A motor fault assessment method based on the current fingerprint algorithm

Abstract

The present invention discloses a method for evaluating motor faults based on the current fingerprint algorithm. The motor fault type, the number of fault warnings and the number of alarms are calculated through the current fingerprint. At the same time, the current fingerprint algorithm is used to obtain the measure characteristics for SPC analysis, and the comprehensive SPC is obtained. The motor health score is calculated by combining the comprehensive SPC score, the number of fault alarms and the number of fault warnings to evaluate the degree of motor faults. The method of the present invention is applicable to detecting various motor faults such as insulation failure, rotor bar breakage, air gap eccentricity and bearing faults, and has the advantages of high detection efficiency, good accuracy, wide application range and easy implementation.

Description

Technical Field

[0001] The present invention belongs to the technical field of motor fault detection, and particularly relates to a method for evaluating motor faults based on a current fingerprint algorithm. Background Art

[0002] As a key device in industrial production, the normal operation of a motor has an important impact on production efficiency and product quality. However, various faults may occur during the operation of the motor, such as insulation failure, rotor bar breakage, air-gap eccentricity, bearing faults, etc. These faults will lead to a decline in motor performance and even damage to the equipment.

[0003] Motor fault assessment refers to the process of analyzing, evaluating, and diagnosing the operating state, performance indicators, etc. of a motor. By implementing the assessment of the motor health, faults and potential hazards of the motor can be detected and eliminated at an early stage. Traditional motor fault assessment methods mainly rely on manual experience and have problems such as poor accuracy; moreover, it is difficult to detect electrical faults of the motor through traditional assessment methods based on vibration sensors. Therefore, there is an urgent need to provide an effective method for assessing electrical and mechanical faults of the motor, which is of great significance for ensuring the normal operation of the motor. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for evaluating motor faults based on a current fingerprint algorithm to solve the problems raised in the background art.

[0005] To achieve the above purpose, on the one hand, the present invention provides a method for evaluating motor faults based on a current fingerprint algorithm, and the method includes the following steps:

[0006] Step 1, collect three-phase current signal data and calculate multiple first measure features of the current, including the eccentricity obtained through Park transformation;

[0007] Step 2, calculate different motor fault frequencies by configuring motor parameters, perform spectral analysis on the three-phase current signal data to obtain a current spectrum diagram, and obtain the alarm numbers and warning numbers of different motor fault types based on the motor fault frequencies;

[0008] Step 3, establish a statistical model of the current signal through the SPC statistical method, and calculate the control charts of each first measure feature of the current signal respectively to monitor abnormal points;

[0009] Step 4, calculate the comprehensive SPC score based on the control chart, and calculate the health score of the motor by combining the alarm number and the warning number;

[0010] In this step 1, the current signals of the d-axis and q-axis of the motor are obtained by performing Park transformation on the three-phase current signal data, the mapping curve between the two is drawn and elliptical fitting is performed on the mapping curve, and the eccentricity of the ellipse is calculated.

[0011] Preferably, in the step 1, the multiple first measurement features further include current value, phase value, unbalance degree, and roughness.

[0012] Preferably, the step 2 includes the following steps:

[0013] Step 2.1, calculate the motor fault frequency by configuring the motor parameters and combining the fault frequency calculation methods for different motor fault types.

[0014] Step 2.2, find the frequencies in the current spectrum that are consistent with the motor fault frequencies, and find the peak closest to this frequency in the current spectrum.

[0015] Step 2.3, compare the peak value of this peak with the preset alarm threshold and preset warning threshold for the corresponding fault to obtain the alarm number and warning number.

[0016] Preferably, the motor parameters include the actual speed of the motor. The vibration spectrum is obtained by collecting vibration signal data and performing spectrum analysis. The frequency where the highest peak in the current spectrum diagram is located is used as the actual current frequency. The speed is calculated using the speed calculation formula based on the actual current frequency. Peaks in the vibration spectrum are searched within the rotation frequency range of the speed, and the actual speed is obtained according to the frequency where the peak in the vibration spectrum is located.

[0017] Preferably, the statistical models of the current signal established by the SPC statistical method are the measure mean calculation model and the standard deviation calculation model for a certain period of history.

[0018] Preferably, in the control chart for calculating each first measurement feature of the current signal, the warning value for calculating the measurement feature is the sum of the measure value mean and 3 times the measure value standard deviation, and the alarm value is the sum of the measure value mean and 6 times the measure value standard deviation.

[0019] Preferably, in calculating the comprehensive SPC score based on the control chart, the SPC scores of each measurement feature are obtained by using a preset SPC score conversion strategy based on the measure value, and the SPC scores of each measurement feature are fused to obtain the comprehensive SPC score.

[0020] Preferably, the SPC score conversion strategy based on the measure value is that when the current measure value is greater than or equal to the corresponding alarm value, the SPC score is set to 0; when the current measure value is less than the corresponding warning value, the SPC score is set to 100; and when the current measure value is between the warning value and the alarm value, the SPC score is calculated by interpolation.

[0021] Preferably, before executing the step 2, the signal-to-noise ratio is calculated using the three-phase current signal data, and it is determined whether the signal-to-noise ratio is greater than a preset threshold. If so, step 2 is executed; otherwise, it ends.

[0022] On the other hand, the present invention provides another motor fault assessment method based on the current fingerprint algorithm, and the method includes the following steps:

[0023] Step 1, collect three-phase current signal data, and calculate multiple first measure features of the current, including the eccentricity obtained through Park transformation;

[0024] Step 2, extract multiple second measure features from the three-phase current signal data, use a fault detection model based on the second measure feature matrix to identify the fault type, set both the fault alarm number and the fault warning number to 0 when the identification result is normal, and obtain the fault alarm number and the fault warning number by using the eccentricity when the fault type is identified;

[0025] Step 3, establish a statistical model of the current signal through the SPC statistical method, and calculate the control charts of the first measure features of the current signal respectively to monitor abnormal points;

[0026] Step 4, calculate the comprehensive SPC score based on the control chart, and calculate the health score of the motor by combining the alarm number and the warning number;

[0027] In this step 1, the current signals of the d-axis and q-axis of the motor are obtained by Park transformation of the three-phase current signal data, the mapping curve of the two is drawn and the mapping curve is ellipse-fitted, and the eccentricity of the ellipse is calculated.

[0028] Compared with the prior art, the beneficial effects of the present invention are:

[0029] The method of the present invention combines self-learning trend analysis, enhances the adaptability of the system to the performance change of the motor and the prediction accuracy, can provide real-time and accurate motor status monitoring and fault prediction, reduce unplanned shutdowns, and has high detection efficiency; and the method of the present invention is applicable to detecting various motor faults such as insulation failure, rotor bar breakage, air gap eccentricity, and bearing faults, and has a wide application range. Description of the Drawings

[0030] Figure 1 It is the flowchart of Embodiment 1 of the present invention.

[0031] Figure 2 It is the calculation flowchart of the eccentricity in the present invention.

[0032] Figure 3 It is the flowchart of Step 3 in this Embodiment 1.

[0033] Figure 4 It is the flowchart of Embodiment 2 of the present invention. Detailed Embodiments

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the protection scope of the present invention.

[0035] Embodiment 1:

[0036] Referring to Figure 1 As shown, a method for motor fault assessment based on the current fingerprint algorithm specifically includes the following five steps.

[0037] Step 1: Collect three-phase current signal data, and calculate multiple first measure features of the current signal based on the three-phase current signal data, namely the current value, phase value, unbalance degree, non-smoothness of the current signal, and eccentricity obtained based on the Park transformation.

[0038] In the present invention, the current signals of three-phase electricity are collected through current transformers. In this embodiment, the sampling frequency is 4000 Hz, and the sampling duration is 10 seconds. In this way, the frequency response range that can be obtained through frequency analysis is 2000 Hz, and the spectral resolution is 0.1 Hz, which can meet the detection requirements of most faults.

[0039] Here, the calculation of the current value, phase value, unbalance degree, and non-smoothness of the current signal is a conventional technical means in the art and will not be elaborated here.

[0040] Combined with Figure 2 As shown, the specific method for obtaining the eccentricity is as follows: Use the Park transformation to convert the current signals of the three phases ABC of the current into the d-axis and q-axis signals. For the corresponding d-axis and q-axis signal data at the same moment, then use the d-axis and q-axis signals at multiple different moments to draw a curve, and fit the curve with an ellipse to calculate the eccentricity of the ellipse. Here, the smaller the eccentricity of the ellipse, the healthier the motor, and vice versa, the more serious the fault.

[0041] Step 2: Calculate different motor fault frequencies by configuring motor parameters, perform spectral analysis on the three-phase current signal data to obtain a current spectrogram, and obtain the alarm numbers and warning numbers of different motor fault types based on the motor fault frequencies.

[0042] In this embodiment, the motor fault types include electrical faults and mechanical faults. Electrical faults include rotor bar breakage faults, air gap eccentricity faults, short-turn faults of low-voltage stator windings, etc. Mechanical faults include ball faults, outer ring faults, inner ring faults of bearing faults, etc. Different fault types have different fault frequency calculation methods:

[0043] ① The calculation method of the rotor bar breakage fault frequency is shown in Equation 1,

[0044] f sb = f 1 (1 ± 2s)(1)

[0045] In Equation 1, s is the unit rotor slip, which is the difference between the synchronous speed of the stator rotating magnetic field of the asynchronous motor and the rotor speed, and f1 represents the base power supply frequency;

[0046] ② The calculation method of the air-gap eccentricity fault frequency is shown in Equation 2,

[0047] f ec = f 1 (R s (1 - s) / p ± η ws ) ± f 1 ((1 - s) / p)(2)

[0048] In Equation 2, f 1 represents the base power supply frequency, R s is the number of rotor slots, which is the same as the number of rotor bars, η ws is a positive integer, representing the fundamental component waveform in the magnetomotive force. For example, when taking the integer 3, it represents the third harmonic. s is the unit rotor slip, and p is the number of pole pairs;

[0049] ③ The calculation method of the short-turn fault frequency of the low-voltage stator winding is shown in Equation 3,

[0050] f st = f 1 [(n / p)(1 - s) ± k](3)

[0051] In Equation 3, f 1 represents the base power supply frequency, n is a positive integer of 1, 2, 3..., k is an odd number of 1, 3, 5..., p is the number of pole pairs, and s is the unit rotor slip;

[0052] ④ The calculation methods of the ball fault frequency, outer race fault frequency, and inner race fault frequency of the bearing fault are shown in Equations (4), (5), and (6) respectively,

[0053] f rb = (D pit / D ball )f rm [1 - (D ball / D pit cosβ)] 2 (4)

[0054] f ro = (n / 2)f rm [1 - (D ball / D pit cosβ)](5)

[0055] f ri = (n / 2)f rm [1 + (D ball / D pit cosβ)] (6)

[0056] where f rm represents that the rotational speed frequency of the rotor is equal to the number of revolutions per minute divided by 60, n is the number of balls, β is the contact angle, D pit is the pitch diameter, and D ball is the ball diameter.

[0057] Referring to Figure 3 shown in the figure, based on the above fault frequency calculation formula, step 2 specifically includes the following steps:

[0058] Step 2.1, calculate the motor fault frequency by configuring motor parameters such as rated power, rated speed, number of motor poles, number of rotor bars, rotor slip, etc., and combining the fault frequency calculation methods for different motor fault types.

[0059] Step 2.2, search for the frequencies in the current spectrum that are the same as the motor fault frequencies calculated in step 2.1, and search for the peak closest to this frequency in the current spectrum;

[0060] Step 2.3, compare the peak value of this peak with the preset alarm threshold and preset warning threshold for the corresponding fault, and finally obtain the alarm number and warning number.

[0061] In step 2.2 of the present invention, when a frequency identical to one of the motor fault frequencies is found in the current spectrum, it indicates that the motor has the corresponding motor fault. The same motor fault frequencies found in the current spectrum can be one or multiple. According to the actual situation, it is necessary to make judgments on alarm and warning for each motor fault frequency found in the current spectrum in step 2.3. When the nearest peak of the motor fault frequency is within the range greater than the warning threshold and less than the alarm threshold, it is a warning, and when it exceeds the alarm threshold, it is an alarm, so as to obtain the warning number and alarm number; it should be noted that for different types of motor faults, their alarm thresholds and warning thresholds are also different; the warning number and alarm number in step 2.3 are respectively the sum of the warning numbers and the sum of the alarm numbers for different types of motor faults.

[0062] In the present invention, it can be known from the above calculation formula about the motor fault frequency that the motor rotor slip is a key factor affecting the fault frequency. The motor rotor slip, that is, the current frequency divided by the number of motor poles, is inconsistent with the actual motor speed frequency. The motor rotor slip is obtained by calculating the actual motor speed. If the measurement of the actual motor speed is inaccurate, it will lead to inaccurate detection of the current fault frequency. As a preferred embodiment of the present invention, the vibration data is obtained by the vibration sensor of the same unit to extract the precise speed of the motor. Specifically, the frequency of the highest peak in the current spectrum diagram is taken as the actual current frequency, and the speed is calculated using the speed calculation formula based on the actual current frequency (synchronous speed = 120*actual current frequency / number of motor poles). The peak of the vibration spectrum is found within the speed frequency range, and the actual speed is obtained according to the frequency of the peak of the vibration spectrum. Here, the speed range is obtained by setting the range threshold o of the speed frequency. For example, the range threshold o of the speed frequency is set to 5Hz, and the speed range is the speed ±5Hz.

[0063] Step 3: Establish a statistical model of the current signal by using the SPC statistical method, and calculate the control chart of each first measurement feature of the current signal to monitor abnormal points.

[0064] SPC refers to statistical process control, which is a process control tool that uses mathematical statistics. Control charts are also called "quality management charts" and "quality assessment charts". They are quality management charts with control limits that are used to analyze and determine whether a process is in a stable state based on mathematical statistics principles.

[0065] In the present invention, step 3 specifically includes the following sub-steps:

[0066] Step 3.1, establish a statistical model, including a mean calculation model and a standard deviation calculation model for measurement values ​​over a period of history;

[0067] Step 3.2, using the mean calculation model and the standard deviation calculation model to respectively calculate the early warning value and the alarm value of the current measurement value of each first measurement feature, the early warning value is the sum of the mean value of the measurement value and 3 times the standard deviation of the measurement value, and the alarm value is the sum of the mean value of the measurement value and 6 times the standard deviation of the measurement value;

[0068] Step 3.3: Perform abnormal monitoring on the current measurement value of the first measurement feature based on the early warning value and the alarm value.

[0069] In the present invention, since the measured values of each measurement feature based on the time series are different, the mean values and standard deviations of the measured values calculated by different measurement features are not necessarily the same, and each measurement feature has its corresponding warning value and alarm value. A corresponding control chart is set for each measurement feature. The warning limit and alarm limit are described in the control chart. The warning limit and alarm limit can be a fluctuating curve based on time variation or a horizontal line. The line type of the warning limit and alarm limit is determined by the specific setting for a certain period of history in the statistical model. When the certain period of history is variable, the line type is a fluctuating curve; otherwise, it is a horizontal line.

[0070] Step 4: Calculate the comprehensive SPC score based on the control chart, and calculate the health score of the motor in combination with the number of fault warnings and the number of fault alarms.

[0071] In the present invention, the SPC scores of each measurement feature are obtained by using a preset SPC score conversion strategy based on the measured values, and the SPC scores of each measurement feature are fused to obtain the comprehensive SPC score. The SPC score conversion strategy based on the measured values is as follows: when the current measured value is greater than or equal to the corresponding alarm value, the SPC score is set to 0; when the current measured value is less than the corresponding warning value, the SPC score is set to 100; and the SPC score when the current measured value is between the warning value and the alarm value is calculated by the interpolation method. As for how to obtain the comprehensive SPC score through data fusion, this is a conventional technical means in the art, and those skilled in the art can set it according to the situation.

[0072] In the present invention, the comprehensive SPC score, the number of fault warnings, and the number of fault alarms are used as parameters for evaluating the health score of the motor to evaluate the degree of motor failure. The lower the health score, the more serious the motor failure. As for how to evaluate the health score of the motor based on the parameters for evaluating the health score of the motor, this is a conventional technical means in the art, such as predicting by setting a deep learning network model or querying the health score of the motor by setting an evaluation table composed of the above three parameters. Those skilled in the art can set it according to the actual situation and will not be elaborated here.

[0073] Further, before performing step 2, calculate the signal-to-noise ratio of the three-phase current signal data, and determine whether the signal-to-noise ratio is greater than a preset threshold. If so, perform step 2; otherwise, end. In the present invention, it is determined whether it is working by calculating the signal-to-noise ratio of the current fingerprint. If it is determined to be working, the subsequent fault detection process is carried out; otherwise, no fault detection is performed.

[0074] Embodiment 2

[0075] Refer to Figure 4 As shown, a method for evaluating motor faults based on the current fingerprint algorithm, which specifically includes the following 5 steps:

[0076] Step 1, collect three-phase current signal data and calculate the first measure features of the current, including current value, phase value, unbalance degree, roughness, and eccentricity obtained through Park transformation;

[0077] Step 2, extract multiple second measure features from the three-phase current signal data, and use a fault detection model based on the second measure feature matrix to identify the fault type; when the identification result is normal, set both the fault alarm count and the fault warning count to 0; when the fault type is identified, compare the eccentricity with the preset alarm threshold and warning threshold to obtain the fault alarm count and the fault warning count;

[0078] Step 3, establish a statistical model of the current signal through the SPC statistical method, and calculate the control charts of the first measure features of the current signal respectively to monitor abnormal points;

[0079] Step 4, calculate the comprehensive SPC score based on the control chart, and calculate the health score of the motor in combination with the fault alarm count and the alarm count.

[0080] In step 2 of the present invention, specifically, extract features from the current signal. The features include maximum value, minimum value, average value, median, peak-to-peak value, average value of absolute values (rectified average value), variance, standard deviation, kurtosis, skewness, root mean square, mean square value, root mean amplitude, waveform factor, peak factor, pulse factor, margin factor, fourth moment, shape factor, center frequency, mean square frequency, root mean square frequency, frequency variance, frequency standard deviation, signal power, signal-to-noise ratio of the signal, signal distortion ratio, spurious-free dynamic range, full power bandwidth, mean value of spectral kurtosis, standard deviation of spectral kurtosis, skewness of spectral kurtosis, kurtosis of spectral kurtosis. These feature measures include time-domain features, frequency-domain features, and time-frequency spectrum features. Then, train an Ada boost multi-classification model with the calibrated data. This model can be used to predict whether there is a fault in the motor and obtain the specific type of fault when there is a fault. These fault types include insulation failure, rotor bar breakage, air gap, bearing fault, etc. This method does not require parameter configuration, and the detection process is relatively simple and fast.

[0081] It should be noted that in this step 2, only one specific fault type can be identified through the Adaboost model. At this time, at least one of the fault alarm count and the fault warning count is 0. For example, when the fault alarm count is 1, the fault warning count must be 0. And different from using the nearest peak of the motor fault frequency in the current spectrum to compare with the preset fault alarm threshold and fault warning threshold to obtain the fault alarm count and the warning count in Embodiment 1, in this embodiment, the eccentricity calculated through Park transformation in step 1 is compared with the preset fault warning threshold and fault alarm threshold to determine the fault alarm count and the warning count.

[0082] In this embodiment, the specific operations in Step 1, Step 3, and Step 4 are the same as those in Embodiment 1, and will not be elaborated here.

[0083] As the second embodiment of the motor fault detection of the present invention, in the case where the motor parameters or the actual motor speed cannot be obtained, a multi-scale detection algorithm in Step 2 is set for fault identification. This method does not require motor configuration parameters and has good robustness and generalization ability.

[0084] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A motor fault assessment method based on current fingerprint algorithm, characterized in that: The method comprises the following steps: Step 1, collecting three-phase current signal data, and calculating multiple first measurement characteristics of the current, including the eccentricity obtained by Park transformation; Step 2, extracting multiple second measurement features from the three-phase current signal data, using a fault detection model based on the second measurement feature matrix to identify the fault type, setting the number of fault warnings and the number of fault alarms to 0 when the identification result is normal, and using the eccentricity to obtain the number of fault warnings and the number of fault alarms when the fault type is identified; Step 3, establishing a statistical model of the current signal by using the SPC statistical method, and calculating the control chart of each first measurement feature of the current signal to monitor abnormal points; Step 4, calculate the comprehensive SPC score based on the control chart, and calculate the health score of the motor based on the number of alarms and warnings; In step 1, the current signals of the motor d-axis and q-axis are obtained by Park transformation of the three-phase current signal data, mapping curves of the two are drawn, and an ellipse is fitted to the mapping curve to calculate the eccentricity of the ellipse.

2. A motor fault assessment method based on current fingerprint algorithm as claimed in claim 1, characterized in that: In step 1, the plurality of first measurement features further include current value, phase value, imbalance, and roughness.

3. A motor fault assessment method based on current fingerprint algorithm as claimed in claim 1, characterized in that: The statistical model of the current signal established by the SPC statistical method includes a measurement mean calculation model and a standard deviation calculation model within a historical period of time.

4. A motor fault assessment method based on current fingerprint algorithm as claimed in claim 1, characterized in that: In the control diagram of each first measurement feature of the calculated current signal, the early warning value of the calculated measurement feature is the sum of the measurement value mean and 3 times the measurement value standard deviation, and the alarm value is the sum of the measurement value mean and 6 times the measurement value standard deviation.

5. A motor fault assessment method based on current fingerprint algorithm as claimed in claim 1, characterized in that: In the calculation of the comprehensive SPC score based on the control chart, the SPC score of each measurement feature is obtained by using a preset SPC score conversion strategy based on the measurement value, and the SPC score of each measurement feature is fused to obtain the comprehensive SPC score.

6. A motor fault assessment method based on current fingerprint algorithm as claimed in claim 5, characterized in that: The SPC score conversion strategy based on the measurement value is to set the SPC score to 0 when the current measurement value is greater than or equal to the corresponding alarm value, set the SPC score to 100 when the current measurement value is less than the corresponding warning value, and calculate the SPC score when the current measurement value is between the warning value and the alarm value by interpolation.

7. A motor fault assessment method based on current fingerprint algorithm as claimed in claim 1, characterized in that: Before executing step 2, the signal-to-noise ratio is calculated using the three-phase current signal data to determine whether the signal-to-noise ratio is greater than a preset threshold. If so, step 2 is executed, otherwise the process ends.

Images