation. Variance is calculated by dividing the sum of the quadratic deviations of all measurements from the arithmetic mean by the number of measurements.

[0067] For the anomaly detector feature according to i), the frequency resolution can be higher because the WH frequency position does not change with slip, i.e. more dynamic drives can be classified and significantly less computational effort is required. Evaluating only the WH frequency can also be advantageous for one or more other spectrum types.

[0068] According to i), the evaluation of the characteristics of the algorithm, e.g. the characteristics of the anomaly detector or the classification according to the WH frequency (in any spectral type or a combination thereof), is based on the following knowledge: According to conventional MCSA, slip-related faults in air gap variations manifest themselves at the rotor's rotational frequency (German: Rotorpassierfrequenz) induced stator currents and also superimposed at multiples of the network frequency (WH frequency).

[0069] Advantageous embodiments and developments of the invention are specified in the dependent claims.

[0070] According to a preferred design of the present invention, the amplitude of the calculated spectral variable is normalized to the amplitude at the power supply frequency.

[0071] According to a preferred embodiment of the invention, the spectral variables are determined with a frequency resolution preselected to be 1 Hz in the range of approximately 0.01 to 5 Hz and a time resolution of approximately 1 s.

[0072] According to a preferred embodiment of the present invention, after determining the slip s from the PSH frequency, the amplitude of the current and / or spectral variables at the theoretical frequency is determined, given the known number of rotor bars R and using the determined amplitudes as a feature set. This type of feature is described as an MCSA feature set of the determined spectrum type.

[0073] According to a preferred design of the present invention, the amplitude of the current and / or spectral variable is used as a feature set for all frequencies. The term "all frequencies" here may include so-called "MAX binning" to achieve a reduction in the number of frequencies. MAX binning is performed with a bin length N. bin Execution, where N FFT FFT of frequency N bin Among the samples, the maximum value is set to the Nth bin The new amplitude of each frequency bin is obtained. This can also be performed by overlapping the bins. The method according to the invention significantly reduces the dimensionality of the feature space by binning.

[0074] According to a preferred design, only the features that are significantly different from the corresponding background noise are represented with the measured amplitude. Here, by so-called significance analysis, the environment of the maximum value in the frequency band can be studied, and for example, the standard deviation s of the measured value relative to the detected peak value can be determined, wherein the peak value thus determined must be greater than the environment by more than N xs in order to be considered a significant peak; here, N is an integer in the value range [1, ..., 6]. Therefore, the environment can be set to a fixed minimum value, which is based on the noise limit of the measurement detector signal chain, for example -100 dB. As a result, noise values ​​that vary at insignificant frequencies are not classified as "abnormal" values. However, newly appearing peaks that are different from the noise are regarded as new frequency peaks and are therefore used for fault detection. The method according to the present invention significantly reduces the dimensionality of the feature space by significance analysis. In particular, significance analysis prevents false alarms of anomaly detection, which can be caused by noise values ​​at insignificant frequencies that vary relative to the learned state.

[0075] According to a preferred design of the present invention, the winding harmonics (i.e. the power supply frequency f supply The amplitude of the harmonics of the fundamental frequency (of the harmonics) can be calculated individually or in multiples of three (3n), positive (3n+1), or negative (3n-1) from the harmonics of the fundamental frequency. Alternatively, the distortion coefficient of the harmonics can also be determined relative to the fundamental frequency of the supply voltage. The advantage of the WH amplitude normalized to the fundamental frequency is that it is independent of the slip in terms of frequency position. Therefore, based on these characteristics, fault detection or abnormal deviations can be performed essentially independently of the operating point.

[0076] The present invention is described below with reference to the accompanying drawings, each of which is schematic and not drawn to scale.

[0077] Figure 1 A first design of a device for drive train condition monitoring according to the invention is shown;

[0078] Figure 2 A second design of the device for drive train condition monitoring according to the invention is shown;

[0079] Figure 3A flow chart showing a design of a method for state diagnosis of an asynchronous motor according to the present invention;

[0080] Figure 4 A typical MCSA-based amplitude spectrum is shown;

[0081] Figure 5 The distortion coefficient is shown as a load-dependent characteristic;

[0082] Figure 6 shows the third winding harmonic as a function of torque;

[0083] Figure 7 The positive, negative and triple WH amplitude diagrams for current at different load stages are shown;

[0084] Figure 8A and 8B An example evaluation of two different asynchronous motors is shown;

[0085] Figure 9A 、 9B and 9C show the confidence matrix of an example supervised classifier;

[0086] Figure 10A shows the current amplitude spectrum type for different shaft alignment settings;

[0087] Figure 10B Shows the type of admittance spectrum for different shaft alignment settings;

[0088] Figure 10C The phase angle spectrum type for different shaft alignment settings is shown;

[0089] Figure 11 shows example anomaly detection results for detecting false alignments;

[0090] Figure 12 A type of phase angle spectrum between the three WH current components is shown.

[0091] Figure 1 The device is shown with an asynchronous motor M, which is electrically connected via a supply line 11 to a voltage source 10, which provides a voltage with a frequency f supply Supply voltage U supply The current I flowing through the power supply line 11 is measured by a current sensor 14, such as a shunt, a current transformer or a Hall effect sensor. The voltage U applied to the power supply line 11 is measured by a voltage sensor 15, such as a shunt.

[0092] The current and voltage values ​​are scanned at frequency f A [1 / T] Synchronize with duration T A [T] records. Therefore, the number of sample values ​​is NA =f A *T A For example, when f A =3200Hz and T A = 1.28 seconds, the current and voltage are N A =4096=2 12 The current and voltage measurement values ​​recorded by the sensors 14 , 15 are transmitted by the sensors 14 , 15 to a calculation unit 16 .

[0093] The method steps of calculating the current spectrum using FFT, calculating one or more time-dependent spectral variables, analyzing the calculated spectral variables, and generating a message for reporting an anomaly are performed by a computing unit 16. To this end, computing unit 16 includes a processor 18 and a data memory 20. Data memory 20 stores software containing algorithms for performing the method steps. This software is executed by processor 18. Input values ​​can be transmitted to computing unit 16 via an input / output unit 22 connected to the computing unit (e.g., a PC). Upon detecting an anomaly, computing unit 16 transmits a corresponding message to input / output unit 22.

[0094] Figure 2 An alternative arrangement is shown in which an asynchronous motor M is configured as the driving machine of a drive train D, which also includes a mechanical transmission G and a working machine W. The torque provided by the asynchronous motor M is transmitted to the working machine W via the transmission G. The working machine W can be a conveyor belt, rollers, or drums. In addition to the asynchronous motor M of the drive train D, other electrical loads 24 are also connected to the power supply network N.

[0095] Figure 3 A flow chart showing one embodiment of the method according to the invention for the condition diagnosis of an asynchronous motor to which a three-phase voltage U is applied is shown. R,S,T (t), with supply frequency fU.

[0096] If T current values ​​I of the current I flowing through the motor M are detected t , which flows through the asynchronous motor M according to the supply voltage U applied to the asynchronous motor M. Therefore, T current values ​​I t The time series I(1…T) during which the asynchronous motor M is in a good state. During the recording phase, the current value I describing the “good state” of the motor is recorded. t These current values ​​I t This is then used to train anomaly detectors. The following aspects are crucial:

[0097] In a first step 310, the voltage U R,S,T (t) Voltage time series U t and due to the voltage U applied to the asynchronous motor MR,S,T (t) and the three-phase motor current I flowing through the asynchronous motor M R,S,T (t) Current time series I t .

[0098] In the second step 320, the time series I of the detection is obtained by FFT. t ,U t Calculate the current spectrum I(f,t) and voltage spectrum U(f,t).

[0099] In a third step 330 , one or more of the following time-dependent spectral variables 330 . 1 to 330 . 6 are calculated:

[0100] 330.1) Calculate the current component I a (f,t),I b (f,t), where I a (f,t) and I b (f,t) is obtained by applying Park transform from I R,S,T form;

[0101] 330.2) Calculate the total current amplitude A I (f,t)=√((I a (f,t)) 2 +(I b (f,t)) 2 );

[0102] 330.3) Calculate the phase angle φ between the current components ab :φ ab (f,t)=arctan(I a / I b );

[0103] 330.4) Calculate the phase angle φ between the current and voltage components UI :φ UI (f,t)=arctan(U(f,t) / I(f,t));

[0104] 330.5) Calculate the instantaneous spectral power P(f,t)=U(f,t)*I(f,t);

[0105] 330.6) Calculate the admittance X(f,t) = I(f,t) / U(f,t).

[0106] In a fourth step 340, the calculated spectral variables are analyzed for anomalies. The fourth step 340, "analyzing the calculated spectral variables," may include the following steps: forming a feature vector from one or more time-dependent spectral variables; and using the formed feature vector as an input value of an ML algorithm, i.e., as training data or test data.

[0107] In the fifth step 350, if an abnormality is confirmed, the abnormality is reported.

[0108] Figure 4 A typical MCSA-based amplitude spectrum (amplitude as a function of frequency) of a current signal is shown before (indicated by crosses) and after (indicated by solid lines) significance analysis. Here, the amplitude of non-significant frequencies is set to 10 in order to reduce the noise of the FFT feature. -12 .

[0109] In addition, the so-called winding harmonics (i.e. the power supply frequency f supply The amplitude of the harmonics of the fundamental frequency (WH) can be determined as a characteristic. This can be calculated individually or as tripled WH (3n), positive WH (3n+1), or negative WH (3n-1) harmonics of the fundamental frequency. Alternatively, the distortion coefficient of the harmonics can be determined relative to the fundamental frequency of the supply voltage. The WH amplitude normalized to the fundamental frequency has the advantage that it is independent of the slip in terms of frequency position. Therefore, these characteristics can be used to detect faults or abnormal deviations largely independently of the operating point.

[0110] Figure 5 The distortion coefficient (THD) is shown as a load-dependent characteristic as a function of torque for a well-aligned drive train ("Healthy") and two increasingly severe misalignments ("Misalignment 1" and "Misalignment 2"). The distortion coefficient and THD represent the relationship of unwanted harmonics / nonlinear distortion to the original signal. The THD states are distinguishable because the distribution width or variance for each torque is smaller than the distance between adjacent states, but it is still load-dependent; a rising separation line can be drawn.

[0111] Figure 6 The third winding harmonic is shown as a function of torque. It can be seen that the phase angle φ between the current components ab In the data, starting from approximately 10 Nm, the third winding harmonics act as a largely load-independent characteristic (with only slight variations relative to the torque (load)) for the well-aligned drive train "Healthy" and two increasingly severe misalignments, "Misalignment 1" and "Misalignment 2," as a function of torque. In particular, the value of the severe misalignment "Misalignment 2" is higher than that of the other two states starting from approximately 10 Nm, but does not rise significantly with load level. Therefore, the third winding harmonics characteristic can be used as a distinguishing characteristic regardless of the exact load.

[0112] Figure 7A graph shows the positive, negative, and triple WH current amplitudes for seven different load stages, where the load is first stepped up to the nearest maximum stage and then stepped down to the nearest minimum stage. Curve A0 depicts the amplitude at a grid frequency of 50 Hz. The graph shows that, in particular, the triple WH amplitude is essentially independent of the load. Consequently, variables derived from the triple WH amplitude also exhibit little load dependence, making them useful as features for fault detection without prior determination of the exact load state.

[0113] Figure 8A and Figure 8B Shown is an example evaluation of two different asynchronous motors, which are connected to a load via a coupling. Figure 8A It involves an asynchronous motor with the number of pole pairs p=1 and the number of rotor bars R=20. Figure 8B This involves an asynchronous motor with a pole pair number p = 2 and a rotor bar number R = 28. A comparison is considered for a good state with perfect alignment (OK) and two states with increasingly severe misalignment (ALGN1, ALGN2). The corresponding current spectra (samples versus frequency) and the resulting confidence matrix (predicted labels versus true labels) are shown for a) current amplitude ("Current") and b) voltage amplitude ("Voltage"), taking into account "labels" (supervised learning). Using the voltage spectrum type, the confidence matrix demonstrates almost perfect classification.

[0114] Figure 9A 、 Figure 9B and Figure 9C Confidence matrices of an exemplary supervised classifier are shown for ball bearings damaged to varying degrees of severity (true and predicted labels, from 0 = OK to 9 = terminal, respectively), for various spectrum types. Figure 9A The confidence matrix related to the current I(f,t) is represented by the spectrum type "FFTI_Spec; FFT current spectrum". Figure 9B By spectrum type "FFTPhaseI α ,I β ;FFT phase I α , I β The spectrum refers to the phase angle φ between the current components. ab The confidence matrix of (f,t). Figure 9C The confidence matrix for using “all” available spectrum types 1) to 6) is indicated by the spectrum type “Best Accuracy Spec”, as listed below:

[0115] 1) Current component I a (f,t),I b (f,t), where I a(f,t) and I b (f,t) is obtained by applying Park transform from I R,S,T form;

[0116] 2) Total current amplitude AI(f,t)=√((I a (f,t)) 2 +(I b (f,t)) 2 );

[0117] 3) Phase angle φ between current components ab :φ ab (f,t)=arctan(I a / I b );

[0118] 4) Phase angle φ between current and voltage components UI :φ UI (f,t)=arctan(U(f,t) / I(f,t));

[0119] 5) Instantaneous spectral power P(f,t)=U(f,t)*I(f,t);

[0120] 6) Admittance X(f,t)=I(f,t) / U(f,t).

[0121] From 9A to 9C, the separability is gradually improved, and the overall classification accuracy increases from 85.1% to 93.6% or 95.9%.

[0122] Figure 10A 、 Figure 10B and Figure 10C Spectrum type showing the shaft alignment status (OK = good shaft alignment = "good status"; ALGN = incorrect alignment) for various settings. Figure 10A A spectrum type showing the current amplitude I(f,t). Figure 10A Shows the spectrum type of admittance X(f,t). Figure 10A Show phase angle φ ab (f, t) spectrum type. Within each state, the load is increased in steps from 0% to 120% of the rated load. The main fault frequency occurs at approximately 175 Hz (slip-related) – see the edge region. In state OK1, a voltage disturbance with a 10 Hz frequency is superimposed. In the X and Phase spectrum types, this voltage disturbance is suppressed and barely measurable or perceptible.

[0123] A particular advantage is that at low loads (left side of each state block), better separability is achieved. At higher loads, the 175 Hz interference line gradually deviates from the slip-affected fault frequency. Advantageously, higher frequency resolution can be selected for the X and Phase spectrum types, and since nothing is measurable, the anomaly detector does not learn the interference line as a false attribute of a good state.

[0124] Figure 11 shows example anomaly detection results for detecting misalignment (top: data scenario = 2, with misalignment state, see Figure 8A ; Figure below: Data scenario = 3, with misalignment state, see Figure 8B ), where only the WH current amplitude is considered during the learning and detection phase ("WH only"). Advantageously, the spectral lines to be evaluated have no slip dependency. This allows for higher frequency resolution, classification of more dynamic drives, and significantly reduced computational effort. Evaluating only the WH frequency can also be advantageous for one or more other spectral types.

[0125] Figure 12 The phase angle φ between the three WH current components for 450 Hz is shown ab Spectral types of (f, t) for broken bar scenarios in the following states: OK, BB1, BB1.5, and BB2 (OK = Good State; BB = Broken Bar; BB1 = One rotor bar broken in the motor; BB1.5 = One rotor bar broken in the motor; BB2 = Two rotor bars broken in the motor). Within each state, the motor load increases from left to right. The rotor ratio, which varies only electrically, results in a significant shift in the phase of the three-way WH, from approximately 50° to 25°, due to one or two broken rotor bars.

Claims

1. A method for diagnosing the state of an asynchronous motor (M) to which a power supply frequency (f U ) of the three-phase voltage U R,S,T (t), the method comprises the following steps: a) Detection voltage U R,S,T (t) Voltage time series (U t ) and due to the voltage U applied to the asynchronous motor (M) R,S,T (t) and the three-phase motor current I flowing through the asynchronous motor (M) R,S,T (t) the current time series (I t ); b) Using FFT based time series detection (I t ,U t ) Calculate the current spectrum I(f,t) and voltage spectrum U(f,t); c) Calculate one or more of the following time-dependent spectral variables 1–6: 1) Current component I a (f,t),I b (f,t), where I a (f,t) and I b (f,t) is transformed by Park from I R,S,T form; 2) Total current amplitude A I (f,t)=√((I a (f,t)) 2 +(I b (f,t)) 2 ); 3) Phase angle φ between current components ab (f,t)=arctan(I a / I b ); 4) Phase angle φ between current and voltage components UI (f,t)=arctan(U(f,t) / I(f,t)); 5) Instantaneous spectral power P(f,t)=U(f,t)*I(f,t); 6) Admittance X(f,t)=I(f,t) / U(f,t), d) spectral variables calculated for anomaly analysis; and e) Report any anomalies when they are identified.

2. The method according to claim 1, comprising the steps of: The amplitudes of the calculated spectral variables are normalized to the amplitudes at the power supply frequency.

3. The method according to any one of the preceding claims, comprising the steps of: The spectral variables are determined with a frequency resolution in the range of approximately 0.01 to 5 Hz, preferably 1 Hz, and a time resolution of approximately 1 s.

4. The method according to any one of the preceding claims, comprising the steps of: After the slip s has been determined from the PSH frequency, the amplitudes of the current and / or spectral variables at the target frequency are determined, knowing the number of rotor bars R and using the determined amplitudes as a characteristic set.

5. The method according to any one of the preceding claims, comprising the steps of: The amplitude of the current and / or the spectral variable is used as a characteristic set for all frequencies.

6. The method according to any one of the preceding claims, comprising the steps of: Only the amplitudes of features and measurements that are significantly different from the corresponding background noise are indicated.

7. The method according to any one of the preceding claims, comprising the steps of: The amplitudes of the winding harmonics, ie the harmonics of the supply frequency, are determined as a characteristic group.

Images