Electromechanical equipment fault diagnosis system and method based on machine learning

Abstract

The invention relates to the technical field of motor testing, in particular to an electromechanical equipment fault diagnosis system and method based on machine learning, and the system comprises a high-frequency signal injection module which injects a high-frequency signal in the range of 10 kHz to 50 kHz into a motor through a power supply line, analyzes the electromagnetic response of the signal, and obtains electromagnetic parameter monitoring data; and based on the electromagnetic parameter monitoring data, statistical characteristics are extracted and pattern recognition is carried out, and an early fault characteristic signal is generated. According to the invention, through high-frequency signal injection and differential signal amplification, early recognition and accurate positioning of motor faults are realized. By injecting high-frequency signals, changes in electromagnetic parameters can be captured, which often occur prior to physical fault performance. Differential amplification further enhances the ability to extract useful signals from these small changes, which can be identified and responded before the fault develops to a more serious stage.

Description

Technical Field

[0001] The present invention relates to the technical field of motor testing, and in particular to a machine learning-based electromechanical equipment fault diagnosis system and method. Background Art

[0002] The field of motor testing technology focuses on the performance analysis, fault diagnosis and maintenance of motors and their related systems. This technology uses various test methods and tools to evaluate the electrical and mechanical characteristics of motors, including resistance testing, insulation testing, vibration analysis, temperature monitoring and more advanced diagnostic techniques such as motor current signature analysis (MCSA) and thermal imaging technology. The main purpose of motor testing is to ensure the efficiency of equipment operation, prevent failures, extend equipment life and improve safety. The electromechanical equipment fault diagnosis system is designed to identify and analyze potential faults of electromechanical equipment (such as motors, generators and other rotating equipment) in real time.

[0003] Although existing motor testing technologies are capable of evaluating and maintaining motor performance, they are insufficient in quickly responding to environmental changes and diagnosing early faults. Existing methods rely on physical testing and parameter monitoring, which can only detect problems when the fault has already begun to affect machine operation. For example, vibration analysis and temperature monitoring show obvious abnormalities only after the problem occurs, resulting in a slow response to rapidly developing faults. This delayed fault response not only increases maintenance costs, but may also lead to greater equipment damage or safety risks due to the failure to solve small problems in a timely manner. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose a machine learning-based electromechanical equipment fault diagnosis system and method.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: A mechanical and electrical equipment fault diagnosis system based on machine learning includes: A high-frequency signal injection module injects a high-frequency signal in the range of 10kHz to 50kHz into the motor through the power supply line, analyzes the electromagnetic response of the signal, and obtains electromagnetic parameter monitoring data; based on the electromagnetic parameter monitoring data, extracts statistical features and performs pattern recognition to generate early fault characteristic signals; The dynamic feature analysis module transforms the early fault feature signal, extracts the envelope analysis data of the signal, and establishes the fault signal envelope analysis result; processes the fault signal envelope analysis result by differential signal amplification to obtain the differential amplification fault feature; A fault diagnosis module, based on the differential amplification fault feature, constructs a feature vector and inputs it into a support vector machine classifier to calculate and obtain a fault identification result; The fault threshold dynamic adjustment module adjusts the fault diagnosis threshold in real time according to the motor type and operating conditions, and establishes dynamic fault threshold parameters.

[0006] Preferably, the steps of acquiring the electromagnetic parameter monitoring data are: A high-frequency signal in the range of 10kHz to 50kHz is injected into the motor through the power line, and the collected electromagnetic waveform data is used as the basis for analysis to form initial electromagnetic response data; The initial electromagnetic response data is subjected to spectrum analysis to calculate the electromagnetic parameter intensity, and the calculation formula is: in, Represents the frequency The electromagnetic parameter intensity under Representative time point at frequency The signal strength, For the The acquisition time point of each signal, is the total number of signal acquisitions; Based on the electromagnetic parameter intensity, the electromagnetic response characteristics of the signal are analyzed to obtain electromagnetic parameter monitoring data.

[0007] Preferably, the step of acquiring the early fault characteristic signal is: Based on the electromagnetic parameter monitoring data, outliers are removed and normalized to obtain preprocessed electromagnetic data; Based on the pre-processed electromagnetic data, the statistical measure of the fault characteristics is extracted, and the calculation formula is: in, A statistical measure that characterizes the fault, After processing data points, for The average value of for The standard deviation of is the total number of samples; Based on the statistical measurement, a decision tree is applied to perform pattern recognition, identify fault characteristic signals, and obtain early fault characteristic signals.

[0008] Preferably, the steps of obtaining the fault signal envelope analysis result are: Based on the early fault characteristic signal, a Hilbert transform is performed to map the real signal at each time point to a complex plane to obtain an analytical signal; Based on the analytical signal, the envelope of the signal is calculated, and the calculation formula is: in, represents the envelope of the signal, is the signal strength at time j, is the phase angle at time j, is the total number of sample points; Based on the signal envelope, continuous wavelet transform is applied to perform multi-scale analysis to identify abnormal changes in the signal and form the fault signal envelope analysis results.

[0009] Preferably, the step of acquiring the differential amplifier fault feature is: Based on the fault signal envelope analysis result, differential signal amplification is performed to obtain an amplified fault signal; Based on the amplified fault signal, a high-pass filter is applied to generate a differential amplified fault signal by filtering out low-frequency components in the signal and retaining high-frequency components; Based on the differential amplification fault signal, a feature extraction process is performed to identify fault features by detecting and analyzing peak values ​​and waveform patterns in the signal to form differential amplification fault features.

[0010] Preferably, the steps of obtaining the fault identification result are: Collecting the differential amplification fault features, performing standardization processing, and obtaining a preprocessed feature vector; Based on the preprocessed feature vector, a support vector machine classifier is configured and trained, wherein the support vector machine classifier uses a radial basis function as a kernel function to obtain a configured and trained SVM classifier; Based on the configuration and the trained SVM classifier, the test data is fault classified, the fault type is identified and classified, and a fault identification result is formed.

[0011] Preferably, the steps of obtaining the dynamic fault threshold parameter are: Collect information about the motor's current load level, operating speed, temperature, and motor type to form an input data set for fault diagnosis; Based on the input data set of the fault diagnosis, a new fault diagnosis threshold is calculated, and the calculation formula is: in, is the new fault diagnosis threshold, is the initial threshold, , , represent load level, speed and temperature respectively, , , is the adjustment factor, is the normalization factor; According to the new fault diagnosis threshold, the fault diagnosis threshold setting is updated to form a dynamic fault threshold parameter.

[0012] The present invention provides a method for diagnosing mechanical and electrical equipment faults based on machine learning, comprising the following steps: Based on the electromagnetic parameter monitoring data, the statistical characteristics of the electromagnetic parameters are extracted, the fault mode is identified using the statistical characteristics, the abnormal signal mode is obtained through pattern recognition, and the early fault characteristic signal is generated; Perform frequency domain transformation on the early fault characteristic signal, extract the signal envelope data, perform signal envelope analysis, generate signal envelope analysis results, use differential amplification to process the signal envelope analysis results, and obtain differential amplification fault characteristics; The differential amplifier fault feature is used as input data to construct a feature vector, which is input into the support vector machine classifier. The support vector machine classifies the input data, calculates and obtains the fault probability and type, and outputs the fault identification result. According to the fault identification result, referring to the type and operating conditions of the motor, the fault diagnosis threshold is dynamically adjusted, the fault threshold is updated in real time, the fault judgment parameters are reset, and the dynamic fault threshold parameters are established and updated.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, early identification and precise positioning of motor faults are achieved through high-frequency signal injection and differential signal amplification. By injecting high-frequency signals, changes in electromagnetic parameters can be captured, which usually occur before physical faults manifest. Differential amplification further enhances the ability to extract useful signals from these tiny changes, and can identify and respond to faults before they develop into more serious stages. In addition, the method of dynamically adjusting the fault diagnosis threshold automatically adjusts the threshold according to real-time operating data, improves adaptability and response speed in a changing operating environment, and thus has advantages in ensuring equipment safety and operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0016] See also Figure 1 The present invention provides a technical solution: a mechanical and electrical equipment fault diagnosis system based on machine learning comprises: The high-frequency signal injection module injects a high-frequency signal in the range of 10kHz to 50kHz into the motor through the power supply line, analyzes the electromagnetic response of the signal, and obtains electromagnetic parameter monitoring data; based on the electromagnetic parameter monitoring data, it extracts statistical features and performs pattern recognition to generate early fault characteristic signals; The dynamic feature analysis module transforms the early fault feature signal, extracts the signal envelope analysis data, and establishes the fault signal envelope analysis result; it uses differential signal amplification to process the fault signal envelope analysis result and obtains the differential amplification fault feature; The fault diagnosis module constructs a feature vector based on the differential amplification fault feature and inputs it into the support vector machine classifier to calculate and obtain the fault identification result; The fault threshold dynamic adjustment module adjusts the fault diagnosis threshold in real time according to the motor type and operating conditions, and establishes dynamic fault threshold parameters.

[0017] The steps for obtaining electromagnetic parameter monitoring data are as follows: A high-frequency signal in the range of 10kHz to 50kHz is injected into the motor through the power line, and the collected electromagnetic waveform data is used as the basis for analysis to form initial electromagnetic response data; Perform spectrum analysis on the initial electromagnetic response data and calculate the electromagnetic parameter intensity. The calculation formula is: in, Represents the frequency The electromagnetic parameter intensity under Representative time point at frequency The signal strength, For the The acquisition time point of each signal, is the total number of signal acquisitions; Based on the electromagnetic parameter intensity, the electromagnetic response characteristics of the signal are analyzed to obtain electromagnetic parameter monitoring data.

[0018] Specifically, the execution process of injecting high-frequency signals through power lines includes setting a signal generating device with a transmission frequency in the range of 10kHz to 50kHz at the power line port of the motor, ensuring that different frequency bands are covered in segments within the injection range, recording the electromagnetic waveform data at the corresponding time and synchronously recording the injection power, injection duration and sampling time and other information, and then comparing each record with the pre-established upper and lower frequency limits for verification, for example, comparing the injection frequency with the 10kHz to 50kHz range, and comparing the injection power with the RF power range of 0W to 5W, recording all data sets that meet the above range requirements and storing them as original electromagnetic waveform data, in order to check for drift or distortion of the injected signal, monitoring the working stability of the injection device and performing frequency recovery when deviation occurs, temporarily marking the data with obvious drift in the statistical record, combining the information of all sampling times, and concatenating a complete electromagnetic waveform data set in chronological order to summarize the initial electromagnetic response data.

[0019] The benefit of the formula is that by fusing time domain and frequency domain information, the electromagnetic response characteristics of the motor at a given frequency can be quantitatively analyzed, so that fault characteristics can be identified for different frequency components in subsequent steps; The acquisition steps are as follows: at each sampling moment, k pairs of frequencies are The component amplitude is analyzed, and the amplitude at that moment is extracted according to the electromagnetic waveform data collected previously, and mapped to The specific value needs to be calculated based on the sampling rate and amplitude. For example, under the condition of sampling rate of 400kS / s, if the amplitude of the signal collected at a certain moment is 1.2V, the recorded ; The acquisition step is to read the time value from the time base record at the sampling time k. For example, at the kth sampling point, the time record is ,in Provided by the sampling rate, for example, a sampling rate of 400 kS / s corresponds to ; The acquisition step is to obtain the total amount of data by multiplying the sampling time by the sampling rate, and use it as the total number of signal acquisitions. For example, when sampling 1.28ms, you can get points; Calculation process: According to frequency Extract all , and based on For index terms Calculate, assuming that Hz, , then at a certain k value When s, we have: Compare this value with Multiply them together and add them up; Divide the accumulated result by , and then take the square of its complex modulus, as The value of Repeat the above calculation process, traverse different f and obtain a series of ; In the specific example, if k=50 , the exponential term is about 0.997 after calculation, and the product is accumulated into the partial sum. After all 512 sampling points are accumulated, it is divided by 512 and then modulo squared to obtain ; The result shows that at f=30000Hz, the electromagnetic parameter intensity is about 1.14, which is used to further determine whether there is a possibility of a significant increase in amplitude at the corresponding frequency point.

[0020] The execution process based on electromagnetic parameter strength requires traversing different frequencies. After that, all the obtained strength values ​​are sorted according to the frequency index, and the sections with higher strength are collected and compared to see if their values ​​exceed the strength threshold set in advance with reference to the historical fault data. For example, the strength threshold is set between 0.9 and 1.2. If the strength value calculated at a certain frequency point is If it exceeds 1.2, it will be marked in the intensity anomaly list. If it is lower than 0.9, it is in the low intensity range and is recorded routinely. At the same time, the recorded intensity value needs to be compared with the control data collected under the same conditions in the early stage. For example, the intensity recorded at 30kHz is compared with the intensity distribution in the historical record range of 20kHz to 40kHz. If this intensity shows a continuous increasing trend, it is classified as a potential anomaly. The time series waveform of this section is analyzed simultaneously to identify whether there are stray peaks or bandwidth mutations. Finally, the extracted intensity and time series characteristics are summarized to form electromagnetic response characteristic data, and the electromagnetic parameter monitoring data is summarized.

[0021] The steps to obtain the early fault characteristic signal are: Based on the electromagnetic parameter monitoring data, outliers are removed and normalized to obtain preprocessed electromagnetic data; Based on the preprocessed electromagnetic data, the statistical measurement of the fault characteristics is extracted, and the calculation formula is: in, A statistical measure that characterizes the fault, After processing data points, for The average value of for The standard deviation of is the total number of samples; Based on statistical measurement, decision tree is applied to perform pattern recognition, identify fault characteristic signals, and obtain early fault characteristic signals.

[0022] Specifically, based on the electromagnetic parameter monitoring data, it is necessary to remove outliers and perform normalization processing to obtain preprocessed electromagnetic data. During the execution process, all previously collected electromagnetic parameter monitoring data are first arranged in chronological order and the original values ​​of each sampling point are read in turn, and compared with the established multi-dimensional detection range, where, for example, the current is compared with the 0A to 5A range, the voltage is compared with the 0V to 24V range, the temperature is compared with the 0°C to 90°C range, etc., and the values ​​exceeding the above range are marked as abnormal and the sampling points and corresponding timestamps where these abnormal marks appear are recorded. Subsequently, the overall abnormal distribution of the data segment is statistically analyzed based on the abnormal records accumulated during multiple sampling processes, and the samples that obviously deviate from the mainstream distribution are eliminated or stored separately to compare the abnormal characteristics under different situations. After the outliers are removed, the remaining data is numerically standardized using the normalization method. This processing process requires the calculation of each electromagnetic parameter separately. The maximum and minimum values ​​of the number are determined and the distribution interval is confirmed. The data is then mapped to the interval of 0 to 1 using a linear normalization method. Multiple rounds of inspection are carried out in combination with the distribution characteristics of various electromagnetic parameters to prevent the distortion of the normalized results caused by the omission of very few abnormal isolated points. In the process, the minimum, maximum and average values ​​of the samples before and after normalization are cross-compared. If the values ​​show a significant deviation or unreasonable concentration trend, the recorded abnormal points and their surrounding sampling values ​​are checked again to identify the parts that may be incorrectly marked, and a relatively clean normalized data set with good distribution coherence is obtained. In order to facilitate the subsequent fault feature extraction, the normalized data is smoothed according to the time dimension, the moving average is calculated in each sampling window, and the positions with large fluctuations are observed in a graded manner, so as to obtain a relatively stable electromagnetic waveform sequence. Finally, the time series data obtained by the above-mentioned removal of outliers and normalization processing flow are integrated and output as preprocessed electromagnetic data.

[0023] The benefit of the formula is that it can reflect the change in the sharpness of the distribution of electromagnetic data through the cubic ratio between the absolute deviation and the standard deviation, and can identify the sudden change of the fault characteristics in the numerical distribution by combining with the total number of samples; The acquisition steps are as follows: The sampling point value is defined , each point corresponds to a specific electromagnetic amplitude; The steps to obtain are: Sum and divide by , the value is obtained from the measured data; The steps to obtain are to first calculate according to the statistical method The sum is then divided by , then take the square root to get the standard deviation; The acquisition step is to count the number of sampling points of the current preprocessed data, which comes from the time series segmentation result of the aforementioned electromagnetic parameter monitoring data acquisition; Calculation process: Setting the actual test Sample points are obtained from monitoring data , ; Traversal From 1 to 400, calculate point by point And divide it by 0.12, then cube the result of the ratio, which is recorded as ; All Add and divide by 400 to get , the process can be expressed as: Select a few values ​​from the sample for example. hour ,but , 0.06 divided by 0.12 is 0.5, 0.5 cubed is 0.125, after accumulating all sample points in sequence, if the result is about 162, then ; The results show that the cusp metric of the electromagnetic data at the current sample size is about 0.405, which is in the medium dispersion range. A value greater than 0.8 indicates that the distribution has a more obvious peak component, while a value less than 0.2 indicates that the data is relatively smooth and stable.

[0024] Based on statistical metrics, it is necessary to apply decision trees for pattern recognition to identify fault characteristic signals. During the execution process, the statistical metrics are first It is correlated with the corresponding electromagnetic data sequence and a set of characteristic values ​​is extracted for each sampling window. For example, multiple parameters such as T value, current amplitude range and voltage fluctuation amplitude are recorded in each time slice. Then, the training data set used in the decision tree definition is queried and the splitting threshold configured on the decision tree node is compared to obtain the statistical measure. The combined values ​​of each parameter are compared with the threshold range summarized in advance one by one. For example, the interval of T is divided into three levels: 0 to 0.3, 0.3 to 0.6, and 0.6 to 1.0, so as to judge the branch path through the tree node. For some other parameters, corresponding intervals are also set up and compared item by item on each tree node. When the corresponding condition is triggered, a branch is selected, and the next level of nodes are continued to be deepened. When the leaf node is finally reached through layer-by-layer splitting, the fault judgment label is obtained. Then, the multiple sampling windows assigned to the same leaf node are aggregated and a statistical list of fault areas is generated. If it is found that the labels of a certain time period show abnormal characteristics, they are marked as fault characteristic signal time slices. Combined with the high-frequency and low-frequency characteristic distributions of electromagnetic data, the distribution positions of each abnormal feature in the time series are listed. Then, these abnormal features are integrated and their cumulative occurrence times in the continuous sampling period are confirmed. It is analyzed whether they continue to reach the specified condition threshold. For example, when the statistical metric is higher than 0.7 and the current amplitude exceeds 3.5A and is continuously sampled for multiple times, it can be directly classified as a fault characteristic signal. Finally, these judgment results are summarized as early fault characteristic signals.

[0025] The steps to obtain the fault signal envelope analysis results are as follows: Based on the early fault characteristic signal, Hilbert transform is performed to map the real signal at each time point to the complex plane to obtain the analytical signal; Based on the analytical signal, the envelope of the signal is calculated using the following formula: in, represents the envelope of the signal, is the signal strength at time j, is the phase angle at time j, is the total number of sample points; Based on the signal envelope, continuous wavelet transform is applied to perform multi-scale analysis to identify abnormal changes in the signal and form the fault signal envelope analysis results.

[0026] Specifically, based on the early fault characteristic signal, combined with the electromagnetic timing data and fault characteristic information obtained previously, first read the corresponding real signal value one by one according to the recorded time index and calibrate its amplitude range and sampling time at each reading. By comparing the pre-established valid range, such as the current range of 0A to 5A, the voltage range of 0V to 24V, etc., the sampling value is marked to ensure that there is no missed reading or wrong reading. Then, all real signal values ​​are matched point by point in time series order and mapped to the complex plane. At this time, it is necessary to determine the phase data at the same time for each sampling point. If it is found that the phase of some sampling points deviates significantly from the main distribution during the phase measurement process, it is necessary to list the sampling point separately in an index and review several sampling points around it. Next, the amplitude and phase data of each sampling point are inferred. The coordinate position of the sampling point in the complex plane is derived, the amplitude is used as the modulus and the phase is used to define the polar angle, and then all the sampling points are merged to obtain a complex signal sequence, and whether there is an obvious jump or intermittent record in the entire signal sequence, by arranging these complex sequences in order and performing statistics on the distance between adjacent sampling points, to check whether there is a phase discontinuity caused by a sudden increase or decrease in the sampling points, after completing the coordinate mapping of all sampling points, a data structure on the complex plane is established, and an analytical signal is generated and stored in a subsequent list of analytical signals to be used in an indexed manner corresponding to the original time series, and finally the analytical signal is associated with the previously saved early fault characteristic signal, and the corresponding relationship between the two at the same time series position is recorded, so that these analytical signals can be further processed in other steps to obtain analytical signals.

[0027] The benefit of the formula is that by using the cumulative average processing of the components of the real signal on the complex plane, the amplitude and phase of different sampling points can be unified into a concise envelope amplitude measurement, so as to provide a more intuitive reference when judging the high-frequency and low-frequency components in the subsequent steps; The steps to obtain are: read the time The amplitude value at the time is provided by the amplitude field in the analytical signal previously mapped to the complex plane. The specific value is obtained by the amplitude detection instrument in the previous test, for example, at time The recorded amplitude was 0.83; The steps to obtain are, from the same moment The phase information is read and scaled uniformly from 0 to The value comes from the phase detection result performed previously, for example, at time Phase measurement radian; The steps to obtain are to count all valid sampling points and exclude the number of abnormal points previously marked as deviating from the main distribution, and then use the obtained count as For example, after testing, there are about 300 remaining time points. ; Calculation process: Each moment Sum and divide Get A in the form of: Each moment Sum and divide Get B in the form of: make , calculated according to the above definition, if in a given actual example, there is , and through statistics we get: but: and then: This result indicates that the current signal envelope amplitude value is about 0.80. If the same operation is performed on different time periods in the future and similar or higher amplitude values ​​are obtained, it means that there are envelope components with stronger amplitudes in those time periods; Based on the signal envelope, continuous wavelet transform is applied to perform multi-scale analysis to identify abnormal changes in the signal. During the execution process, the signal envelope sequence obtained previously and the amplitude value at the corresponding moment are first listed one by one in the same data table, and the phase information, electromagnetic vibration frequency and possible associated current reference value at that moment are retained in the table. At the same time, before the continuous wavelet transform operation, this part of the signal envelope should be divided into frequency bands and the frequency range should be segmented from lower frequency to higher frequency. The amplitude change in the envelope sequence is read for each segment in turn and compared with the corresponding range of the set multi-scale frequency band. For example, 10Hz to 500Hz is called the low frequency band, 500Hz to 2000Hz is called the medium frequency band, and 2000Hz to 10000Hz is called the high frequency band. Then, in each segment, The envelope amplitude distribution is smoothed and the local average amplitude and center frequency are calculated. Compared with the recorded statistical distribution data, if the envelope amplitude in a certain section is found to increase continuously and exceed the established segmentation threshold, it will be marked. The threshold is set based on the vibration and noise data collected during the continuous operation of the machine for 300 hours. The envelope amplitude of the same frequency band is summarized and the quantiles are obtained. The signal segments at these marks are input into the wavelet transform process for multi-scale expansion. The continuous evolution of the amplitude on the time axis is observed at different scales. The possible transient spikes are compared multiple times through the envelope distribution of adjacent scales. In combination with the parallel recording content of current and voltage in the reference time series, it is checked whether there is a synchronous anomaly. The fault signal envelope analysis results are formed by further inspection of these abnormal points or sections.

[0028] The steps to obtain the differential amplifier fault characteristics are: Based on the fault signal envelope analysis result, differential signal amplification is performed to obtain an amplified fault signal; Based on the amplified fault signal, a high-pass filter is applied to generate a differential amplified fault signal by filtering out the low-frequency components in the signal and retaining the high-frequency components; Based on the differential amplification fault signal, a feature extraction process is performed to identify the fault characteristics by detecting and analyzing the peak value and waveform mode in the signal to form the differential amplification fault signature.

[0029] Specifically, based on the fault signal envelope analysis results, first read the fault signal envelope analysis data formed previously one by one according to the time index, and check the amplitude information and phase indicators corresponding to each time point while reading, and compare these information with the previously recorded current and voltage reference values ​​one by one. For example, compare the current reference value with the 0A to 5A interval, and compare the voltage reference value with the 0V to 24V interval to determine whether the corresponding time point belongs to the normal range or the deviation range. When multiple deviations or continuous fluctuations occur, mark the interval to enter the differential operation sequence. Then, when performing differential processing on these marked intervals, select the difference in the amplitude values ​​of adjacent sampling points in the same time window and record the difference result. If the difference is multiple times in the same window, When the specified value is exceeded, it is necessary to query the origin of the specified value. For example, the amplitude differential distribution data collected by the device after 300 hours of continuous operation is subjected to quantile statistics and a certain quantile is set as the threshold. The differential values ​​exceeding the threshold are marked and added to the amplification list, and then the marked differential values ​​are amplified one by one. The amplified fault signal is obtained by multiplying the differential value with the amplification factor. The selection of the amplification factor needs to be combined with the differential amplitude distribution range obtained previously. A suitable gain is determined through segmented cumulative calculation, so that after amplification, it still maintains appropriate recognizability in medium-amplitude scenarios, and does not saturate in large-amplitude scenarios. After the differential amplification is completed, the differential amplification result set is updated and the final amplified fault signal is saved.

[0030] Based on the amplified fault signal and combined with the previously determined amplification coefficient distribution, a high-pass filtering operation is performed on each amplified data segment. First, the time domain signal is segmented and classified before filtering, and a specific segment center frequency is given. 0Hz to 50Hz is defined as a low frequency band, 50Hz to 500Hz is defined as a medium frequency band, 500Hz and above is defined as a high frequency band, etc. Then, during filtering, the high frequency band components are retained and the energy components below the segment center are filtered out. During the filtering process, a cutoff frequency needs to be determined based on the long-term accumulated motor operation vibration frequency statistics and the cutoff frequency is used as the main basis for the filtering parameters. For example, based on The vibration data samples extracted earlier show that most normal vibrations of the equipment are concentrated below 200Hz, so the cutoff frequency of the high-pass filter is set between 200Hz and 300Hz. The specific threshold is fine-tuned by judging whether it frequently crosses this frequency point during continuous operation. After the low-frequency components in the signal are attenuated, observe whether the amplitude of the high-frequency part is maintained in the normal range. For example, compare the high-frequency amplitude with the high-frequency amplitude distribution range recorded in advance. If there is a significant increase multiple times and the amplitude exceeds the reference range, it will be marked. Finally, the result retained and output by the high-pass filter will be attributed to the differential amplification fault signal and recorded.

[0031] Based on the differential amplification fault signal, when performing feature extraction, the differential amplification fault signal obtained above is first matched with the time axis, and its amplitude peak points and waveform change trends in several sampling windows are read one by one. At the same time, the current amplitude or voltage amplitude corresponding to this period of time is recorded and compared with the current 0A to 5A range and the voltage 0V to 24V range of the normal operation of the equipment for multiple times. If the peak values ​​detected in multiple places exceed the reference peak threshold calculated by the previous sample distribution, these peak points are used as candidates for fault features, and further compared with the peak change amplitude and occurrence frequency in adjacent sampling windows. If a relatively concentrated and large-amplitude peak appears multiple times in the same window, it is marked in the fault feature list. At the same time, combined with the extreme value position of the waveform in the rising and falling segments, it is detected whether it meets the fault features of waveform mutation or jitter type. Each candidate peak and its surrounding waveform pattern features are summarized and the repeated distribution of peaks is evaluated in the cross-window scenario. If a waveform pattern of similar intensity appears in multiple time windows, it is further confirmed that the pattern belongs to a stable fault feature. Finally, all peaks and waveform patterns that meet the fault feature identification conditions are recorded to generate differential amplification fault features.

[0032] The steps to obtain the fault identification results are as follows: Collect the differential amplification fault features, perform standardization processing, and obtain the pre-processed feature vector; Based on the preprocessed feature vector, a support vector machine classifier is configured and trained. The support vector machine classifier uses a radial basis function as a kernel function to obtain a configured and trained SVM classifier. Based on the configured and trained SVM classifier, the test data is fault classified to identify and classify the fault type and form a fault identification result.

[0033] Specifically, to collect the differential amplifier fault characteristics, first read the time series data and corresponding amplitude information of each differential amplifier fault characteristic obtained earlier one by one, aggregate these amplitude information according to the same time index and summarize them in the form of a list, and then check the numerical range of each record in the summary list, for example, compare the current amplitude with the range of 0A to 5A, and compare the voltage amplitude with the range of 0V to 24V, so as to identify whether there is data out of range. If it is found that it is out of range, mark it in the summary list and record the amplitude status of the sample point separately. Next, calculate the statistics of the entire differential amplifier fault characteristic set, including the arithmetic mean, maximum value, minimum value, variance, etc., use these statistics to evaluate the gap and distribution concentration between the fault characteristics, and then based on experience or history, A distribution threshold interval is set based on the historical analysis results, and samples that are obviously at extreme value positions are marked as candidate anomalies for further review. After the initial marking is completed, the standardization process is started. By subtracting the corresponding mean value from each feature data and then dividing it by the corresponding standard deviation, each sample is mapped to a standard normal distribution interval with a mean of 0 and a standard deviation of 1. For example, when processing the current amplitude, the mean value is first subtracted and then divided by the standard deviation. The same method is used for the voltage amplitude and vibration amplitude. After the process is completed, a set of numerical values ​​with a more compact range and not affected by the original dimension is formed. At the same time, the time index of the record is retained in each record to maintain its reference in subsequent stages. Finally, all differential amplification fault features that have undergone the standardization process are merged and arranged in a fixed order to obtain the preprocessed feature vector.

[0034] Based on the preprocessed feature vectors, each feature vector record is first assigned its corresponding label information in the training set. If the real labels of the equipment in different fault scenarios have been collected before, the corresponding records are retrieved from the label set and attached to the end of the training data table. Then, before training, each feature vector is compared one by one to check whether the numerical distribution of fields such as current amplitude, voltage amplitude, and vibration parameters is consistent with the common feature intervals in the label. If there is an inconsistency and the value exceeds the set detection range, the record is marked in the training data table and its belonging label is checked to see if it needs to be corrected. After all samples have been verified, the corresponding records are compared according to the facts. First, the collected differential amplification fault feature data and corresponding label information are split into a training set and a validation set. Then, the support vector machine classifier with radial basis function as the kernel function is called and the feature vectors and labels in the training set are gradually input. The batch gradient method is used for iterative training. The classification decision boundary is updated multiple times during the iteration process and the classification accuracy is evaluated with the validation set at the end of each iteration. When the evaluation result is stable, the training is stopped and the final classification accuracy is recorded to confirm the classification effect of the model in fault type identification. Finally, the support vector machine classifiers with stable classification performance are packaged uniformly and the configuration information is recorded and regarded as SVM classifiers that have been configured and trained.

[0035] Based on the configured and trained SVM classifier, the feature vectors of each record in the test data set are first subjected to the same normalization preprocessing as the previously generated standardized parameters, including the same mean and standard deviation adjustment of differential feature items such as current amplitude, voltage amplitude and vibration. These processed vector data are then sent to the trained SVM classifier, and the previously trained radial basis function is used to check the test data for identification one by one. The classification label results generated for each identification are summarized and compared with the true labels of the test samples recorded in advance. If the classification label is inconsistent with the true label, it is marked in the test data set for subsequent review. When all test samples have been identified, statistics are taken based on the comparison results to check the accuracy and confusion of the classifier on various fault types. If the accuracy exceeds the preset threshold, the classifier is included in the subsequent available list and applied to the real-time monitoring of the equipment. Otherwise, it is necessary to trace back the distribution of the training set and the feature processing method to find the cause. Finally, the fault classification results of each test sample are classified to display different fault types and their distribution on the time axis to form a fault identification result.

[0036] The steps for obtaining dynamic fault threshold parameters are as follows: Collect information about the motor's current load level, operating speed, temperature, and motor type to form an input data set for fault diagnosis; Based on the input data set of fault diagnosis, a new fault diagnosis threshold is calculated. The calculation formula is: in, is the new fault diagnosis threshold, is the initial threshold, , , represent load level, speed and temperature respectively, , , is the adjustment factor, is the normalization factor; According to the new fault diagnosis threshold, the fault diagnosis threshold setting is updated to form a dynamic fault threshold parameter.

[0037] Specifically, the current load level, operating speed, temperature and type information of the motor are collected. During the execution process, the indicators related to the motor load are read separately and compared with the established load range list, such as comparing the load value with the range of 0% to 120%. If the load data exceeds 120%, the sample point is marked in the record and archived for subsequent analysis. The motor operating speed is recorded and compared with the range of 0rpm to 10000rpm, and the temperature is compared with the range of 0℃ to 90℃. It is also verified whether the type information of the motor matches the pre-confirmed type list in the database. If a type code is found not in the list, it is listed as a type to be confirmed, and then the above-mentioned types that meet the range or are to be confirmed are listed. The recognized data items are integrated into the input data set of fault diagnosis in turn. In order to ensure the correspondence between information such as speed, load and temperature in different time periods, it is necessary to combine and record the data at each data sampling moment to form a complete record. After the summary is completed, all records are checked line by line for the relationship between their load, operating speed and temperature. For example, the average load and operating speed corresponding to the temperature not exceeding 40°C are counted, and the load and speed are compared when the temperature is higher than 40°C. If the load continues to increase or the operating speed changes continuously and too quickly under high temperature conditions, the time period is marked so that correlation analysis can be performed in combination with the diagnosis threshold calculation in the subsequent steps, and finally an input data set for fault diagnosis is formed.

[0038] The formula is beneficial in that it combines the three elements of load level, speed and temperature and introduces multiple adjustment coefficients to dynamically control the threshold, making the threshold for fault judgment more flexible and in line with actual working conditions; The acquisition step is to set the initial threshold when the device leaves the factory or is first deployed. This value is obtained through industrial verification data and is allowed to be slightly fine-tuned in the later stage. The steps to obtain the value are as follows: the actual load monitoring device records the average load level of the motor over a period of time, and compares this value with the rated load to obtain a proportional value. For example, if the measured load is 80% of the rated load, then ; The acquisition steps are as follows: the motor speed is collected by the speed sensor, and compared with the maximum speed of 10000rpm, and the result is converted into a quantitative value between 0 and 1. For example, when the running speed is 5800rpm in this stage, ; The acquisition steps are as follows: the temperature sensor measures the motor temperature. If the current temperature is 35°C, , take the actual temperature number and assign it directly; , , The steps for obtaining the adjustment coefficient are to sort out multiple rounds of industrial test data, calculate the sensitivity of the failure rate per unit time to the above three items of load, speed and temperature, and combine the weighted average processing to obtain the corresponding adjustment coefficient; The acquisition step is to evaluate and sort out the threshold correction range in multiple operation results. For example, in the continuous operation of the equipment for 300 hours, the fluctuation range of the threshold before and after each failure is recorded, and the correction coefficient is calculated; The steps of obtaining are to obtain factors with the same dimension as each element by normalizing and summarizing each parameter in the range of 0 to 1 or 0 to the rated value, and use them for the denominator to make overall balance; Calculation process: Given an example, let , ; Load Level , speed quantization ,temperature ; Through the fault sensitivity analysis previously organized, we can obtain , , And record it as reference value; From the existing normalization standard ; calculate ,calculate , and then calculate ; Add the numerators, , , ,sum ; Divide by ,result , then multiply by ,get ; Finally, Add together and get ; The results show that under the current load, speed and temperature conditions, the new fault diagnosis threshold is calculated to be 1.908. If the fault index obtained in subsequent monitoring exceeds 1.908, it can be determined to be abnormal.

[0039] According to the new fault diagnosis threshold, when updating the fault diagnosis threshold setting, it is necessary to re-check all registered motor operation data. During the execution process, the sampled data is read one by one and compared with the threshold of 1.908. If the corresponding fault indicator value exceeds 1.908, the data is marked as suspected abnormal in the summary record. At the same time, it is cross-checked with the fault threshold of the earlier stage to determine whether fault signs have appeared in a similar range before. If it is found that there is a difference between this mark and the previous result, the difference record is split and its abnormal mark is retained under the current threshold scheme. If the fault indicator of the same type of equipment or the same type of information exceeds the threshold for many times, the number of times the limit is exceeded is counted and the current fault symptom conclusion is reported. In order to ensure the continuity of the next step of diagnosis, it is also necessary to record the speed, load, temperature and other information of the motor when the limit is exceeded for further analysis. Finally, all the latest judgments are saved in the updated fault diagnosis threshold parameters.

Claims

1. A mechanical and electrical equipment fault diagnosis system based on machine learning, characterized in that: The system comprises: A high-frequency signal injection module injects a high-frequency signal in the range of 10kHz to 50kHz into the motor through the power supply line, analyzes the electromagnetic response of the signal, and obtains electromagnetic parameter monitoring data; based on the electromagnetic parameter monitoring data, extracts statistical features and performs pattern recognition to generate early fault characteristic signals; The dynamic feature analysis module transforms the early fault feature signal, extracts the envelope analysis data of the signal, and establishes the fault signal envelope analysis result; processes the fault signal envelope analysis result by differential signal amplification to obtain the differential amplification fault feature; A fault diagnosis module, based on the differential amplification fault feature, constructs a feature vector and inputs it into a support vector machine classifier to calculate and obtain a fault identification result; The fault threshold dynamic adjustment module adjusts the fault diagnosis threshold in real time according to the motor type and operating conditions, and establishes dynamic fault threshold parameters.

2. The electromechanical equipment fault diagnosis system based on machine learning according to claim 1 is characterized in that: The steps for acquiring the electromagnetic parameter monitoring data are as follows: A high-frequency signal in the range of 10kHz to 50kHz is injected into the motor through the power line, and the collected electromagnetic waveform data is used as the basis for analysis to form initial electromagnetic response data; The initial electromagnetic response data is subjected to spectrum analysis to calculate the electromagnetic parameter intensity. The calculation formula is: in, Represents the frequency The electromagnetic parameter intensity under Representative time point at frequency The signal strength, For the The acquisition time point of the signal, is the total number of signal acquisitions; Based on the electromagnetic parameter intensity, the electromagnetic response characteristics of the signal are analyzed to obtain electromagnetic parameter monitoring data.

3. The electromechanical equipment fault diagnosis system based on machine learning according to claim 1 is characterized in that: The steps of obtaining the early fault characteristic signal are: Based on the electromagnetic parameter monitoring data, outliers are removed and normalized to obtain preprocessed electromagnetic data; Based on the pre-processed electromagnetic data, the statistical measure of the fault characteristics is extracted, and the calculation formula is: in, A statistical measure that characterizes the fault, After processing data points, for The average value of for The standard deviation of is the total number of samples; Based on the statistical measurement, a decision tree is applied to perform pattern recognition, identify fault characteristic signals, and obtain early fault characteristic signals.

4. The electromechanical equipment fault diagnosis system based on machine learning according to claim 1, characterized in that: The steps for obtaining the fault signal envelope analysis result are: Based on the early fault characteristic signal, a Hilbert transform is performed to map the real signal at each time point to a complex plane to obtain an analytical signal; Based on the analytical signal, the envelope of the signal is calculated using the following formula: in, represents the envelope of the signal, is the signal strength at time j, is the phase angle at time j, is the total number of sample points; Based on the signal envelope, continuous wavelet transform is applied to perform multi-scale analysis to identify abnormal changes in the signal and form the fault signal envelope analysis results.

5. The electromechanical equipment fault diagnosis system based on machine learning according to claim 1, characterized in that: The steps of obtaining the differential amplifier fault characteristics are as follows: Based on the fault signal envelope analysis result, differential signal amplification is performed to obtain an amplified fault signal; Based on the amplified fault signal, a high-pass filter is applied to generate a differential amplified fault signal by filtering out low-frequency components in the signal and retaining high-frequency components; Based on the differential amplification fault signal, a feature extraction process is performed to identify fault features by detecting and analyzing peak values ​​and waveform patterns in the signal to form differential amplification fault features.

6. The electromechanical equipment fault diagnosis system based on machine learning according to claim 1, characterized in that: The steps of obtaining the fault identification result are: Collecting the differential amplification fault features, performing standardization processing, and obtaining a preprocessed feature vector; Based on the preprocessed feature vector, a support vector machine classifier is configured and trained, wherein the support vector machine classifier uses a radial basis function as a kernel function to obtain a configured and trained SVM classifier; Based on the configuration and the trained SVM classifier, the test data is fault classified, the fault type is identified and classified, and a fault identification result is formed.

7. The electromechanical equipment fault diagnosis system based on machine learning according to claim 1, characterized in that: The steps for obtaining the dynamic fault threshold parameter are: Collect information about the motor's current load level, operating speed, temperature, and motor type to form an input data set for fault diagnosis; Based on the input data set of the fault diagnosis, a new fault diagnosis threshold is calculated, and the calculation formula is: in, is the new fault diagnosis threshold, is the initial threshold, , , represent load level, speed and temperature respectively, , , is the adjustment factor, is the normalization factor; According to the new fault diagnosis threshold, the fault diagnosis threshold setting is updated to form a dynamic fault threshold parameter.

8. A method for diagnosing electromechanical equipment faults based on machine learning, characterized in that: The electromechanical equipment fault diagnosis system based on machine learning according to any one of claims 1 to 7 comprises the following steps: Based on the electromagnetic parameter monitoring data, the statistical characteristics of the electromagnetic parameters are extracted, the fault mode is identified using the statistical characteristics, the abnormal signal mode is obtained through pattern recognition, and the early fault characteristic signal is generated; Perform frequency domain transformation on the early fault characteristic signal, extract the signal envelope data, perform signal envelope analysis, generate signal envelope analysis results, use differential amplification to process the signal envelope analysis results, and obtain differential amplification fault characteristics; The differential amplifier fault feature is used as input data to construct a feature vector, which is input into the support vector machine classifier. The support vector machine classifies the input data, calculates and obtains the fault probability and type, and outputs the fault identification result. According to the fault identification result, referring to the type and operating conditions of the motor, the fault diagnosis threshold is dynamically adjusted, the fault threshold is updated in real time, the fault judgment parameters are reset, and the dynamic fault threshold parameters are established and updated.

Images