A motor mechanical defect classification diagnosis method based on threshold invariant multi-current features

Abstract

Predicting mechanical defects in electric motors can significantly reduce unplanned production stoppages in industrial manufacturing. This application addresses common mechanical defects in high-power motors, such as motor jamming, overload, stator-rotor misalignment, rotor imbalance, and combined mechanical defects, by providing a cross-condition mechanical defect classification and diagnosis method based on domain-invariant current characteristics. This method includes: in-depth analysis of the formation mechanism of motor mechanical defects; obtaining defect-related features using a mechanical defect type analysis method; comparing the feature values ​​with normal motor thresholds to determine the type of mechanical defect and obtaining a first diagnostic result; using an interval abnormal frequency feature diagnosis method to establish an abnormal point distribution shape fitting curve to determine the type of mechanical defect and obtain a second diagnostic result; using a multi-source adversarial domain adaptive mechanical defect method based on multi-classifier alignment to determine the type of mechanical defect and obtain a third diagnostic result; and fusing the three diagnostic results and establishing a mechanical defect decision-making method based on actual operating conditions.

Description

Technical Field

[0001] This invention belongs to the field of big data and artificial intelligence, and specifically relates to a method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics. Background Technology

[0002] With the continuous development of modern industry, electric motors have become a common and crucial piece of equipment in industrial production. Timely detection of mechanical defects in motors, accurate identification of the type of defect, and timely repair are of great significance for extending the service life of motors and ensuring their normal and stable operation. However, mechanical defects in motors generally occur internally, and these subtle changes pose potential risks to the motor's use, yet workers cannot perceive these subtle changes with the naked eye. If workers can promptly detect mechanical defects in motors, accurately identify the type of defect, and repair them, they can extend the motor's service life and ensure its normal and stable operation. Summary of the Invention

[0003] The purpose of this invention is to provide a method for classifying and diagnosing mechanical defects in motors based on the characteristic of invariant threshold current. This method can promptly detect mechanical defects in motors, accurately determine the type of mechanical defect, and perform repairs to extend the service life of the motor and ensure its normal and stable operation.

[0004] To address the aforementioned technical problems, this invention provides a method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics, comprising:

[0005] For assembly mechanical defects of high-power motors: motor jamming, overload, stator-rotor misalignment, rotor imbalance, and assembly mechanical defects; (high-power motors refer to motors with a power of 200 KW or more)

[0006] Mechanical defect type analysis method is used to obtain mechanical defect characteristics related to mechanical defects, including time domain and frequency domain electrical signal characteristics, as well as amplitude modulation and frequency modulation characteristics of current signals and shaft system characteristics. The characteristic values ​​are compared with the data when the motor is working normally to determine the type of mechanical defect of the motor and obtain the first diagnostic result.

[0007] The interval abnormal frequency feature diagnosis method is used to count the number of abnormal points, establish a fitting curve of the distribution pattern of abnormal points to determine the type of mechanical defect of the motor, and obtain the second diagnostic result.

[0008] A deep learning framework is constructed, and a multi-source adversarial domain adaptive mechanical defect method based on multi-classifier alignment is used to determine the type of mechanical defect in the motor and obtain a third diagnostic result.

[0009] The first, second, and third diagnostic results are combined, and the type of mechanical defect is determined based on the actual working conditions.

[0010] Optionally, the formation mechanism and manifestation characteristics of mechanical defects include:

[0011] Overload can be determined by the magnitude of the current. If the motor is not operating at 60%-80% of its rated load, its efficiency will decrease.

[0012] Misalignment of the stator and rotor, resulting in misalignment of the centers of mass of the stator and rotor, and axial misalignment of the stator and rotor, leads to axial movement and large axial vibration of the motor.

[0013] Rotor imbalance, or uneven mass distribution of the motor rotor, can cause excessive vibration during operation and may also produce abnormal noises.

[0014] Optionally, abnormal features and data are extracted during mechanical defect analysis to obtain a first diagnostic result, including:

[0015] Choose one of the following methods from the mechanical defect type analysis methods: overcurrent comparison method, phase sequence detection method, frequency conversion analysis method, fundamental frequency analysis method, harmonic amplitude comparison method, and upper and lower limit frequency point method. Obtain the mechanical defect-related characteristics, including time-domain and frequency-domain electrical signal characteristics, as well as amplitude modulation and frequency modulation characteristics of current signals and shaft system characteristics. Compare the mechanical defect characteristic values ​​with the data when the motor is operating normally to determine the type of mechanical defect of the motor and obtain the first diagnostic result.

[0016] Optionally, an interval anomaly frequency feature diagnosis method is used to obtain a second diagnostic result, including:

[0017] During different working periods, the amplitude-frequency characteristics of the motor during normal operation are extracted, the normal operating baseline of the motor is calculated, and at least three baselines are merged as the threshold for judging the mechanical defects of the motor. The spectrum of the mechanical defects of the motor is sampled, the sampling results are compared with the baseline, the number of abnormal frequency points is counted, and the abnormal point distribution shape fitting curve is established to judge the mechanical defects of the motor and obtain the second diagnostic result.

[0018] Optionally, deep learning methods are used to diagnose the types of mechanical defects in the motor and obtain a third diagnostic result, including:

[0019] A sub-network is constructed for each source domain, and a feature set is formed based on the domain-invariant features of different mechanical defect types in each source domain. The distribution distance between all target features and source domain features is calculated sequentially. The entropy of the distribution distance vector is minimized so that the sub-network learns features that are more representative of mechanical defects, thereby determining the type of mechanical defect and obtaining the third diagnostic result.

[0020] Optionally, the first diagnostic result, the second diagnostic result, and the third diagnostic result are integrated, and a mechanical defect decision-making method is established based on the actual working conditions, including:

[0021] The first diagnostic result is compared with the second diagnostic result. If the mechanical defect results are consistent, the motor mechanical defect can be determined. If the mechanical defect results are inconsistent, they are compared with the third diagnostic result. If the third diagnostic result is consistent with the first or second diagnostic result, the specific working conditions, such as temperature, humidity, and pressure, are considered to analyze and diagnose the motor mechanical defect. If the first, second, and third diagnostic results are inconsistent, the mechanical defect characteristics of the three are combined, and the combined characteristics are used again to diagnose the motor mechanical defect using deep learning.

[0022] This invention provides a method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics. The method includes using overcurrent comparison, phase sequence detection, frequency conversion analysis, fundamental frequency analysis, harmonic amplitude comparison, and upper / lower limit frequency point methods from mechanical defect type analysis to obtain defect features related to mechanical defects. These features include time-domain and frequency-domain electrical signal characteristics, as well as amplitude modulation and frequency modulation characteristics of voltage and current signals, and shaft system characteristics. The feature values ​​are compared with data from normal motor operation to determine the type of mechanical defect, obtaining a first diagnostic result. A second diagnostic result is obtained by using an interval abnormal frequency feature diagnosis method to count the number of abnormal points and establish an abnormal point distribution shape fitting curve. A third diagnostic result is obtained by constructing a deep learning framework and using a multi-source adversarial domain adaptive mechanical defect method based on multi-classifier alignment. Finally, the first, second, and third diagnostic results are fused, and a mechanical defect decision-making method is established based on actual operating conditions to determine the type of mechanical defect in the motor.

[0023] By diagnosing and integrating mechanical defect types three times, the accuracy of mechanical defect type diagnosis is improved, ensuring timely repair when mechanical defects occur in the motor, which helps to extend the service life of the motor and ensure the normal and stable operation of the motor. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating the method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics, as provided in an embodiment of this application.

[0026] Figure 2 This is a schematic diagram of the overall framework of the interval abnormal frequency feature diagnosis method provided in the embodiments of this application.

[0027] Figure 3 This is a schematic diagram of the overall framework of the multi-source adversarial domain adaptive mechanical defect method based on multi-classifier alignment provided in the embodiments of this application. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics, as provided in an embodiment of this application. The method may include:

[0030] S 11: Mechanical defect characteristics are obtained by using mechanical defect type analysis method, and the type of mechanical defect of motor is determined based on the characteristic value to obtain the first diagnostic result.

[0031] It should be noted that to obtain the time-domain and frequency-domain electrical signal characteristics related to mechanical defects, as well as the amplitude modulation, frequency modulation, and shaft system characteristics of the current signal, multiple methods from mechanical defect type analysis are required to process the original features. These methods include the phase imbalance method, overcurrent comparison method, phase sequence detection method, rotational frequency analysis method, fundamental frequency analysis method, harmonic amplitude comparison method, and upper and lower limit frequency point method. Then, the obtained feature values ​​are compared with the threshold values ​​when the motor is operating normally. Based on the deviation from the threshold values, the type of mechanical defect in the motor is determined, and a preliminary diagnostic result is obtained.

[0032] To determine the type of mechanical defect in a motor using the phase imbalance method: First, calculate the difference between the position of the waveform peaks in each phase voltage and the number of timing points between two peaks. Then, sort the timing positions of the first peak in each of the three phases. Finally, subtract the phase of each peak in each phase from the phase of each peak in the adjacent phase and take the average value to obtain the phase-related average value. Determine whether a three-phase phase imbalance mechanical defect exists based on the obtained average value.

[0033] The overcurrent comparison method is used to determine the type of mechanical defects in a motor: the effective value of the current in each phase is calculated, and the presence of an overcurrent mechanical defect is determined by comparing the effective value of the current measured on any phase line with the rated current value.

[0034] The phase sequence detection method is used to determine the type of mechanical defects in a motor: First, the average value of the original current data for each phase (excluding the DC signal) is subtracted. Then, two current waveforms for each phase are selected using a zero-crossing method and converted into square waves. Finally, the waveform fluctuations of the three-phase power supply are encoded to obtain the phase sequence characteristics. The presence of mechanical defects in the three-phase power supply phase sequence is determined by comparing the phase sequence characteristics with the standard phase sequence.

[0035] To determine the type of mechanical defect in a motor using frequency rotation analysis: First, calculate the threshold line: Sum the 1st and 3rd frequency amplitudes of normal motor data, smooth the sum, and then take the average. Draw a centroid line. Train the smoothed data using a normalized model to obtain a normalized dataset. Establish the alarm threshold line using the average value + 2.5 * standard deviation.

[0036] Then, the theoretical rotation frequency is calculated using the original single-phase current data, i.e., the fundamental frequency / number of pole pairs. The actual rotation frequency is recorded as the frequency with the highest amplitude near the theoretical rotation frequency. Based on the actual rotation frequency, the two rotation frequencies and their amplitudes around the current fundamental frequency are found, i.e., the harmonic amplitude content of the ±1st order rotation frequency (number of pole pairs ±1). The sum of the ±1st order rotation frequency amplitudes is then averaged with the sum of the historical ±1st order amplitudes to obtain the standard value. The standard value is compared with the alarm threshold line to determine whether there is a mechanical defect of rotational imbalance.

[0037] To determine the type of mechanical defect in a motor using fundamental frequency analysis: The original current signal is transformed using a Park vector transform to obtain i_d and i_q. The squares of i_d and i_q are then summed and the square root is taken to generate a new Park dataset. A Fourier transform is performed on this dataset to obtain a frequency-amplitude table. The maximum amplitude value within a given frequency range is then found in the table. The presence of a mechanical defect due to axial misalignment between the stator and rotor is determined based on the maximum amplitude value.

[0038] Determining the type of mechanical defect in a motor using the harmonic amplitude comparison method: First, train the model by repeatedly calculating the amplitude of the Xth harmonic at the rotational frequency (X: number of pole pairs - 1). Take the average of these multiple harmonic amplitudes and then use the average + 6 * ... Mean square error and average Value + 8 *Note the threshold line and outlier threshold line when calculating standard deviation.

[0039] Then, when the motor develops a mechanical defect, the amplitude of the Xth harmonic of the rotational frequency is calculated to obtain the operating harmonic amplitude. Based on the operating harmonic amplitude and the threshold line, it is determined whether there is a mechanical defect of uneven air gap.

[0040] Using the upper and lower frequency point method to determine the type of mechanical defects in motors: RMS demodulation: In order to achieve the purpose of one waveform per window, the data is split according to the parameter window length (data cutting window length: sampling frequency / fundamental frequency) to obtain a series of subsets. The sum of the squares of all subsets is divided by the number of subsets, and then the square root is taken to obtain the demodulated data. The average value of the demodulated data is then subtracted from the demodulated data to remove the DC component.

[0041] First, the theoretical frequency transition is calculated using the fundamental frequency / pole logarithm. Then, the extracted raw feature data is demodulated using RMS. Next, the demodulated dataset is subjected to Fourier transform to obtain Fourier data. Finally, the upper and lower limit frequency points of the specified frequency band are determined.

[0042] Upper limit: Number of pole pairs * Maximum slip * Theoretical speed.

[0043] Lower limit: Number of pole pairs * (fundamental frequency / original data length).

[0044] Find the frequency point with the maximum amplitude in the Fourier data based on the upper and lower limit frequency points, and record the amplitude corresponding to this frequency point as the maximum amplitude. Finally, use the fundamental frequency amplitude of the current divided by the maximum amplitude to obtain the working value. Determine whether there is a mechanical defect such as rotor bar breakage based on the working value.

[0045] S 12: Using the interval abnormal frequency characteristic diagnosis method, the number of abnormal points is counted, a distribution shape fitting curve is established to determine the type of mechanical defect of the motor, and a second diagnostic result is obtained.

[0046] like Figure 2 As shown, Figure 2 This is a schematic diagram of the overall framework of the interval abnormal frequency feature diagnosis method provided in the embodiments of this application. Obtaining a second diagnostic result through this method may include:

[0047] It should be noted that this method primarily targets three-phase current data. The first step is training the model baseline: When the motor is operating normally without mechanical defects, three-phase data is accumulated over a fixed time period. The load and power frequency for this period are calculated using a sliding time window with a certain window width and step size. If the load and power frequency remain relatively constant, the average value and range of the power frequency and load are recorded. The baseline is then calculated using the accumulated data over the fixed time period: First, the data undergoes a Park vector transformation. Then, the data for this fixed time period is divided into multiple data points of equal duration, and windowed Fourier transforms are performed on each. Next, the spectrum data after the Fourier transform is downsampled, using a unit frequency as the window width. The maximum frequency amplitude within each window is taken. Then, the average amplitude and standard deviation at each frequency are calculated. Finally, the average amplitude + 2.5 * standard deviation at each frequency is taken as the baseline. During the first few hours of model operation, the above steps are performed once every hour, for a total of at least three baseline calculations. If the power frequency and load of the three baselines are similar, the baselines are merged; otherwise, the baseline calculation continues until at least three baselines for a given operating condition are obtained and merged.

[0048] Next, data processing is performed when a mechanical fault occurs. Three-phase current data is extracted when the motor experiences a mechanical defect. The data is then subjected to Park vector transformation, dividing the data within a specific timeframe into multiple short data points of equal duration. Each short data point undergoes a windowed Fourier transform, and the resulting spectral data is downsampled, with a window width of unit frequency. The maximum frequency amplitude within each window is then recorded. Next, the number of outliers is counted. The downsampled spectral data is compared to the baseline at each frequency, and the number of short data points of equal duration within multiple timeframes with spectral amplitudes greater than the baseline is counted. If the proportion of spectral amplitudes greater than the baseline exceeds a certain level, that frequency is recorded as an outlier. Then, the number of outliers within a defined interval width of a certain frequency is counted. Finally, a Gaussian function is used to fit the outliers, and the mean, standard deviation, and peak height of the fitting results are used to comprehensively determine the type of mechanical defect, obtaining a second diagnostic result.

[0049] S 13: Use a multi-source adversarial domain adaptive mechanical defect diagnosis method based on multi-classifier alignment to determine the type of mechanical defect in the motor and obtain a third diagnostic result.

[0050] like Figure 3 As shown, Figure 3 This is a schematic diagram of the overall framework of the multi-source adversarial domain adaptive mechanical defect method based on multi-classifier alignment provided in the embodiments of this application. The method may include:

[0051] Set source domain data with labels And unlabeled target domain data and unlabeled target domain data Number of training iterations Batch size Randomly initialize the common feature extractor Private Feature Extractor Classifier Feature Reconstructor (2) Set the number of iterations (3) From the first Source domain datasets Randomly collect a batch of data (4) From the target domain dataset Randomly collect a batch of data (5) Input the source domain samples and target domain samples into the common feature extractor to obtain common features. and (6) Input the common features of the source domain samples into the private feature extractor to obtain the source domain sample features. and calculate (7) The source domain sample features The data is fed into a classifier to obtain the classification result. and calculate (8) The source domain sample features The data is fed into the reconstructor to obtain the reconstructed features. and calculate (9) Input the common features of the target domain samples into the private feature extractor to obtain the target domain sample features. (10) The source domain sample features The data is fed into a classifier to obtain the classification result. and calculate (11) Source domain sample features The data is fed into a discriminator to obtain reconstructed features. And calculate (12) Update the common feature extractor using formula (1). Private Feature Extractor Classifier The parameters and the feature reconstructor updated by formula (2) (13) Repeat (1)-(12) for the parameters, until the program iteration is complete.

[0052]

[0053] It is important to note that when using a multi-classifier aligned, multi-source adversarial domain adaptive mechanical defect diagnosis method to determine the type of mechanical defects in motors, the common feature extractor is shared by all source and target domains, extracting their common features. The private feature extractor, classifier, and reconstructor correspond one-to-one with each source domain. Each batch of source and target domain data from motor operation is input, and this data is processed by the common and private feature extractors to reach the classifier. Multiple classifiers produce consistent predictions for the same target domain. Different source domains are input into a specific classifier and then adversarially challenged against the target domain. The differences obtained from the adversarial comparison determine the type of mechanical defect, resulting in a third diagnostic result. To improve the accuracy of mechanical defect feature identification, this method employs a feature reconstructor to reconstruct features from both the source and target domains, resulting in a more stable decision boundary.

[0054] S 14: Integrate the three diagnostic results and establish a mechanical defect decision-making method based on actual working conditions.

[0055] It should be noted that the three diagnostic results obtained from the same segment of the motor may not be completely consistent. Therefore, fusing these three diagnostic results to obtain a more accurate diagnosis is crucial. This invention provides a fusion method as follows: First, the first and second diagnostic results are compared. If the results are consistent, the defect type can be determined. If the results are inconsistent, the third diagnostic result is referenced. If all three results are inconsistent, the mechanical defect features of the three results can be combined to further utilize deep learning to determine the mechanical defect type. Finally, a decision method is derived from the results.

[0056] This application employs a mechanical defect type analysis method, an interval abnormal frequency feature diagnosis method, and a multi-source adversarial domain adaptive mechanical defect diagnosis method based on multi-classifier alignment to determine the type of mechanical defect in the motor. Finally, the results of these three methods are combined with the operating conditions to obtain the final determination. This multi-layered determination method improves the accuracy of the results. Even when a mechanical defect occurs in the motor and its external manifestations are not obvious, it can accurately analyze the motor's mechanical defects, ensuring timely repair when mechanical defects occur, thus extending the motor's service life and ensuring its normal and stable operation.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that the elements inherent in a process, method, article, or apparatus that includes a list of elements are included. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Additionally, portions of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0058] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

Claims

1. A method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics, characterized in that, include: Addressing assembly mechanical defects in high-power motors: motor jamming, overload, stator-rotor misalignment, rotor imbalance, and assembly mechanical defects; Mechanical defect type analysis method is used to obtain mechanical defect characteristics related to mechanical defects, including time domain and frequency domain electrical signal characteristics, as well as amplitude modulation and frequency modulation characteristics of current signals and shaft system characteristics. The characteristic values ​​are compared with the data when the motor is working normally to determine the type of mechanical defect of the motor and obtain the first diagnostic result. The interval abnormal frequency feature diagnosis method is adopted. First, the model baseline is trained. Three-phase data within a fixed time period are accumulated when the motor is working normally. The load and power frequency of the time period are calculated using a sliding time window with a certain width and step size. If the load and power frequency remain basically unchanged, the average value and range of the power frequency and load are recorded. Park vector transformation is performed on the accumulated data within the fixed time period. Then, the data is divided into multiple data of equal duration, and windowed Fourier transform is performed on each. Next, the transformed spectrum data is downsampled, with the window width being the unit frequency width. The maximum frequency amplitude within each window is taken, and the average value and standard deviation of the frequency amplitude at each frequency are calculated. The average value + 2.5 * standard deviation is taken as the baseline at that frequency. In the first few hours of model operation... At any given time, a baseline calculation is performed every hour to obtain at least three baselines. If the power frequency and load of the three baselines are similar, the baselines are merged; otherwise, the baseline calculation continues until at least three baselines can be merged under a certain operating condition. When a mechanical fault occurs in the motor, its three-phase current data is extracted and subjected to the same Park vector transformation, windowed Fourier transform, and downsampling as above. The downsampled spectrum amplitude data is compared with the corresponding baseline by frequency. If the proportion of spectrum amplitude greater than the baseline is higher than a certain level, the frequency is recorded as an abnormal frequency point. Then, with a certain frequency as the interval width, the number of abnormal points in each interval is counted. Finally, a Gaussian function is used to fit the abnormal points. The mean, standard deviation, and peak height of the fitting results are used to comprehensively judge the type of mechanical defect and obtain the second diagnostic result. A deep learning framework is constructed, and a multi-source adversarial domain adaptive mechanical defect method based on multi-classifier alignment is used to determine the type of mechanical defect in the motor and obtain a third diagnostic result. If the first and second diagnostic results are consistent, the mechanical defects of the motor can be determined. If they are inconsistent, they need to be compared with the third diagnostic result. If all three diagnostic results are inconsistent, the mechanical defect features of the three are combined, and the combined features are used again to diagnose the mechanical defects of the motor using deep learning.

2. The method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics as described in claim 1, characterized in that, The formation mechanism and characteristics of mechanical defects include: Overload can be determined by the magnitude of the current. If the motor is not operating at 60%-80% of its rated load, its efficiency will decrease. Misalignment of the stator and rotor, resulting in misalignment of the centers of mass of the stator and rotor, and axial misalignment of the stator and rotor, leads to axial movement and large axial vibration of the motor. Rotor imbalance, or uneven mass distribution of the motor rotor, causes excessive vibration and even abnormal noise during operation.

3. The method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics as described in claim 1, characterized in that, Extracting abnormal features and data from mechanical defects, and using mechanical defect analysis methods to obtain the first diagnostic result, including: Choose one of the following methods from the mechanical defect type analysis methods: overcurrent comparison method, phase sequence detection method, frequency conversion analysis method, fundamental frequency analysis method, harmonic amplitude comparison method, and upper and lower limit frequency point method. Obtain the mechanical defect-related characteristics, including time-domain and frequency-domain electrical signal characteristics, as well as amplitude modulation and frequency modulation characteristics of current signals and shaft system characteristics. Compare the mechanical defect characteristic values ​​with the data when the motor is operating normally to determine the type of mechanical defect of the motor and obtain the first diagnostic result.

4. The method for classifying and diagnosing mechanical defects in motors based on threshold-invariant multi-current characteristics as described in claim 1, characterized in that, The method of deep learning is used to diagnose the mechanical defect type of motor and obtain a third diagnostic result. This includes: constructing a sub-network for each source domain and forming a feature set based on the domain-invariant features of different mechanical defect types in each source domain; calculating the distribution distance between all target features and source domain features in turn, minimizing the entropy of the distribution distance vector so that the sub-network learns features that are more representative of mechanical defects, thereby determining the mechanical defect type and obtaining a third diagnostic result.

Images