Abnormality detection method and system based on motor operation vibration sound

Abstract

The present invention proposes a method and system for detecting anomalies based on the vibration sound of motor operation, which relates to the field of motor technology. The method includes: obtaining an original motor signal, determining the amplitude of the original motor signal, and marking the abnormal signal therein; determining the spectral distribution characteristics of the vibration signal in the abnormal signal, performing multi-scale decomposition on the vibration signal to obtain a number of sub-band signals, and integrating the sub-band signals to obtain a multi-band combined signal; using a variational mode decomposition algorithm to separate the multi-band combined signal into a number of intrinsic mode components, integrating the intrinsic mode components to obtain a reconstructed signal; performing time domain and frequency domain analysis on the reconstructed signal to obtain frequency domain distribution characteristics; inputting the frequency domain distribution characteristics into a trained motor anomaly detection model to obtain anomaly detection results. The present invention can effectively extract multidimensional abnormal characteristics during motor operation and improve the accuracy and reliability of anomaly detection.

Description

Technical Field

[0001] The present invention relates to the field of motor technology, and in particular to a method and system for detecting anomalies based on the vibration sound of motor operation. Background Art

[0002] As the core power equipment of modern industrial production, the operating status of the motor is directly related to the stability and safety of the entire equipment system. With the continuous improvement of industrial automation, the production losses caused by motor failure are becoming increasingly serious, making motor anomaly detection technology a key link in ensuring the continuity of industrial production.

[0003] The current mainstream motor anomaly detection methods mainly rely on traditional frequency domain analysis and time domain statistical feature extraction. These methods often show obvious limitations when processing multi-source signals in complex environments, making the current mainstream motor anomaly detection methods perform unsatisfactorily in practical applications.

[0004] Therefore, how to effectively extract and retain the key abnormal features in the motor vibration sound signal in a complex environment has become an important factor in improving the accuracy and reliability of motor abnormality detection. Summary of the Invention

[0005] In view of this, the present invention proposes an abnormality detection method and system based on the vibration sound of motor operation, so as to solve the problem that the current motor abnormality detection method often shows obvious limitations when processing multi-source signals in complex environments.

[0006] The technical solution of the present invention is achieved as follows:

[0007] According to a first aspect, an embodiment of the present invention provides a method for detecting abnormalities based on vibration sound during motor operation, the method comprising:

[0008] Acquire the original motor signal during the operation of the motor to be detected, determine the amplitude of the original motor signal, and mark the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal;

[0009] Determine the spectrum distribution characteristics of the vibration signal in the abnormal signal, perform multi-scale decomposition on the vibration signal based on the spectrum distribution characteristics to obtain a number of sub-band signals, and integrate the sub-band signals based on their energy proportions to obtain a multi-band combined signal;

[0010] The variational mode decomposition algorithm is used to separate the multi-band combination signal into several eigenmode components. The eigenmode components are integrated according to the degree of correlation between the eigenmode components and the multi-band combination signal to obtain the reconstructed signal.

[0011] Performing time domain analysis on the reconstructed signal to obtain time domain abnormality features, and performing frequency domain analysis on the time domain abnormality features to obtain frequency domain distribution features; the time domain abnormality features carry the abnormal time period of the original motor signal, and the frequency domain distribution features carry the distribution law of nonlinear changes in the frequency domain;

[0012] The frequency domain distribution features are input into the trained motor anomaly detection model to obtain the anomaly detection results output by the motor anomaly detection model.

[0013] In combination with the first aspect, in the first implementation of the first aspect, determining the spectral distribution characteristics of the vibration signal in the abnormal signal, performing multi-scale decomposition on the vibration signal according to the spectral distribution characteristics to obtain a plurality of sub-band signals, and integrating the sub-band signals according to the energy proportions of the sub-band signals to obtain a multi-band combined signal, specifically includes:

[0014] Obtaining the vibration signal from the abnormal signal, performing spectrum analysis on the vibration signal, and obtaining the spectrum distribution characteristics of the vibration signal;

[0015] According to the spectrum distribution characteristics and the adaptive wavelet transform algorithm, the vibration signal is decomposed into multiple scales to obtain several sub-band signals;

[0016] The energy proportion of each sub-band signal is determined, the sub-band signals whose energy proportion exceeds a preset proportion are screened out, and the sub-band signals exceeding the preset proportion are integrated into a multi-band combined signal through signal recombination.

[0017] In combination with the first aspect, in the second implementation of the first aspect, the multi-band combination signal is separated into a plurality of eigenmode components using a variational mode decomposition algorithm, and the eigenmode components are integrated according to the degree of correlation between the eigenmode components and the multi-band combination signal to obtain a reconstructed signal, specifically including:

[0018] Perform frequency domain separation processing on the multi-band combined signal to obtain a preliminary separation signal containing abnormal features;

[0019] The variational mode decomposition algorithm is used to separate the multi-band composite signal into several eigenmode components;

[0020] determining a correlation coefficient between each eigenmode component and the abnormal feature, and determining a correlation degree between each eigenmode component and the multi-band combined signal based on the correlation coefficient;

[0021] The eigenmode components exceeding the preset correlation degree are integrated to obtain the denoised reconstructed signal.

[0022] In combination with the first aspect, in the third implementation of the first aspect, the time domain analysis of the reconstructed signal is performed to obtain the time domain abnormality feature, and the frequency domain analysis of the time domain abnormality feature is performed to obtain the frequency domain distribution feature, specifically including:

[0023] Segmenting the reconstructed signal according to a preset sliding window to obtain a plurality of time domain windows, and determining time domain statistical parameters within each time domain window; the time domain statistical parameters include mean, variance, deviation, and kurtosis;

[0024] Determine the change rate of the time domain statistical parameter between time domain windows adjacent in time, and determine the potential abnormal period in the time domain window according to the change rate;

[0025] Clustering and classifying the time domain windows determined as potential abnormal periods to determine several groups of time domain abnormality features; the time domain windows within each group of the time domain abnormality features have similar statistical characteristics;

[0026] Frequency domain analysis is performed on each set of time domain anomaly features to identify nonlinear changes in the time domain anomaly features and obtain frequency domain distribution features.

[0027] In combination with the third embodiment of the first aspect, in the fourth embodiment of the first aspect, frequency domain analysis is performed on each group of time domain abnormal features to identify nonlinear changes in the time domain abnormal features to obtain frequency domain distribution features, specifically including:

[0028] Hilbert-Huang transform is used to perform frequency domain decomposition on each set of time domain anomaly features to obtain the instantaneous frequency and instantaneous amplitude information of each set of time domain anomaly features;

[0029] Construct a time-frequency spectrum of each group of time-domain anomaly features based on the instantaneous frequency and instantaneous amplitude information, and extract the frequency distribution in the time-frequency spectrum;

[0030] Determining a frequency offset of each frequency distribution, and determining a ratio between the frequency offset and a preset reference frequency, and marking a frequency distribution whose ratio exceeds a preset ratio as a frequency domain abnormal feature;

[0031] All frequency domain abnormal features are collected to obtain the frequency domain distribution law of the frequency domain abnormal features, feature positioning is performed on the frequency domain distribution law, the distribution law of the frequency domain abnormal features in the frequency domain is obtained, and the frequency domain distribution features are obtained.

[0032] In combination with the first aspect, in a fifth implementation of the first aspect, the motor abnormality detection model is trained based on sample motor signals obtained from the test motor in a test environment and is constructed using a support vector machine classifier.

[0033] In combination with the fifth implementation of the first aspect, in the sixth implementation of the first aspect, the motor abnormality detection model is trained in the following manner:

[0034] Acquire a sample motor signal of the test motor under a test environment, and determine a sample frequency domain distribution feature in the sample motor signal;

[0035] Normal frequency domain distribution features are obtained, and a support vector machine classifier is used to perform supervised training on the sample frequency domain distribution features and the normal frequency domain distribution features, to obtain a preset hyperplane and a preset safety margin, and to obtain a trained support vector machine classifier.

[0036] The trained support vector machine classifier is used to classify and predict the frequency domain distribution characteristics of the samples, determine the distance value from the frequency domain distribution characteristics of the samples to the preset hyperplane, and generate anomaly detection results based on the distance value and the preset safety margin. The anomaly detection results are verified to train a motor anomaly detection model.

[0037] In combination with the first aspect, in a seventh implementation of the first aspect, obtaining the original motor signal during the operation of the motor to be detected, determining the amplitude of the original motor signal, and marking the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal specifically includes:

[0038] A multi-channel sensor array is used to obtain the original motor signal during the operation of the motor to be tested;

[0039] Determine a first amplitude of a vibration signal and a second amplitude of a sound characteristic signal in the original motor signal;

[0040] The raw motor signal whose first amplitude exceeds a first preset amplitude and / or whose second amplitude exceeds a second preset amplitude is marked as an abnormal signal.

[0041] In combination with the seventh embodiment of the first aspect, in the eighth embodiment of the first aspect, marking the original motor signal whose first amplitude exceeds the first preset amplitude and / or whose second amplitude exceeds the second preset amplitude as an abnormal signal specifically includes:

[0042] marking the original motor signal whose first amplitude exceeds the first preset amplitude and / or whose second amplitude exceeds the second preset amplitude as a candidate signal;

[0043] Determine the root mean square value of the candidate abnormal signal, and mark the candidate signal whose root mean square value exceeds a preset value as a risk signal;

[0044] Identify abnormal signals from risk signals based on business scenarios.

[0045] According to a second aspect, an embodiment of the present invention provides an abnormality detection system based on the vibration sound of a motor operation, the system comprising:

[0046] A signal acquisition module is used to acquire the original motor signal during the operation of the motor to be tested, determine the amplitude of the original motor signal, and mark the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal;

[0047] The signal decomposition module is used to determine the spectral distribution characteristics of the vibration signal in the abnormal signal, perform multi-scale decomposition on the vibration signal according to the spectral distribution characteristics, obtain a number of sub-band signals, and integrate the sub-band signals according to the energy proportion of the sub-band signals to obtain a multi-band combined signal;

[0048] A signal separation module is used to separate the multi-band combination signal into a number of eigenmode components using a variational mode decomposition algorithm, and integrate the eigenmode components according to the degree of correlation between the eigenmode components and the multi-band combination signal to obtain a reconstructed signal;

[0049] A signal analysis module is used to perform time domain analysis on the reconstructed signal to obtain time domain abnormality features, and to perform frequency domain analysis on the time domain abnormality features to obtain frequency domain distribution features; the time domain abnormality features carry the abnormal time period of the original motor signal, and the frequency domain distribution features carry the distribution pattern of nonlinear changes in the frequency domain;

[0050] The anomaly detection module is used to input the frequency domain distribution features into the trained motor anomaly detection model to obtain the anomaly detection results output by the motor anomaly detection model.

[0051] The method and system for detecting abnormalities based on motor vibration sound of the present invention have the following beneficial effects compared with the prior art:

[0052] The present invention determines the amplitude of an original motor signal and marks any original motor signal with an amplitude exceeding a preset amplitude as an abnormal signal. Multi-scale decomposition is then performed on the abnormal signal to determine sub-band signals of the main frequency components in the original signal. These sub-band signals are integrated to obtain a multi-band combined signal. A variational mode decomposition algorithm is then used to remove noise interference components from the multi-band combined signal. Time and frequency domain analysis are then performed to extract features from nonlinear data and obtain the distribution pattern of nonlinear changes in the frequency domain. The frequency domain distribution range of the abnormal feature is then determined, resulting in a frequency domain distribution feature. Finally, the frequency domain distribution feature is input into a trained motor abnormality detection model to obtain an abnormality detection result output by the motor abnormality detection model. The motor abnormality detection model is trained based on sample motor signals obtained from a test motor in a test environment and constructed using a support vector machine classifier. This results in more accurate abnormality detection results. The present invention achieves accurate identification of motor abnormal states, effectively extracts multi-dimensional abnormal features during motor operation, significantly improves the accuracy and reliability of abnormality detection, and provides important technical support for motor fault diagnosis and predictive maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 This is a flow chart of a method for detecting abnormalities based on motor vibration sound according to the present invention;

[0055] Figure 2 This is a second flow chart of the method for detecting abnormalities based on motor vibration sound according to the present invention;

[0056] Figure 3 This is a third flow chart of the method for detecting abnormalities based on motor operation vibration sound of the present invention;

[0057] Figure 4 This is a fourth flow chart of the method for detecting abnormalities based on motor operation vibration sound of the present invention;

[0058] Figure 5 FIG5 is a flowchart of the method for detecting abnormality based on the vibration sound of a motor in accordance with the present invention;

[0059] Figure 6 The figure is a schematic structural diagram of the abnormality detection system based on the vibration sound of motor operation according to the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] As the core power equipment of modern industrial production, the operating status of the motor is directly related to the stability and safety of the entire equipment system. With the continuous improvement of industrial automation, the production losses caused by motor failure are becoming increasingly serious, making motor anomaly detection technology a key link in ensuring the continuity of industrial production.

[0062] The current mainstream motor anomaly detection methods mainly rely on traditional frequency domain analysis and time domain statistical feature extraction. These methods often show obvious limitations when processing multi-source signals in complex environments, making the current mainstream motor anomaly detection methods perform unsatisfactorily in practical applications.

[0063] The main reason is that these methods have difficulty effectively distinguishing useful signals from interfering noise, resulting in low detection accuracy and high false alarm rates. Furthermore, the motor vibration sound signal itself has complex nonlinear characteristics, which contain rich information about the device status. However, this critical information is often obscured by various environmental noises and irrelevant signals. Traditional denoising methods, while removing noise, tend to filter out weak signals that carry abnormal characteristics, resulting in the loss of key abnormal information. For example, the filtering technology used in the motor anomaly detection process often uses a linear processing method with fixed parameters, which cannot adapt to the dynamic changes in signal characteristics during motor operation.

[0064] It's also important to note that the characteristics of motor anomaly signals often exhibit nonlinear transient variations, which are difficult to accurately capture using traditional linear analysis methods. When signal processing fails to effectively preserve these key anomaly characteristics, subsequent detection algorithms lose the basis for accurate judgment, further exacerbating the difficulty of achieving high-precision anomaly detection in high-noise environments and significantly reducing anomaly detection reliability.

[0065] In summary, how to effectively extract and retain the key abnormal features in the motor vibration sound signal in a complex environment has become an important factor in improving the accuracy and reliability of motor abnormality detection.

[0066] The anomaly detection method based on motor operation vibration sound provided in this manual is designed to effectively extract multi-dimensional anomaly characteristics during motor operation, significantly improve the accuracy and reliability of anomaly detection, and provide important technical support for motor fault diagnosis and predictive maintenance.

[0067] See also Figure 1 , Figure 1 The figure is a flow chart of a method for detecting abnormalities based on the vibration sound of a motor in accordance with the present invention. The method may include the following steps:

[0068] S101: Acquire an original motor signal during the operation of a motor to be tested, determine the amplitude of the original motor signal, and mark an original motor signal with an amplitude exceeding a preset amplitude as an abnormal signal. The original motor signal includes an acquisition channel, a vibration signal, and a sound characteristic signal.

[0069] In this embodiment, a multi-channel sensor array is used to acquire the vibration signal and the sound characteristic signal during the operation of the motor, thereby obtaining the above-mentioned original motor signal.

[0070] Preferably, the multi-channel sensor array is an array consisting of a number of vibration sensors (for example, 8) and sound sensors (for example, 4), which are respectively installed near the motor bearing seat and the housing. The sensor signals collected by these sensors can be controlled by a high-speed signal acquisition card with a synchronous clock signal to ensure that the signals of each channel are aligned in time. For example, the sampling frequency of the multi-channel sensor array is set to 10 kHz to capture high-frequency vibration signals and sound characteristic signals.

[0071] Specifically, step S101 includes:

[0072] S1011. Use a multi-channel sensor array to obtain the original motor signal during the operation of the motor to be detected.

[0073] S1012: Determine a first amplitude of the vibration signal and a second amplitude of the sound characteristic signal in the original motor signal.

[0074] In this embodiment, the original motor signal is digitized, for example, by using a 16-bit precision analog-to-digital converter to convert the original motor signal in analog signal form into a digital signal form, and then the amplitude of the original motor signal is calculated in real time to obtain an original signal sequence containing timestamp information and amplitude information.

[0075] S1013 : Mark the original motor signal whose first amplitude exceeds a first preset amplitude and / or whose second amplitude exceeds a second preset amplitude as an abnormal signal.

[0076] It is understood that the first preset amplitude is the amplitude range corresponding to the vibration signal, for example, the first preset amplitude is set to -3.5V to 3.5V, and the first preset amplitude is the amplitude range corresponding to the sound characteristic signal, for example, the first preset amplitude is set to -1.5V to 1.5V. If it is calculated that the first amplitude of the vibration signal obtained by a certain channel exceeds the first preset amplitude and / or the second amplitude of the sound characteristic signal exceeds the second preset amplitude, the original motor signal in this time period is marked as an abnormal signal.

[0077] Considering that the indicators of motor abnormality signals may be different in different business scenarios, more specifically, step S1013 may further include:

[0078] S10131. Mark the original motor signal whose first amplitude exceeds the first preset amplitude and / or whose second amplitude exceeds the second preset amplitude as a candidate signal. When storing the candidate signal, each candidate abnormal signal is also given candidate tag information. For example, the vibration signal of an original motor signal collected by channel 1 reaches an amplitude of 6.2V at t=10.23 seconds and lasts for 0.05 seconds. The original motor signal is marked as a candidate abnormal signal. The corresponding candidate tag information includes the start timestamp, end timestamp, duration, channel number and amplitude signal, etc., as shown in [ =10.23, =10.28, Channel 1, 6.2V], where Indicates the start timestamp, Indicates the end timestamp.

[0079] S10312. Determine the root mean square (RMS) value of the candidate signal, and mark the candidate signal whose RMS value exceeds a preset value as a risk signal.

[0080] Next, the candidate signals are analyzed using a sliding window algorithm (e.g., a window size of 1 second and a step size of 0.5 seconds) to determine the RMS of the candidate signals within the window. Candidate abnormal signals whose RMS exceeds the configured preset value are further marked as high-risk risk signals.

[0081] S10133. Determine abnormal signals from risk signals based on business scenarios.

[0082] In this embodiment, risk signals can be associated with business information related to the business scenario, such as motor load signals. If the motor load signals simultaneously exceed 80% of the rated value, the priority is increased and the corresponding risk signal is identified as an abnormal signal, triggering subsequent in-depth analysis.

[0083] It should be noted that the business information in different business scenarios may be different, depending on the specific business scenario. In this way, the abnormal signal can also be associated with the business information and saved. When extracting the signal later, it can be segmented according to the time interval corresponding to the abnormal signal to obtain a structured signal group.

[0084] In this way, a complete logical chain is formed from signal acquisition to abnormality marking to abnormal signal generation, ensuring the efficiency and accuracy of signal processing.

[0085] S102. Determine the spectrum distribution characteristics of the vibration signal in the abnormal signal, perform multi-scale decomposition on the vibration signal according to the spectrum distribution characteristics to obtain a plurality of sub-band signals, and integrate the sub-band signals according to the energy proportions of the sub-band signals to obtain a multi-band combined signal.

[0086] In this embodiment, an adaptive wavelet transform algorithm is used to perform multi-scale decomposition of the vibration signal based on the spectral distribution characteristics of the vibration signal. The main frequency components are then determined by calculating the energy proportion of each sub-band signal obtained by the decomposition. The main frequency components are then reorganized and integrated to obtain a multi-band combined signal. The multi-band combined signal obtained in this way can better reflect the signal characteristics, that is, it contains the original abnormal characteristics, and is therefore more suitable for subsequent fault diagnosis and feature extraction.

[0087] S103. Using a variational mode decomposition (VMD) algorithm, the multi-band combination signal is separated into a number of eigenmode components. The eigenmode components are integrated according to the degree of correlation between the eigenmode components and the multi-band combination signal to obtain a reconstructed signal.

[0088] In this embodiment, the noise interference portion of the multi-band combined signal is subjected to frequency domain separation processing to determine the potential interference frequency band. The multi-band combined signal is then separated using the VMD algorithm to obtain a number of eigenmode components (functions). The correlation coefficient of each modal function is calculated to determine its degree of correlation with the original abnormal feature. If the correlation coefficient is lower than a preset correlation degree, the eigenmode component is eliminated, and the denoised reconstructed signal is obtained based on the eigenmode components whose correlation coefficient is not lower than the preset correlation degree.

[0089] S104. Perform time domain analysis on the reconstructed signal to obtain time domain abnormality features, and perform frequency domain analysis on the time domain abnormality features to obtain frequency domain distribution features, wherein the time domain abnormality features carry the abnormal time period of the original motor signal, and the frequency domain distribution features carry the distribution law of nonlinear changes in the frequency domain.

[0090] In this embodiment, time domain analysis is performed on the reconstructed signal to obtain several time domain windows and their corresponding time domain statistical parameters. The time domain window containing the abnormal time period is then determined based on the time domain statistical parameters, thereby obtaining a time domain anomaly feature. Frequency domain analysis is then performed on the time domain anomaly feature to extract features from the nonlinear data and obtain the frequency domain distribution pattern of the nonlinear changes in the time domain anomaly feature. That is, the frequency domain distribution range of the frequency domain anomaly feature is determined, thereby obtaining a frequency domain distribution feature. The frequency domain distribution feature obtained in this manner can improve the accuracy and reliability of anomaly detection when performing motor anomaly analysis, providing important technical support for motor fault diagnosis and predictive maintenance.

[0091] S105 , inputting the frequency domain distribution features into the trained motor abnormality detection model to obtain an abnormality detection result output by the motor abnormality detection model.

[0092] In this embodiment, the motor abnormality detection model is trained based on the sample motor signals obtained from the test motor in the test environment. It can be understood that the sample motor signals will also be processed from step S102 to step S104, that is, the training samples finally used are the sample frequency domain distribution characteristics of the test motor.

[0093] The sample motor signal of the test motor can be pre-stored in the electronic device or acquired in real time from the test environment. There is no restriction on the specific acquisition method of the sample motor signal, as long as the electronic device can acquire the sample motor signal.

[0094] In this embodiment, a support vector machine (SVM) classifier is used to build a motor anomaly detection model. The motor anomaly detection model learns the characteristic boundaries between normal and abnormal states through training samples, thereby obtaining an anomaly detection result of the motor to be detected.

[0095] The present invention provides an abnormality detection method based on motor operation vibration sound. The method determines the amplitude of an original motor signal and marks an original motor signal with an amplitude exceeding a preset amplitude as an abnormal signal. The abnormal signal is then subjected to multi-scale decomposition to determine sub-band signals of the main frequency components in the original signal. These sub-band signals are integrated to obtain a multi-band combined signal. The noise interference components in the multi-band combined signal are then removed using a variational mode decomposition algorithm. Time domain and frequency domain analysis are then performed to extract features from nonlinear data and obtain the distribution pattern of nonlinear changes in the frequency domain. The frequency domain distribution range of the abnormal feature is determined to obtain a frequency domain distribution feature. Finally, the frequency domain distribution feature is input into a trained motor abnormality detection model to obtain an abnormality detection result output by the motor abnormality detection model. The motor abnormality detection model is trained based on sample motor signals obtained from a test motor in a test environment and constructed using a support vector machine classifier. The abnormality detection result obtained in this way is more accurate. The present invention achieves accurate identification of motor abnormal conditions and can effectively extract multi-dimensional abnormal features during motor operation, significantly improving the accuracy and reliability of abnormality detection, and providing important technical support for motor fault diagnosis and predictive maintenance.

[0096] See also Figure 2 , the method may further comprise the following steps:

[0097] S201: Acquire the original motor signal during the operation of the motor to be detected, determine the amplitude of the original motor signal, and mark the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal. For details, refer to step S101.

[0098] S2021. Obtain a vibration signal from the abnormal signal, perform spectrum analysis on the vibration signal, and obtain spectrum distribution characteristics of the vibration signal.

[0099] Assume that the vibration signal is a time-domain sequence with a sampling frequency of 1000Hz and a duration of 1 second. The vibration signal contains 2048 sampling points and is composed of multiple frequency components. By performing spectral analysis on the vibration signal and calculating the spectrum using the Fast Fourier Transform algorithm, we obtain the spectral distribution within the 0-500Hz range. The spectral distribution characteristics show that the main energy of the vibration signal is concentrated near 50Hz, 120Hz, and 300Hz, with a total energy of 1000 units.

[0100] S2022. Perform multi-scale decomposition on the vibration signal according to the spectrum distribution characteristics and adopt an adaptive wavelet transform algorithm to obtain a plurality of sub-band signals.

[0101] After determining the spectrum distribution characteristics, the adaptive wavelet transform algorithm is then used, the DB4 wavelet basis function is selected, the number of decomposition layers is set to 5, and sub-band signals with frequency bands of 0-31.25Hz, 31.25-62.5Hz, 62.5-125Hz, 125-250Hz and 250-500Hz are generated.

[0102] S2023: Determine the energy proportion of each sub-band signal, screen out sub-band signals whose energy proportion exceeds a preset proportion, and integrate the sub-band signals exceeding the preset proportion into a multi-band combined signal through signal recombination.

[0103] After multi-scale decomposition, the energy and energy proportion of each sub-band signal are determined. For example, the energy of the sub-band signal corresponding to 0-31.25Hz is 50 units, the energy of the sub-band signal corresponding to 31.25-62.5Hz is 200 units, the energy of the sub-band signal corresponding to 62.5-125Hz is 400 units, the energy of the sub-band signal corresponding to 125-250Hz is 250 units, and the energy of the sub-band signal corresponding to 250-500Hz is 100 units. The energy proportions of these sub-band signals are 5%, 20%, 40%, 25%, and 10%, respectively. Sub-band signals that exceed a preset ratio, such as an energy proportion of 23%, are retained. These retained sub-band signals are referred to as key band signals. Next, the retained sub-band signals are recombined and integrated using a signal reconstruction algorithm, such as an inverse wavelet transform, to obtain a multi-band composite signal containing the main frequency components.

[0104] In this embodiment, to ensure the signal quality of the integrated multi-band combination signal, the mean-square error (MSE) between the multi-band combination signal and the original motor signal is calculated to determine the degree of difference between the two signals, thereby determining whether the multi-band combination signal can effectively preserve the signal characteristics of the original motor signal.

[0105] Preferably, for the multi-band combination signal, a low-pass filtering algorithm may be used to perform denoising processing on the multi-band combination signal, so as to better perform subsequent analysis of the vibration signal.

[0106] S203: Separate the multi-band composite signal into several eigenmode components using a variational mode decomposition algorithm, and integrate the eigenmode components according to their correlation with the multi-band composite signal to obtain a reconstructed signal. For details, refer to step S103.

[0107] S204: Perform time domain analysis on the reconstructed signal to obtain time domain abnormal features, and perform frequency domain analysis on the time domain abnormal features to obtain frequency domain distribution features. For details, refer to step S104.

[0108] S205: Input the frequency domain distribution features into the trained motor abnormality detection model to obtain the abnormality detection result output by the motor abnormality detection model. For details, refer to step S105.

[0109] See also Figure 3 , the method may further comprise the following steps:

[0110] S301: Acquire the original motor signal during the operation of the motor to be detected, determine the amplitude of the original motor signal, and mark the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal. For details, refer to step S101.

[0111] S302: Determine the spectral distribution characteristics of the vibration signal in the abnormal signal, perform multi-scale decomposition on the vibration signal based on the spectral distribution characteristics to obtain a number of sub-band signals, and integrate the sub-band signals based on their energy proportions to obtain a multi-band combined signal. For details, refer to step S102.

[0112] S3031. Perform frequency domain separation processing on the multi-band combined signal to obtain a preliminary separation signal containing original abnormal features.

[0113] In this embodiment, a bandpass filter bank or other method is used to perform frequency domain separation on a multi-band signal, generating a preliminary separated signal containing frequency band identification information. This information indicates whether the frequency band is a potential interference region, that is, whether it contains noise interference components. For example, the preliminary separated signal contains significant 50Hz and 100Hz main frequency components, as well as random noise interference.

[0114] S3032. Use a VMD algorithm to separate the multi-band combined signal into a number of eigenmode components.

[0115] In this embodiment, when the VMD algorithm is used for iterative optimization, if the iterative convergence error is greater than a preset convergence threshold, the number of modes is increased and the decomposition is performed again to obtain a stable intrinsic mode component, namely, an eigenmode function.

[0116] Preferably, VMD is performed on the multi-band composite signal using the optimal modal layer calculation method to obtain the optimal modal decomposition layer number and generate K eigenmodal components. Assuming that the optimal modal decomposition layer number determined by the optimal modal layer calculation method is 5 layers, the VMD algorithm can decompose the signal into five eigenmodal components, namely mode 1 to mode 5.

[0117] S3033. Determine the correlation coefficient between each eigenmode component and the original abnormal feature, and determine the degree of association between each eigenmode component and the multi-band combined signal based on the correlation coefficient.

[0118] In this embodiment, the Pearson algorithm can be used to calculate the correlation between each eigenmode component and the waveform corresponding to the original abnormal feature, and then the correlation coefficient between each eigenmode component and the original abnormal feature can be obtained. Then, the degree of association between each eigenmode component and the multi-band combination signal can be obtained based on the correlation coefficient.

[0119] S3034. Integrate the intrinsic mode components that exceed the preset correlation degree to obtain a denoised reconstructed signal.

[0120] Since the multi-band combination signal contains the original abnormal features, accordingly, the higher the correlation degree of the intrinsic mode component, the higher the similarity with the multi-band combination signal, and the less the noise interference component mixed in it, which means that the intrinsic mode component is valid data. Conversely, the higher the correlation degree of the intrinsic mode component, the lower the similarity with the multi-band combination signal, which means that the intrinsic mode component can be classified as a noise interference component.

[0121] In this embodiment, a linear superposition reconstruction algorithm is used to perform weighted summation on the time-domain waveforms of the retained eigenmode components to integrate the eigenmode components. To verify the denoising effect of the integrated reconstructed signal, the MSE of the reconstructed signal can be compared with the multi-band combined signal.

[0122] S304: Perform time domain analysis on the reconstructed signal to obtain time domain abnormal features, and perform frequency domain analysis on the time domain abnormal features to obtain frequency domain distribution features. For details, refer to step S104.

[0123] S305: Input the frequency domain distribution features into the trained motor abnormality detection model to obtain the abnormality detection result output by the motor abnormality detection model. For details, refer to step S105.

[0124] See also Figure 4 , the method may further comprise the following steps:

[0125] S401: Acquire the original motor signal during the operation of the motor to be detected, determine the amplitude of the original motor signal, and mark the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal. For details, refer to step S101.

[0126] S402: Determine the spectral distribution characteristics of the vibration signal in the abnormal signal, perform multi-scale decomposition on the vibration signal based on the spectral distribution characteristics to obtain a number of sub-band signals, and integrate the sub-band signals based on their energy proportions to obtain a multi-band combined signal. For details, refer to step S102.

[0127] S403: Separate the multi-band composite signal into several eigenmode components using a variational mode decomposition algorithm, and integrate the eigenmode components according to their correlation with the multi-band composite signal to obtain a reconstructed signal. For details, refer to step S103.

[0128] S4041. Segment the reconstructed signal according to a preset sliding window to obtain a plurality of time domain windows, and determine the time domain statistical parameters in each time domain window, wherein the time domain statistical parameters include mean, variance, deviation, and kurtosis.

[0129] In this embodiment, a sliding window is preset with a window size of 128 sampling points and a sliding step size of 64 sampling points to ensure that adjacent time domain windows overlap to capture continuous changes. Time domain statistical features, i.e., time domain statistical parameters, are calculated for each time domain window. For example, in the first time domain window, the mean is calculated to be 0.23, the variance is 1.45, the skewness is 0.67, and the kurtosis is 2.89. Subsequent time domain windows are then calculated sequentially to form a series of time domain statistical parameters.

[0130] In this embodiment, a linear interpolation method may be used to smooth the statistical parameters within the abnormal period to generate a continuous set of time-domain feature vectors and determine the final representation of the time-domain statistical parameters.

[0131] S4042: Determine the change rate of the time domain statistical parameters between temporally adjacent time domain windows, and determine a potential abnormal period in the time domain window according to the change rate.

[0132] Next, the rate of change of the time domain statistical parameters between temporally adjacent time domain windows is analyzed. The rate of change characterizes the transient characteristic interval of the signal. If the rate of change exceeds a preset percentage, both time domain windows are marked as potential abnormal periods. This process is repeated until the rate of change of the time domain statistical parameters between all time domain windows is calculated. In this way, all potential abnormal periods are marked. The representation of the potential abnormal period can be [mean 0.23, variance 1.45, skewness 0.67, kurtosis 2.89, abnormality mark 1].

[0133] In order to determine the transient characteristic interval, in this embodiment, a dynamic threshold algorithm is adopted to dynamically adjust the value of the preset percentage based on the standard deviation of the historical window data.

[0134] In this way, a complete logic chain is formed from signal feature extraction to motor abnormal state detection to ensure data availability.

[0135] S4043. Cluster and classify the time domain windows determined as potential abnormal time periods to determine several groups of time domain abnormality features, wherein the time domain windows within each group of time domain abnormality features have similar statistical features.

[0136] It can be understood that each group of time-domain anomaly features is a time-domain anomaly feature set.

[0137] S4044. Perform frequency domain analysis on each set of time domain anomaly features to identify nonlinear changes in the time domain anomaly features and obtain frequency domain distribution features.

[0138] In this embodiment, the Hilbert-Huang transform is used to perform frequency domain analysis on each group of time domain abnormal features to obtain the instantaneous frequency and instantaneous amplitude information of each group of time domain abnormal features, and then a time-frequency spectrum is constructed based on the instantaneous frequency and instantaneous amplitude information to identify the nonlinear change pattern. For the nonlinear change pattern, the ratio between the frequency offset and the preset reference frequency is calculated. If the ratio exceeds the preset ratio, it is marked as a frequency domain abnormal feature. Then, a frequency domain distribution analysis tool is used to locate the frequency domain abnormal feature, and the distribution law of the frequency domain abnormal feature in the frequency domain is obtained, thereby determining the frequency domain distribution feature.

[0139] More specifically, step S4044 includes:

[0140] S40441. Use Hilbert-Huang transform to perform frequency domain decomposition processing on each group of time domain anomaly features to obtain instantaneous frequency and instantaneous amplitude information of each group of time domain anomaly features.

[0141] For example, a Hilbert-Huang transform is performed on a signal with a sampling frequency of 1000 Hz and a data length of 4096 sampling points. The frequency domain distribution characteristics are decomposed into 10 intrinsic mode functions using empirical mode decomposition. The first three main modes are extracted, and their instantaneous frequency and instantaneous amplitude information are calculated.

[0142] At the same time, the Hilbert-Huang transform can also verify the integrity of each set of frequency domain distribution features in the data.

[0143] S40442. Construct a time-frequency spectrum diagram of each set of time-domain anomaly features based on the instantaneous frequency and instantaneous amplitude information, and extract the frequency distribution in the time-frequency spectrum diagram. The frequency distribution in the time-frequency spectrum diagram is also the nonlinear pattern in the time-frequency spectrum diagram. Feature extraction of the time-frequency spectrum diagram is to extract the distribution law of nonlinear changes in the time-frequency spectrum diagram.

[0144] The horizontal axis of the time-spectrum graph is time (0 to 4.096 seconds) and the vertical axis is frequency (0 to 500 Hz).

[0145] S40443. Determine the frequency offset of each frequency distribution, and determine the ratio between the frequency offset and a preset reference frequency, and mark the frequency distribution whose ratio exceeds the preset ratio as a frequency domain abnormal feature.

[0146] That is, if the ratio of the frequency offsets of a certain frequency distribution exceeds a preset ratio, the frequency distribution is marked as a frequency domain abnormal feature.

[0147] S40444. Gather all frequency domain abnormal features to obtain the frequency domain distribution law of the frequency domain abnormal features, perform feature location on the frequency domain distribution law, obtain the distribution law of the frequency domain abnormal features in the frequency domain, and obtain the frequency domain distribution features.

[0148] The frequency domain distribution pattern obtained by aggregating all frequency domain anomaly features can further combine the original anomaly features with business scenarios. For example, frequency offset may be associated with equipment speed fluctuations, and a frequency offset greater than 10% may indicate bearing wear risk. The generated frequency domain distribution pattern can be expressed as [instantaneous frequency 53.5, instantaneous amplitude 0.87, anomaly flag 1], which is used for subsequent data analysis to ensure the completeness of the logical chain from frequency domain analysis to equipment status assessment.

[0149] In this embodiment, a frequency domain distribution analysis tool is used to locate the frequency domain distribution law obtained based on the frequency domain abnormality feature, so as to determine the frequency domain distribution feature.

[0150] S405: Input the frequency domain distribution features into the trained motor abnormality detection model to obtain the abnormality detection result output by the motor abnormality detection model. For details, refer to step S105.

[0151] See also Figure 5 In this method, the motor anomaly detection model is trained in the following way:

[0152] S501 : Obtain a sample motor signal of a test motor under a test environment, and determine a sample frequency domain distribution feature in the sample motor signal.

[0153] The sample motor signal includes positive samples, i.e., normal samples, and negative samples, i.e., abnormal samples. In order to ensure the effectiveness of model training, the sample frequency domain distribution characteristics are determined from the negative samples in the sample motor signal.

[0154] The method of obtaining the frequency domain distribution characteristics of the samples can be found in steps S102 to S04 and will not be elaborated here.

[0155] S502. Obtain normal frequency domain distribution features, use an SVM classifier to perform supervised training on the sample frequency domain distribution features and the normal frequency domain distribution features, learn the feature boundaries of the normal state and the abnormal state, obtain a preset hyperplane and a preset safety margin, and obtain a trained SVM classifier.

[0156] It can be understood that the normal frequency domain distribution features are determined from positive samples in the sample motor signal.

[0157] Among them, the preset hyperplane consists of a 10-dimensional weight vector and bias composition.

[0158] Preferably, the SVM classifier uses a radial basis function kernel and sets the kernel parameters = 0.1 and penalty parameter =100.

[0159] S503. Use the trained SVM classifier to classify and predict the sample frequency domain distribution characteristics, determine the distance value from the sample frequency domain distribution characteristics to the preset hyperplane, and generate anomaly detection results based on the distance value and the preset safety margin, and verify the anomaly detection results to train a motor anomaly detection model.

[0160] Among them, if the distance value is less than the preset safety margin, the abnormal detection result generated is that the motor is in an abnormal working state. Conversely, if the distance value is not less than the preset safety margin, the abnormal detection result generated is that the motor is in a normal working state. If business association is required at the same time, the abnormal detection result can be connected to the equipment maintenance system to trigger an alarm below the safety margin. Combined with historical data analysis, the abnormal frequency (such as twice a week) is optimized for maintenance scheduling.

[0161] The following describes a system provided by an embodiment of the present invention. The system described below and the method described above can refer to each other.

[0162] See also Figure 6 , Figure 6 The following is a schematic diagram showing the structure of a system for detecting abnormalities based on motor vibration and sound according to an embodiment of the present invention. The system may include:

[0163] The signal acquisition module 10 is used to acquire the original motor signal during the operation of the motor to be detected, determine the amplitude of the original motor signal, and mark the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal.

[0164] In this embodiment, a multi-channel sensor array is used to acquire the vibration signal and the sound characteristic signal during the operation of the motor, thereby obtaining the above-mentioned original motor signal.

[0165] Preferably, the multi-channel sensor array is an array consisting of a number of vibration sensors (for example, 8) and sound sensors (for example, 4), which are respectively installed near the motor bearing seat and the housing. The sensor signals collected by these sensors can be controlled by a high-speed signal acquisition card with a synchronous clock signal to ensure that the signals of each channel are aligned in time. For example, the sampling frequency of the multi-channel sensor array is set to 10 kHz to capture high-frequency vibration signals and sound characteristic signals.

[0166] The signal decomposition module 20 is used to determine the spectral distribution characteristics of the vibration signal in the abnormal signal, perform multi-scale decomposition on the vibration signal according to the spectral distribution characteristics to obtain a number of sub-band signals, and integrate the sub-band signals according to the energy proportion of the sub-band signals to obtain a multi-band combined signal.

[0167] In this embodiment, an adaptive wavelet transform algorithm is used to perform multi-scale decomposition of the vibration signal based on the spectral distribution characteristics of the vibration signal. The main frequency components are then determined by calculating the energy proportion of each sub-band signal obtained by the decomposition. The main frequency components are then reorganized and integrated to obtain a multi-band combined signal. The multi-band combined signal obtained in this way can better reflect the signal characteristics, that is, it contains the original abnormal characteristics, and is therefore more suitable for subsequent fault diagnosis and feature extraction.

[0168] The signal separation module 30 is used to separate the multi-band combination signal into a plurality of eigenmode components using the VMD algorithm, and integrate the eigenmode components according to the degree of correlation between the eigenmode components and the multi-band combination signal to obtain a reconstructed signal.

[0169] In this embodiment, the noise interference part of the multi-band combination signal is subjected to frequency domain separation processing to determine the potential interference frequency band therein. The multi-band combination signal is then separated by the VMD algorithm to obtain a number of eigenmode components (components). The correlation coefficient of each modal function is calculated to determine its degree of correlation with the original abnormal feature. If the correlation coefficient is lower than a preset correlation degree, the eigenmode component is eliminated, and then the denoised reconstructed signal is obtained based on the eigenmode components whose correlation coefficient is not lower than the preset correlation degree.

[0170] The signal analysis module 40 is used to perform time domain analysis on the reconstructed signal to obtain time domain abnormality features, and to perform frequency domain analysis on the time domain abnormality features to obtain frequency domain distribution features, wherein the time domain abnormality features carry the abnormal time period of the original motor signal, and the frequency domain distribution features carry the distribution law of nonlinear changes in the frequency domain.

[0171] In this embodiment, time domain analysis is performed on the reconstructed signal to obtain several time domain windows and their corresponding time domain statistical parameters. The time domain window containing the abnormal time period is then determined based on the time domain statistical parameters, thereby obtaining a time domain anomaly feature. Frequency domain analysis is then performed on the time domain anomaly feature to extract features from the nonlinear data and obtain the frequency domain distribution pattern of the nonlinear changes in the time domain anomaly feature. That is, the frequency domain distribution range of the frequency domain anomaly feature is determined, thereby obtaining a frequency domain distribution feature. The frequency domain distribution feature obtained in this manner can improve the accuracy and reliability of anomaly detection when performing motor anomaly analysis, providing important technical support for motor fault diagnosis and predictive maintenance.

[0172] The anomaly detection module 50 is used to input the frequency domain distribution features into the trained motor anomaly detection model to obtain the anomaly detection result output by the motor anomaly detection model.

[0173] In this embodiment, the motor abnormality detection model is trained based on the sample motor signals obtained from the test motor in the test environment. It can be understood that the above-mentioned processing is also performed on the sample motor signals, that is, the training samples finally used are the sample frequency domain distribution characteristics of the test motor.

[0174] The sample motor signal of the test motor can be pre-stored in the electronic device or acquired in real time from the test environment. There is no restriction on the specific acquisition method of the sample motor signal, as long as the electronic device can acquire the sample motor signal.

[0175] In this embodiment, an SVM classifier is used to construct a motor anomaly detection model. The motor anomaly detection model learns the characteristic boundaries between the normal state and the abnormal state through training samples, thereby obtaining an anomaly detection result of the motor to be detected.

[0176] The present invention provides an abnormality detection system based on motor vibration sound. The system determines the amplitude of an original motor signal and marks any original motor signal exceeding a preset amplitude as an abnormal signal. The system then performs multi-scale decomposition on the abnormal signal to determine sub-band signals of the main frequency components in the original signal. These sub-band signals are integrated to obtain a multi-band combined signal. A variational mode decomposition algorithm is then used to remove noise interference components from the multi-band combined signal. Time and frequency domain analysis are then performed to extract features from nonlinear data, obtaining the distribution pattern of nonlinear changes in the frequency domain. This allows the frequency domain distribution range of the abnormal features to be determined, resulting in frequency domain distribution features. Finally, the frequency domain distribution features are input into a trained motor abnormality detection model to obtain an abnormality detection result output by the motor abnormality detection model. The motor abnormality detection model is trained based on sample motor signals obtained from a test motor in a test environment and constructed using a support vector machine classifier. This results in more accurate abnormality detection results. The present invention achieves accurate identification of motor abnormalities, effectively extracts multi-dimensional abnormal features during motor operation, significantly improves the accuracy and reliability of abnormality detection, and provides important technical support for motor fault diagnosis and predictive maintenance.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for detecting abnormalities based on the vibration sound of a motor, characterized in that: The method comprises: Acquire the original motor signal during the operation of the motor to be detected, determine the amplitude of the original motor signal, and mark the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal; Determine the spectrum distribution characteristics of the vibration signal in the abnormal signal, perform multi-scale decomposition on the vibration signal based on the spectrum distribution characteristics to obtain a number of sub-band signals, and integrate the sub-band signals based on their energy proportions to obtain a multi-band combined signal; The variational mode decomposition algorithm is used to separate the multi-band combination signal into several eigenmode components. The eigenmode components are integrated according to the degree of correlation between the eigenmode components and the multi-band combination signal to obtain the reconstructed signal. Performing time domain analysis on the reconstructed signal to obtain time domain abnormality features, and performing frequency domain analysis on the time domain abnormality features to obtain frequency domain distribution features; the time domain abnormality features carry the abnormal time period of the original motor signal, and the frequency domain distribution features carry the distribution law of nonlinear changes in the frequency domain; The frequency domain distribution features are input into the trained motor anomaly detection model to obtain the anomaly detection results output by the motor anomaly detection model.

2. The abnormality detection method based on the vibration sound of the motor as claimed in claim 1, characterized in that: The method of determining the spectrum distribution characteristics of the vibration signal in the abnormal signal, performing multi-scale decomposition on the vibration signal according to the spectrum distribution characteristics to obtain a plurality of sub-band signals, and integrating the sub-band signals according to the energy proportions of the sub-band signals to obtain a multi-band combined signal specifically includes: Obtaining the vibration signal from the abnormal signal, performing spectrum analysis on the vibration signal, and obtaining the spectrum distribution characteristics of the vibration signal; According to the spectrum distribution characteristics and the adaptive wavelet transform algorithm, the vibration signal is decomposed into multiple scales to obtain several sub-band signals; The energy proportion of each sub-band signal is determined, the sub-band signals whose energy proportion exceeds a preset proportion are screened out, and the sub-band signals exceeding the preset proportion are integrated into a multi-band combined signal through signal recombination.

3. The abnormality detection method based on the vibration sound of the motor as claimed in claim 1, characterized in that: The variational mode decomposition algorithm is used to separate the multi-band combination signal into a number of eigenmode components, and the eigenmode components are integrated according to the degree of correlation between the eigenmode components and the multi-band combination signal to obtain a reconstructed signal, specifically including: Perform frequency domain separation processing on the multi-band combined signal to obtain a preliminary separation signal containing the original abnormal characteristics; The variational mode decomposition algorithm is used to separate the multi-band composite signal into several eigenmode components; determining a correlation coefficient between each eigenmode component and the original abnormal feature, and determining a correlation degree between each eigenmode component and the multi-band combined signal based on the correlation coefficient; The eigenmode components exceeding the preset correlation degree are integrated to obtain the denoised reconstructed signal.

4. The abnormality detection method based on the motor running vibration sound according to claim 1, characterized in that: The reconstructed signal is subjected to time domain analysis to obtain time domain abnormality features, and the time domain abnormality features are subjected to frequency domain analysis to obtain frequency domain distribution features, specifically including: Segmenting the reconstructed signal according to a preset sliding window to obtain a plurality of time domain windows, and determining time domain statistical parameters within each time domain window; the time domain statistical parameters include mean, variance, deviation, and kurtosis; Determine the change rate of the time domain statistical parameter between time domain windows adjacent in time, and determine the potential abnormal period in the time domain window according to the change rate; Clustering and classifying the time domain windows determined as potential abnormal periods to determine several groups of time domain abnormality features; the time domain windows within each group of the time domain abnormality features have similar statistical characteristics; Frequency domain analysis is performed on each set of time domain anomaly features to identify nonlinear changes in the time domain anomaly features and obtain frequency domain distribution features.

5. The abnormality detection method based on the vibration sound of the motor as claimed in claim 4, characterized in that: The frequency domain analysis is performed on each set of time domain anomaly features to identify nonlinear changes in the time domain anomaly features and obtain frequency domain distribution features, specifically including: Hilbert-Huang transform is used to perform frequency domain decomposition on each set of time domain anomaly features to obtain the instantaneous frequency and instantaneous amplitude information of each set of time domain anomaly features; Construct a time-frequency spectrum of each group of time-domain anomaly features based on the instantaneous frequency and instantaneous amplitude information, and extract the frequency distribution in the time-frequency spectrum; Determining a frequency offset of each frequency distribution, and determining a ratio between the frequency offset and a preset reference frequency, and marking a frequency distribution whose ratio exceeds a preset ratio as a frequency domain abnormal feature; All frequency domain abnormal features are collected to obtain the frequency domain distribution law of the frequency domain abnormal features, feature positioning is performed on the frequency domain distribution law, the distribution law of the frequency domain abnormal features in the frequency domain is obtained, and the frequency domain distribution features are obtained.

6. The abnormality detection method based on the motor running vibration sound according to claim 1, characterized in that: The motor anomaly detection model is trained based on sample motor signals obtained from the test motor in a test environment and is constructed using a support vector machine classifier.

7. The abnormality detection method based on the vibration sound of the motor as claimed in claim 6, characterized in that: The motor anomaly detection model is trained in the following way: Acquire a sample motor signal of the test motor under a test environment, and determine a sample frequency domain distribution feature in the sample motor signal; Obtain normal frequency domain distribution features, use a support vector machine classifier to perform supervised training on the sample frequency domain distribution features and the normal frequency domain distribution features, obtain a preset hyperplane and a preset safety margin, and obtain a trained support vector machine classifier; The trained support vector machine classifier is used to classify and predict the frequency domain distribution characteristics of the samples, determine the distance value from the frequency domain distribution characteristics of the samples to the preset hyperplane, and generate anomaly detection results based on the distance value and the preset safety margin. The anomaly detection results are verified to train a motor anomaly detection model.

8. The abnormality detection method based on the vibration sound of the motor as claimed in claim 1, characterized in that: The method of obtaining an original motor signal during the operation of the motor to be detected, determining the amplitude of the original motor signal, and marking an original motor signal having an amplitude exceeding a preset amplitude as an abnormal signal specifically includes: A multi-channel sensor array is used to obtain the original motor signal during the operation of the motor to be tested; Determine a first amplitude of a vibration signal and a second amplitude of a sound characteristic signal in the original motor signal; The raw motor signal whose first amplitude exceeds a first preset amplitude and / or whose second amplitude exceeds a second preset amplitude is marked as an abnormal signal.

9. The abnormality detection method based on the vibration sound of the motor as claimed in claim 8, characterized in that: The step of marking the original motor signal whose first amplitude exceeds the first preset amplitude and / or whose second amplitude exceeds the second preset amplitude as an abnormal signal specifically includes: marking the original motor signal whose first amplitude exceeds the first preset amplitude and / or whose second amplitude exceeds the second preset amplitude as a candidate signal; Determine the root mean square value of the candidate abnormal signal, and mark the candidate signal whose root mean square value exceeds a preset value as a risk signal; Identify abnormal signals from risk signals based on business scenarios.

10. An abnormality detection system based on the vibration sound of a motor, characterized by: The system comprises: A signal acquisition module is used to acquire the original motor signal during the operation of the motor to be tested, determine the amplitude of the original motor signal, and mark the original motor signal whose amplitude exceeds a preset amplitude as an abnormal signal; The signal decomposition module is used to determine the spectral distribution characteristics of the vibration signal in the abnormal signal, perform multi-scale decomposition on the vibration signal according to the spectral distribution characteristics, obtain a number of sub-band signals, and integrate the sub-band signals according to the energy proportion of the sub-band signals to obtain a multi-band combined signal; A signal separation module is used to separate the multi-band combination signal into a number of eigenmode components using a variational mode decomposition algorithm, and integrate the eigenmode components according to the degree of correlation between the eigenmode components and the multi-band combination signal to obtain a reconstructed signal; A signal analysis module is used to perform time domain analysis on the reconstructed signal to obtain time domain abnormality features, and to perform frequency domain analysis on the time domain abnormality features to obtain frequency domain distribution features; the time domain abnormality features carry the abnormal time period of the original motor signal, and the frequency domain distribution features carry the distribution pattern of nonlinear changes in the frequency domain; The anomaly detection module is used to input the frequency domain distribution features into the trained motor anomaly detection model to obtain the anomaly detection results output by the motor anomaly detection model.

Images