Motor performance detection method and system based on data analysis

Abstract

The invention relates to the field of data processing, in particular to a motor performance detection method and system based on data analysis. The method comprises the steps that potential fault signals in motor vibration signals are separated, a plurality of windows are arranged, the potential fault signals are segmented to obtain segment signals, the segment signals are clustered to obtain a plurality of clustering clusters, and the clustering method comprises the steps that feature vectors of all the segment signals are constructed, the distance between the feature vectors is calculated, and the distance between the feature vectors is calculated; the dimension of the feature vector comprises periodic intensity and frequency band similarity, according to the feature vector, clustering is carried out by using a clustering algorithm to obtain a plurality of clusters, and the clusters comprise an interference signal cluster and / or a plurality of fault clusters. According to the invention, the influence of the interference signal on the fault detection accuracy can be reduced.

Description

Technical Field

[0001] This application relates to the field of data processing, and particularly to a method and system for detecting motor performance based on data analysis. Background Art

[0002] As a core power component in fields such as industry, agriculture, and household appliances, the performance of the motor directly affects the operating efficiency, reliability, and safety of the equipment. Detecting the performance of the motor is a key link in ensuring product quality, extending service life, and preventing failures. A faulty motor may cause energy waste and even lead to serious safety accidents. Therefore, regularly detecting and evaluating the performance of the motor can timely discover potential problems and avoid downtime losses caused by sudden failures.

[0003] The prior art can monitor the operating state of the motor and diagnose faults through various means. Among them, the method based on vibration signal analysis is one of the commonly used techniques. By installing acceleration sensors or displacement sensors at key parts of the motor, vibration data during the operation of the motor can be collected in real time. These data can reflect the dynamic characteristics of internal components of the motor (such as bearings, rotors, stators, etc.) and possible abnormal conditions. For example, when the motor bearing wears or has defects, it will trigger vibration signals at specific frequencies; while rotor imbalance or misalignment will cause the periodic vibration in the low-frequency band to intensify.

[0004] However, in the detection of motor faults, there may be interference signals in the collected signals. The main reasons for the generation of interference signals are external electromagnetic interference, startup or shutdown of equipment, etc. In the actual operating environment of the motor, the existence of interference signals is inevitable. These signals are not caused by problems with the motor itself, and the interference signals affect the detection accuracy of faults. Summary of the Invention

[0005] To solve the problem that the interference signals affect the detection accuracy of faults, this application provides a method and system for detecting motor performance based on data analysis.

[0006] In the first aspect, this application provides a method for detecting motor performance based on data analysis, adopting the following technical solution: A motor performance detection method based on data analysis includes the following steps: segmenting a vibration signal according to a set window, screening out windows with abnormal signals to obtain multiple segment signals; clustering the segment signals to obtain multiple cluster clusters, and distinguishing interference signals from fault signals according to the cluster clusters; wherein the clustering method is: calculating the similarity between any segment signal and other segment signals, taking other segment signals with a similarity greater than a preset threshold as matching signals of any segment signal, and taking the number of matching signals and the total number of segment signals as the period strength of any segment signal; obtaining the union of the frequency band of any segment signal and the standard frequency band, as well as the intersection of the frequency band and the standard frequency band, and taking the ratio of the intersection to the union as the frequency band similarity; constructing feature vectors of each segment signal, calculating the distance between the feature vectors, the dimension of the feature vector includes period strength and frequency band similarity, and clustering using a clustering algorithm according to the feature vector to obtain multiple cluster clusters, the cluster clusters include interference signal cluster clusters and / or several fault cluster clusters, and each fault cluster cluster expresses a fault type.

[0007] The beneficial effects are: segmenting the vibration signal through the window, and preliminarily filtering out the segment signals with possible faults, focusing on the key areas that may contain faults or interference, and reducing the impact of redundant data. Combining the clustering algorithm to classify the segment signals, it realizes the automatic distinction between normal signals, interference signals and fault signals, and improves the detection efficiency.

[0008] The interference signal and fault signal are determined by combining the cycle strength and frequency band. The interference signal usually appears as high-frequency noise, random fluctuations or periodic but irregular signals, and the spectrum may cover a wide frequency range, which is irrelevant to the characteristic frequency of the normal operation of the motor. The fault signal has a specific frequency characteristic.

[0009] The frequency band similarity is defined as the ratio of the intersection and union of the frequency band and the standard frequency band, and the period strength and the frequency band similarity are used as feature vectors for clustering. This can effectively solve the problem of mismatch between the frequency band range form and the period strength numerical form. By constructing feature vectors and clustering, different types of faults (such as bearing faults, rotor faults, etc.) can be distinguished, providing a basis for subsequent fault diagnosis.

[0010] Optionally, a method for calculating the degree of similarity is: calculating the time series distance between the middle signals of the two windows by a dynamic time warping algorithm, and normalizing the time series distance by positive correlation to obtain the degree of similarity.

[0011] The beneficial effects are as follows: The dynamic time warping algorithm can effectively handle the situation where the lengths of time series are inconsistent or there are non-linear stretches, and is suitable for time offsets or periodic changes that may occur in motor vibration signals. By comparing based on the similarity of signal shapes, it can capture the periodic or impact characteristics in the signal, improve the accuracy of calculating the similarity degree, and has a certain tolerance for noise and small-range disturbances, making it suitable for signal analysis under complex working conditions.

[0012] Optionally, the method for calculating the similarity degree is as follows: Construct a data point sequence of any segment of the signal, calculate the Pearson correlation coefficient of the segment signals in two windows, and use the positively correlated normalized Pearson correlation coefficient as the similarity degree.

[0013] The beneficial effects are as follows: The calculation complexity of the Pearson correlation coefficient is low, making it suitable for real-time processing or applications with large-scale data sets.

[0014] Optionally, the method for screening windows with abnormal signals to obtain multiple segment signals is as follows: Calculate the kurtosis of the segment signals , in response to , denotes a hyperparameter, and there are abnormal signals in the segment signals within this window.

[0015] The beneficial effects are as follows: Kurtosis can effectively reflect the thickness of the tail of the signal distribution, is very sensitive to impact signals or abnormal events, and is suitable for detecting sudden faults during motor operation. By adjusting the hyperparameter ϵ , the screening criteria for abnormal signals can be flexibly controlled to meet the requirements of different working conditions. The calculation formula of kurtosis is clear and easy to implement, making it suitable for online monitoring systems.

[0016] Optionally, the method for screening windows with abnormal signals to obtain multiple segment signals is as follows: Calculate the peak factor of the segment signals. When the peak factor exceeds a preset threshold, it is determined that there are abnormal signals in the segment signals within this window.

[0017] The beneficial effects are as follows: It provides another method for judging whether there are abnormal impacts in the signal. The peak factor can effectively reflect the spike components in the signal, is suitable for detecting impact faults (such as bearing faults, gear faults) during motor operation, and does not require complex feature extraction or model training, with simple calculation and easy implementation.

[0018] Optionally, the peak factor is the ratio of the signal peak to the root mean square or the average absolute value.

[0019] The beneficial effects are as follows: The root mean square is a measure of the signal energy and can reflect the overall strength of the signal. When using the ratio of the peak to the root mean square as the peak factor, this method is more sensitive to the energy distribution of the signal and is suitable for detecting high-energy spike signals.

[0020] The average absolute value is the average of the absolute values of the signal amplitudes, which can directly reflect the amplitude characteristics of the signal; when using the ratio of the peak value to the average absolute value as the peak factor, spike signals can also be detected, but this method is more sensitive to changes in the signal amplitude and is suitable for detecting low-amplitude but sharp abnormal signals.

[0021] Optionally, the fault clustering clusters include bearing fault clusters, rotor fault clusters, and / or gear fault clusters.

[0022] The beneficial effect is: refining the fault types, by further dividing the fault signals into types such as bearing faults, rotor faults, and gear faults, the fault source can be more accurately located.

[0023] Optionally, the distance between the feature vectors is the Euclidean distance, cosine similarity, or Manhattan distance.

[0024] Optionally, the clustering algorithm can be the K-means clustering algorithm or the DBSCAN clustering algorithm.

[0025] In a second aspect, the present application provides a motor performance detection system based on data analysis, adopting the following technical solutions: A motor performance detection system based on data analysis, a processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the motor performance detection method based on data analysis as described above is implemented.

[0026] The beneficial effect is: generating a computer program for the motor performance detection method based on data analysis as described above and storing it in the memory to be loaded and executed by the processor, thereby making a system according to the memory and the processor, which is convenient to use.

[0027] The present application has the following technical effects: Segment the vibration signal through a window and initially screen out the segment signals that may have faults, focusing on the key areas that may contain faults or interferences, and reducing the influence of redundant data. Classify the segment signals by combining a clustering algorithm, realizing the automatic distinction of normal signals, interference signals, and fault signals, and improving the detection efficiency.

[0028] Comprehensively judge interference signals and fault signals based on the cycle strength and frequency band. Interference signals usually appear as high-frequency noise, random fluctuations, or periodic but irregular signals, and their spectra may cover a relatively wide frequency range, which has nothing to do with the characteristic frequencies of the normal operation of the motor. Fault signals have specific frequency characteristics.

[0029] By defining the ratio of the intersection to the union of a frequency band and a standard frequency band as the frequency band similarity, and using the periodic intensity and the frequency band similarity as feature vectors for clustering, the problem of mismatch between the frequency band range form and the periodic intensity numerical form can be effectively solved. By constructing feature vectors and clustering, different types of faults (such as bearing faults, rotor faults, etc.) can be distinguished, providing a basis for subsequent fault diagnosis. Description of the Drawings

[0030] Figure 1 is the method for detecting the performance of an electric motor based on data analysis according to an embodiment of the present application.

[0031] Figure 2 is the flowchart of the method in step S2 of the method for detecting the performance of an electric motor based on data analysis according to an embodiment of the present application. Detailed Embodiments

[0032] The embodiment of the present application discloses a method for detecting the performance of an electric motor based on data analysis, which is used to reduce the influence of interference signals on the detection accuracy of electric motor operation faults. The interference signals usually appear as high-frequency noises, random fluctuations or periodic but irregular signals. Their spectra may cover a relatively wide frequency range and are irrelevant to the characteristic frequencies of the normal operation of the electric motor. Fault signals have specific frequency characteristics and are usually associated with the operating parameters of the electric motor (such as rotational speed). For example, bearing faults may exhibit obvious amplitudes at specific multiple frequencies.

[0033] Refer to Figure 1 , including steps S1 - S2, specifically as follows: S1: Segment the vibration signal according to the set window, and screen out the windows with abnormal signals to obtain multiple segment signals.

[0034] A vibration sensor is set at a position where sensing can be performed, such as the motor housing or the rotating shaft. Set the sampling frequency. According to the actual fault frequency range (for example, the highest fault frequency is 200 Hz), set the sampling rate to 500 Hz (meeting the Nyquist theorem, that is, ≥400 Hz), and the data acquisition volume for 10 seconds is 5000 points to ensure signal integrity.

[0035] When the motor is running normally, it will generate certain vibration signals, which usually have periodicity and regularity. When a fault occurs in the motor, the fault signal will be superimposed on the vibration signal during normal operation. Since the fault signal may be non-linear, intermittent or high-frequency, it may be intertwined with the normal signal. The subsequent clustering algorithm depends on the distribution characteristics of the data. First, the normal signal and the potential fault signal in the adopted signal can be separated, and the potential fault signal can be used as the vibration signal for subsequent analysis. The mixed signal can be decomposed by independent component analysis or principal component analysis to separate the normal operation signal and the potential fault signal. Independent component analysis or principal component analysis is a blind source separation technology used to extract independent source signals from the mixed signal. They do not depend on the frequency characteristics of the signal, but decompose the signal based on statistical independence or variance maximization.

[0036] In other embodiments, this step may not be performed, and the sampling signal collected by the sensor can be directly used as the vibration signal for subsequent processing, omitting the step to make the calculation simpler and the amount of calculation smaller.

[0037] Set a window with a fixed or variable length to segment the potential fault signal. Exemplarily, the fixed window length is 100 ms (assuming a sampling rate of 500 Hz, which corresponds to 50 sampling points).

[0038] In one embodiment, the method for screening the windows with abnormal signals to obtain multiple segment signals is: calculate the kurtosis of the segment signal , in response to , represents a hyperparameter, and there is an abnormal signal in the segment signal within this window. Kurtosis measures the thickness of the tail of the signal distribution and is used to detect whether there are transient shocks in the signal. If the kurtosis of the signal is high (positive kurtosis), it indicates that there may be more spikes or sudden events in the signal. In some special cases, the motor may operate in a special steady state. For example, when the load is extremely small or there is no load and the motor is idling, the vibration is mainly composed of slight mechanical noise. In this state, the vibration signal may be close to a uniform distribution or other distributions with lighter tails, resulting in a kurtosis less than 3. The calculation of kurtosis is a prior art and will not be elaborated here. When (such as ), the kurtosis deviates significantly from the Gaussian distribution, then there may be an abnormal signal in this segment signal.

[0039] In one embodiment, the method for screening the windows with abnormal signals to obtain multiple segment signals is: calculate the peak factor of the segment signal. When the peak factor exceeds a preset threshold (such as 5), it is determined that there is an abnormal signal in the segment signal within this window; the peak factor is used to evaluate the extreme degree of the peak in the signal. When the peak factor exceeds the preset threshold, it is determined that there may be an abnormal signal in the window. The peak factor is the ratio of the signal peak to the root mean square or the average absolute value.

[0040] This application preferably uses the ratio of the signal peak value to the root mean square as the peak factor. The root mean square is a measure of the signal energy and can reflect the overall strength of the signal. When using the ratio of the peak value to the root mean square as the peak factor, this method is more sensitive to the energy distribution of the signal and is suitable for detecting high-energy spike signals. The average absolute value is the average of the absolute values of the signal amplitudes and can directly reflect the amplitude characteristics of the signal; when using the ratio of the peak value to the average absolute value as the peak factor, spike signals can also be detected, but this method is more sensitive to changes in the signal amplitude and is suitable for detecting low-amplitude but sharp abnormal signals.

[0041] In other embodiments, it is also possible to comprehensively measure based on kurtosis and the peak factor. If the peak factor exceeds the threshold and the kurtosis deviates significantly from the Gaussian distribution, it is marked that there is an abnormal signal in this section of the signal.

[0042] S2: Cluster the section signals to obtain multiple clusters, and distinguish interference signals and fault signals according to the clusters.

[0043] By performing clustering analysis on the segmented signal data, the signals are divided into normal signals, interference signals, and different types of fault signals (such as bearing faults, rotor faults, etc.), thereby providing a basis for subsequent fault diagnosis. Refer to Figure 2 , the clustering method includes steps S20 - step S22, which are as follows: S20: Calculate the similarity degree between any section signal and other section signals, use the other section signals with a similarity degree greater than the preset threshold as the matching signals of any section signal, and use the number of matching signals and the total number of section signals as the cycle strength of any section signal.

[0044] In one embodiment, the calculation method of the similarity degree is: calculate the time series distance between the section signals in two windows through the dynamic time warping algorithm, and use the positively correlated normalization of the time series distance as the similarity degree.

[0045] The dynamic time warping algorithm can compare two time series and can handle the situation where the lengths of the time series are inconsistent or non-linearly stretched. The prior art will not be elaborated here. Positively correlated normalization limits the value range of the time series distance to , while keeping the relative relationship between the data unchanged, and can unify the values with different dimensions or scales into the same range for subsequent analysis or comparison. Positively correlated normalization can be min-max normalization, Z-Score standardization, etc., and the prior art will not be elaborated here.

[0046] This application preferably calculates the similarity degree through the dynamic time warping algorithm. The non-linear alignment ability and shape matching ability of the dynamic time warping algorithm are more suitable for such complex working conditions. Since a fixed window is adopted in this application and the lengths of segment signals are basically the same, the similarity degree between two segment signals can also be calculated according to the Pearson correlation coefficient.

[0047] Specifically, the calculation method of the similarity degree is as follows: construct a data point sequence of any segment signal, calculate the Pearson correlation coefficient of the segment signals in two windows, and use the Pearson correlation coefficient after positive correlation normalization as the similarity degree. If a certain data point in the segment signal is missing, the interpolation method of the prior art can be used to supplement the missing data and then calculate the Pearson correlation coefficient.

[0048] The Pearson correlation coefficient is used to measure the linear correlation between two variables, and its value range is , if two segment signals are highly similar, the Pearson correlation coefficient is close to 1; if two segment signals are completely different, the Pearson correlation coefficient is close to 0; if two segment signals are negatively correlated, the Pearson correlation coefficient is close to . Since the similarity degree needs to reflect the positive correlation between signals, only the positive value part of the Pearson correlation coefficient (i.e., within the range of ) is concerned, and the Pearson correlation coefficient is converted into a standardized similarity value to ensure that its range is , and the influence of negative correlation is eliminated.

[0049] So far, the similarity degree between any segment signal and other segment signals is obtained. The other segment signals with a similarity degree greater than a preset threshold (such as 0.7 or 0.8) are used as the matching signals of any segment signal, and the number of matching signals and the total number of segment signals are used as the periodic intensity of any segment signal. A lower periodic intensity indicates a weaker periodicity of the signal, and at this time, the possibility of it being an interference signal is higher. On the contrary, the possibility of it being an interference signal is lower.

[0050] S21: Obtain the union of the frequency band of any segment signal and the standard frequency band, as well as the intersection of the frequency band and the standard frequency band, and use the ratio of the intersection to the union as the frequency band similarity.

[0051] Generally, a fault signal will generate periodic impact signals, while random noise or other external interferences usually do not have obvious periodicity. However, relying solely on the periodic intensity may not be able to accurately distinguish between fault signals and interference signals. Some specific types of interference (such as periodic electromagnetic interference) may cause the periodic intensity of the signal to increase. Therefore, it is also necessary to make a comprehensive judgment by combining the index of frequency band similarity.

[0052] The spectrum distribution of interference signals is usually random and may cover a wide frequency range. Specific frequency components, such as high-frequency impact components or new frequency components, usually appear in the spectrum of fault signals. Therefore, the period strength and frequency band characteristics are combined to comprehensively determine the fault signal and the interference signal.

[0053] However, since the frequency band is a range of values ​​and the period strength is a value, if the subsequent feature vector is constructed directly based on the frequency band and the period strength, the dimensions of the data are inconsistent. The range form of the frequency band does not match the numerical form of the period strength and cannot be directly used for subsequent clustering. Therefore, this application calculates the union of the frequency band and the standard frequency band of any signal segment and the intersection of the frequency band and the standard frequency band, and takes the ratio of the intersection to the union as the frequency band similarity.

[0054] The ratio of the intersection and union of the frequency band and the standard frequency band is defined as the frequency band similarity, which can effectively solve the problem of mismatch between the frequency band range form and the period intensity numerical form. Regarding the calculation of frequency band similarity, for example, the frequency band is , the standard frequency band is , frequency band similarity = 40-20 / 50-10 = 0.5.

[0055] The standard frequency band expresses the standard frequency band of the fault, which can reflect the frequency band of the signal corresponding to the typical fault. There may be multiple subsequent fault clusters, so the number of standard frequency bands can also be multiple. For example, the unbalanced frequency is usually the fundamental frequency corresponding to the motor speed (such as 50Hz), and the frequency multiplication components may appear at 2 times or 3 times the speed frequency. The gear fault signal has strong periodicity, and the meshing frequency and its harmonics appear in the spectrum. The meshing frequency is determined by the number of teeth and the speed of the gear, and the harmonic components may appear at integer multiples of the meshing frequency. The present application can set the standard frequency band according to the frequency band corresponding to the historical fault signal.

[0056] S22: construct the feature vector of each signal segment, calculate the distance between the feature vectors, the dimension of the feature vector includes periodic intensity and frequency band similarity, and use a clustering algorithm to cluster the feature vectors to obtain multiple clusters, which include interference signal clusters and / or several fault clusters. Each fault cluster expresses a fault type.

[0057] The distance between feature vectors can be Euclidean distance, cosine similarity or Manhattan distance. The distance between feature vectors reflects the degree of difference between signals. The smaller the distance, the more similar the two signals are; the larger the distance, the greater the difference between the two signals. Clustering is performed using a clustering algorithm to obtain multiple clusters.

[0058] The clustering algorithm can be the K-means clustering algorithm or the DBSCAN clustering algorithm. The prior art will not be elaborated here. The K-means algorithm requires specifying the number of clusters in advance. However, in practical applications, the types of faults may be unknown, and the number of clusters cannot be accurately set, resulting in relatively low accuracy of the results. This application preferably uses the DBSCAN clustering algorithm, which can automatically identify the number of clusters and is more suitable for processing clusters with irregular shapes.

[0059] The clustering result is an interference signal clustering cluster and / or several fault clustering clusters.

[0060] Interference signal clustering cluster: low periodic intensity and low frequency band similarity. The signal does not have obvious periodic characteristics, and its spectrum distribution is random. Among them, the impact signal may be caused by the normal operating signal. During the normal operation of the motor, due to factors such as its mechanical structure, load change, or speed fluctuation, some short-term and high-frequency vibration impact signals may be generated. These impact signals do not necessarily mean that there is a fault, but may be part of the normal operation of the motor and are also classified as interference signals.

[0061] Fault clustering cluster: strong periodic intensity and high frequency band similarity. The data in the fault clustering cluster has a high similarity with the standard frequency band, reflecting the existence of typical fault characteristics in the signal, and these characteristics highly match the known fault types. This high similarity indicates that the spectrum distribution of the signal is concentrated in specific frequency ranges, which are consistent with the standard frequency band (learned by the fault classification model), thereby effectively distinguishing fault signals from interference signals. The number of fault clustering clusters may be multiple. Exemplarily, the fault clustering clusters can be bearing fault clusters, rotor fault clusters, and / or gear fault clusters.

[0062] The embodiment of this application also discloses a motor performance detection system based on data analysis, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the motor performance detection method based on data analysis according to this application.

[0063] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. Their settings and functions are known in the art, so they will not be elaborated here.

[0064] In this application, the foregoing memory may be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as, a resistive random access memory, a dynamic random access memory, a static random access memory, etc., or any other medium that can be used to store the required information and can be accessed by an application program, a module, or both. Any such computer storage medium may be part of the device or accessible or connectable to the device.

[0065] The above are all the preferred embodiments of this application, and the protection scope of this application is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of this application shall be covered within the protection scope of this application.

Claims

1. A motor performance detection method based on data analysis, characterized in that: Includes steps: The vibration signal is segmented according to the set window, and the window with abnormal signal is screened out to obtain multiple segment signals; the segment signals are clustered to obtain multiple clusters, and the interference signal and the fault signal are distinguished according to the clusters; The clustering method is as follows: calculate the similarity between any segment signal and other segment signals, take other segment signals with a similarity greater than a preset threshold as matching signals of any segment signal, and take the ratio of the number of matching signals to the total number of segment signals as the periodic strength of any segment signal; Obtain the union of the frequency band of any signal segment and the standard frequency band and the intersection of the frequency band and the standard frequency band, and use the ratio of the intersection to the union as the frequency band similarity; Construct the feature vector of each signal segment and calculate the distance between the feature vectors. The dimension of the feature vector includes periodic intensity and frequency band similarity. According to the feature vector, clustering algorithm is used to obtain multiple clusters. The clusters include interference signal clusters and / or several fault clusters. Each fault cluster expresses a fault type.

2. The motor performance detection method based on data analysis according to claim 1 is characterized in that: The similarity is calculated by using a dynamic time warping algorithm to calculate the time series distance between the mid-term signals in the two windows, and then normalizing the time series distance to obtain the similarity.

3. The motor performance detection method based on data analysis according to claim 1 is characterized in that: The similarity degree is calculated by constructing a data point sequence of any signal segment to calculate the Pearson correlation coefficient of the mid-range signals in two windows, and normalizing the Pearson correlation coefficient to positive correlation as the similarity degree.

4. The motor performance detection method based on data analysis according to claim 1 is characterized in that: The method of filtering the window with abnormal signals to obtain multiple segment signals is: calculating the kurtosis of the segment signal , in response to , Represents a hyperparameter, and the segment signal within this window contains abnormal signals.

5. The motor performance detection method based on data analysis according to claim 1 is characterized in that: The method for screening the window with abnormal signals to obtain multiple segment signals is: calculating the peak factor of the segment signal, and when the peak factor exceeds a preset threshold, determining that the segment signal in the window has an abnormal signal.

6. The motor performance detection method based on data analysis according to claim 5 is characterized in that: The crest factor is the ratio of the signal's peak value to its RMS or mean absolute value.

7. The motor performance detection method based on data analysis according to claim 1 is characterized in that: The fault clustering clusters include bearing fault clusters, rotor fault clusters and / or gear fault clusters.

8. The motor performance detection method based on data analysis according to claim 1 is characterized in that: The distance between feature vectors is Euclidean distance, cosine similarity, or Manhattan distance.

9. The motor performance detection method based on data analysis according to any one of claims 1 to 8, characterized in that: The clustering algorithm is the K-means clustering algorithm or the DBSCAN clustering algorithm.

10. The motor performance detection system based on data analysis is characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the motor performance detection method based on data analysis according to any one of claims 1 to 9 is implemented.

Images