Hydropower station auxiliary machine equipment real-time monitoring method and system based on Internet of Things

Abstract

The invention provides a hydropower station auxiliary machine equipment real-time monitoring method and system based on the Internet of Things, and relates to the technical field of information, and the method comprises the steps: constructing a pressure oil pump fault diagnosis model for a temperature and vibration related feature subset, and training a temperature and vibration feature sample of a preset fault type, establishing a mapping relation between the temperature and vibration characteristics of the pressure oil pump and the fault type; on the basis of the pressure oil pump fault diagnosis model, typical faults of bearing abrasion and poor gear meshing of the pressure oil pump are classified and recognized, and the temperature and vibration fault diagnosis result of the pressure oil pump is determined through the mapping relation between the temperature and vibration characteristics of the pressure oil pump and the fault type; and according to the fault diagnosis results of the temperature and vibration of the pressure oil pump, carrying out comprehensive decision making on a plurality of diagnosis results, according to the preset confidence of different diagnosis results, giving corresponding weight coefficients, and generating a final health state evaluation result of the pressure oil pump. The operation reliability and the maintenance efficiency of the pressure oil pump are improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a real-time monitoring method and system for auxiliary equipment of a hydropower station based on the Internet of Things. Background Art

[0002] The medium-pressure oil pump of the auxiliary equipment of the hydropower station is one of the key equipment, and its operating status directly affects the safe and stable operation of the hydropower station. In the actual operation process, the oil pump often has problems such as bearing wear and poor gear meshing, which leads to increased vibration and temperature rise of the equipment. In severe cases, it may even cause equipment failure, affecting the normal production of the hydropower station. Traditional oil pump monitoring mainly relies on temperature sensors to judge the operating status of the equipment by monitoring temperature changes, but it is difficult to fully reflect the health status of the equipment, especially the capture of early signs of failure, based on temperature information alone. In order to achieve comprehensive monitoring and health management of the oil pump, it is urgent to introduce new monitoring methods. Vibration is another important representation of the operating status of the oil pump. The vibration signal during the operation of the oil pump can be obtained through the vibration sensor, and then the dynamic characteristics of the equipment can be analyzed. However, in actual applications, the vibration signal is easily affected by external interference, the data quality is difficult to guarantee, and the vibration characteristics under different fault modes are not obvious. It is difficult to accurately diagnose the fault type with a single vibration signal. Therefore, how to effectively integrate temperature and vibration signals under complex working conditions, explore the intrinsic relationship between the two, and construct a temperature-vibration joint monitoring model is a key technical problem that needs to be solved to achieve comprehensive monitoring and health management of the oil pump. At the same time, based on massive monitoring data, how to establish an efficient data analysis algorithm to accurately identify the abnormal working status of the oil pressure pump is also an issue worthy of in-depth study. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides a real-time monitoring method for auxiliary equipment of a hydropower station based on the Internet of Things, which mainly includes:

[0004] Collect the temperature data and vibration data of the oil pump, pre-process the collected temperature data and vibration data, remove the abnormal values ​​and noise interference, obtain the time series data of the oil pump temperature and vibration, and perform feature extraction. Through time domain analysis and frequency domain analysis, extract the statistical characteristics and spectrum characteristics of the oil pump temperature and vibration signals, and obtain the characteristic vector of the oil pump temperature and vibration characteristics;

[0005] According to the historical working data of the oil pump, the correlation analysis is performed on the characteristic vectors of the temperature and vibration characteristics of the oil pump, and the correlation coefficient matrix between different characteristics is calculated. By screening out the features in the matrix whose correlation between temperature and vibration characteristics is higher than the preset threshold, the temperature and vibration related feature subset is obtained;

[0006] A fault diagnosis model for the oil pump is constructed for the subset of temperature and vibration related features. By training the temperature and vibration feature samples of the preset fault types, a mapping relationship between the temperature and vibration features of the oil pump and the fault type is established.

[0007] Based on the fault diagnosis model of the oil pump, the typical faults of the oil pump such as bearing wear and gear meshing failure are classified and identified. The fault diagnosis results of the oil pump temperature and vibration are determined through the mapping relationship between the oil pump temperature and vibration characteristics and the fault type.

[0008] According to the fault diagnosis results of the oil pump temperature and vibration, a comprehensive decision is made on multiple diagnosis results, and corresponding weight coefficients are assigned according to the preset confidence of different diagnosis results to generate the final health status assessment result of the oil pump;

[0009] Based on the health status assessment results of the oil pressure pump, combined with the pre-set health thresholds and alarm rules, it is determined whether the oil pressure pump is in an abnormal working state. If the health index of the oil pressure pump is lower than the preset threshold or the corresponding alarm rules are triggered, a fault warning signal is issued to prompt maintenance or inspection of the oil pressure pump.

[0010] The present invention provides a real-time monitoring system for auxiliary equipment of a hydropower station based on the Internet of Things, which mainly includes:

[0011] The data acquisition and preprocessing module is used to collect the temperature data and vibration data of the oil pressure pump, preprocess the collected temperature data and vibration data, remove the abnormal values ​​and noise interference, obtain the time series data of the oil pressure pump temperature and vibration, and perform feature extraction. Through time domain analysis and frequency domain analysis, the statistical characteristics and spectrum characteristics of the oil pressure pump temperature and vibration signals are extracted to obtain the characteristic vector of the oil pressure pump temperature and vibration characteristics;

[0012] A correlation analysis module is used to perform correlation analysis on the characteristic vectors of the temperature and vibration characteristics of the oil pressure pump according to the historical working data of the oil pressure pump, calculate the correlation coefficient matrix between different characteristics, and obtain the temperature and vibration related feature subset by screening out the features in the matrix whose correlation between the temperature and vibration characteristics is higher than the preset threshold;

[0013] The fault diagnosis modeling module is used to build a fault diagnosis model for the oil pump based on the temperature and vibration related feature subsets. By training the temperature and vibration feature samples of the preset fault types, the mapping relationship between the temperature and vibration features of the oil pump and the fault type is established.

[0014] The typical fault classification and identification module is used to classify and identify the typical faults of bearing wear and gear meshing failure of the oil pump based on the oil pump fault diagnosis model, and determine the fault diagnosis results of the oil pump temperature and vibration through the mapping relationship between the oil pump temperature and vibration characteristics and the fault type;

[0015] The comprehensive decision-making module is used to make a comprehensive decision on multiple diagnosis results based on the fault diagnosis results of the oil pump temperature and vibration, assign corresponding weight coefficients according to the preset confidence of different diagnosis results, and generate the final health status assessment result of the oil pump;

[0016] The health status assessment and early warning module is used to determine whether the oil pressure pump is in an abnormal working state based on the health status assessment results of the oil pressure pump, combined with the pre-set health thresholds and alarm rules. If the health index of the oil pressure pump is lower than the preset threshold or the corresponding alarm rules are triggered, a fault warning signal is issued to prompt maintenance or inspection of the oil pressure pump.

[0017] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0018] The present invention discloses a real-time monitoring method and system for auxiliary equipment of a hydropower station based on the Internet of Things. The method collects temperature and vibration data of a pressure oil pump, performs preprocessing and feature extraction on the data, and obtains a feature vector. The relevant feature subsets of temperature and vibration are screened out through correlation analysis, and a fault diagnosis model is constructed. The model divides the fault types into bearing, pump body, seal and motor faults, and classifies and identifies typical faults such as bearing wear and poor gear meshing. According to the diagnosis results, the present invention comprehensively considers the historical occurrence frequency, severity and impact of different fault types, assigns weight coefficients, and generates the final health status assessment results. If the health index is lower than a preset threshold or the alarm rule is triggered, a fault warning signal is issued. By integrating temperature and vibration characteristics, the present invention realizes accurate diagnosis and timely warning of pressure oil pump faults, effectively improves the operating reliability and maintenance efficiency of the pressure oil pump, reduces the risk of equipment failure, and provides an important basis for preventive maintenance of the pressure oil pump. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flow chart of a method for real-time monitoring of auxiliary equipment of a hydropower station based on the Internet of Things. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0021] like Figure 1The real-time monitoring method for auxiliary equipment of a hydropower station based on the Internet of Things in this embodiment may specifically include:

[0022] S101. Collect temperature data and vibration data of the oil pressure pump, pre-process the collected temperature data and vibration data, remove abnormal values ​​and noise interference, obtain time series data of the oil pressure pump temperature and vibration, and perform feature extraction. Through time domain analysis and frequency domain analysis, extract the statistical characteristics and spectral characteristics of the oil pressure pump temperature and vibration signals, and obtain the characteristic vector of the oil pressure pump temperature and vibration characteristics.

[0023] According to the Nyquist sampling theorem, the temperature data and vibration data of the oil pressure pump are collected and marked with timestamps. The collection frequency of the temperature data is to collect a first preset number of data points per second, and the collection frequency of the vibration data is to collect a second preset number of data points per second; the temperature data is filtered by a Butterworth low-pass filter, and the outliers of the filtered temperature data are eliminated according to the principle of three times the standard deviation to obtain the processed temperature data; the vibration data is subjected to four-layer wavelet decomposition using the db four-wavelet basis function, and the processed vibration data is obtained by removing the high-frequency noise components; the processed temperature data is segmented and calculated according to the preset time window to obtain the time domain eigenvalues, and the processed vibration data is subjected to fast Fourier transform to obtain the frequency domain eigenvalues; the time domain eigenvalues ​​and the frequency domain eigenvalues ​​are normalized to the maximum and minimum values ​​to obtain the normalized eigenvalues, and the feature combination with the correlation coefficient greater than the preset threshold is selected by calculating the Pearson correlation coefficient between the normalized eigenvalues, and the feature combination is subjected to dimensionality reduction processing using the principal component transformation to obtain the temperature vibration eigenvector of the oil pressure pump.

[0024] Specifically, according to the Nyquist sampling theorem, the oil pump temperature data acquisition frequency is calculated to be 1024 data points per second, and the vibration data acquisition frequency is 2048 data points per second. The temperature data and vibration data are time-stamped and stored in a numerical array. The temperature data marked with the timestamp is filtered according to the Butterworth low-pass filter, and the temperature data abnormal values ​​are judged and eliminated based on the principle of three times the standard deviation within the 10-second time window. The db4 wavelet basis function is used to perform a 4-layer wavelet decomposition on the vibration data to remove high-frequency noise components. The filtered temperature data is segmented according to a fixed time window of 60 seconds, and the four time domain eigenvalues ​​of the temperature data, namely the mean, variance, kurtosis, and skewness, are calculated. The denoised vibration data is subjected to a 512-point fast Fourier transform to obtain a spectrum diagram, and the frequencies and amplitudes corresponding to the first 10 maximum amplitude points in the spectrum diagram are extracted. The extracted temperature time domain eigenvalues ​​and vibration frequency domain eigenvalues ​​are normalized to their maximum and minimum values. The correlation between eigenvalues ​​is calculated based on the Pearson correlation coefficient, and the feature combination with a correlation coefficient greater than 0.7 is selected. The normalized eigenvalues ​​are reduced in dimension using principal component transformation, and the oil pump temperature vibration eigenvector is constructed. During the oil pump data collection process, the temperature data collection frequency is 1024 Hz for sampling and determination of the temperature spectrum, and the temperature sensor sensitivity is set to 0.01 degrees Celsius. The vibration data collection frequency is 2048 Hz for sampling and determination of the vibration spectrum, and the vibration sensor sensitivity is set to 0.001 mm / s2. The data storage format is a floating point array, and each data point contains three attribute values: temperature value, vibration value, and collection time stamp. In the data filtering preprocessing stage, the temperature data is filtered out with a 5th-order Butterworth low-pass filter to remove frequency components above 50 Hz. After filtering, the temperature data is calculated with a standard deviation of 2.5 degrees Celsius based on a mean of 36 degrees Celsius within a 10-second window. Data points beyond the range of three times the standard deviation are determined as outliers. The vibration data is decomposed using the four-layer decomposition of the db4 wavelet basis function to remove noise interference with a frequency higher than 500 Hz. In the data feature extraction stage, the temperature data is segmented according to a fixed window of 60 seconds, and the four statistical values ​​of temperature mean, variance, kurtosis, and skewness are calculated. The vibration data is subjected to a 512-point fast Fourier transform to obtain a spectrum diagram, and the top 10 frequency points with the largest amplitude are extracted. The frequency range is 0-1000 Hz, and the corresponding amplitude range is 0-10 mm / s2. In the feature dimensionality reduction processing stage, the maximum and minimum values ​​of the 14-dimensional temperature vibration eigenvalues ​​are normalized and mapped to the 0-1 interval. Eight main features are obtained based on the 0.7 correlation coefficient threshold. The 8-dimensional features are reduced to a 4-dimensional feature vector through principal component transformation. The feature vector contains four principal components: temperature mean, temperature variance, vibration main frequency, and vibration amplitude. The cumulative contribution rate reaches 95%.

[0025] S102. Based on the historical working data of the oil pressure pump, a correlation analysis is performed on the characteristic vectors of the temperature and vibration characteristics of the oil pressure pump, and a correlation coefficient matrix between different characteristics is calculated. The characteristics whose correlation between temperature and vibration characteristics in the matrix is ​​higher than a preset threshold are screened out to obtain a subset of temperature and vibration related characteristics.

[0026] The temperature feature vector and vibration feature vector in the history database of the oil pressure pump are obtained, the feature vectors are aligned according to the millisecond timestamp, and a standardized feature matrix is ​​obtained through a sliding time window; for the standardized feature matrix, the four time domain features of mean, variance, kurtosis and skewness in the temperature matrix and the four frequency domain features of main frequency, frequency bandwidth, amplitude and waveform factor in the vibration matrix are extracted, and the standardized eigenvalues ​​are obtained by Z-score standardization; according to the standardized eigenvalues, the cross-correlation coefficient matrix between the temperature feature and the vibration feature is calculated, and the lagged correlation between the features is obtained within the time delay range; if the absolute value of the correlation coefficient is greater than a preset first threshold or the absolute value of the lagged correlation coefficient is greater than a preset second threshold, the corresponding feature combination is added to the candidate feature subset; for the feature combination in the candidate feature subset, the nonlinear correlation is verified by mutual information entropy, and the temperature-vibration related feature subset is obtained by arranging in descending order according to the information gain ratio.

[0027] Specifically, the temperature feature vectors and vibration feature vectors with timestamps for the past 30 days are read from the historical database of the oil pump, the data are aligned according to the millisecond timestamp, the feature vectors are segmented using a 1-hour sliding time window, the abnormal data with amplitudes exceeding the range of plus or minus three times the standard deviation are eliminated, and the standardized feature matrix is ​​obtained by maximum and minimum value normalization. For the four time domain features of mean, variance, kurtosis, and skewness in the standardized temperature feature matrix and the four frequency domain features of main frequency, frequency bandwidth, amplitude, and waveform factor in the vibration feature matrix, the 4x4 dimensional cross-correlation coefficient matrix between the temperature feature and the vibration feature is calculated based on Z-score standardization to eliminate the dimensional effect, and the lag correlation between the features is calculated by sliding within the time delay range of 0 to 4 hours. The 16 cross terms of the temperature feature and the vibration feature in the calculated correlation coefficient matrix are traversed. If the absolute value of the correlation coefficient is greater than 0.85 or the absolute value of the lag correlation coefficient is greater than 0.8, the corresponding feature combination is added to the candidate feature subset, and the repeated feature combinations with a correlation coefficient difference less than 0.1 are eliminated from the candidate feature subset. The nonlinear correlation of feature combinations in the candidate feature subsets was verified based on mutual information entropy. The feature combinations with normalized mutual information entropy values ​​greater than 0.7 were arranged in descending order according to the information gain ratio. The top 8 feature combinations were selected to construct the temperature vibration related feature subset, and the feature correlation calculation results were updated every hour. During the historical data collection and reading stage of the oil pressure pump, a temperature and vibration data record is read from the database every millisecond in the past 30 days. The temperature data accuracy is 0.01 degrees Celsius, the vibration data accuracy is 0.001 millimeters per square second, the timestamp format uses a 13-bit Unix timestamp, and the data storage structure is in the form of key-value pairs. The average temperature data is 36 degrees Celsius, the standard deviation is 2.5 degrees Celsius, the root mean square value of the vibration data is 0.5 millimeters per square second, and the peak-to-peak value is 2 millimeters per square second. The data is divided into continuous time series segments according to a fixed window of 3600 seconds. The data that exceeds the temperature range of 35-38 degrees Celsius and the vibration amplitude range of 2 millimeters per square second is marked as abnormal and eliminated. The temperature data is normalized to the 0-1 range through maximum and minimum value mapping, and the vibration data is normalized to the negative 1 to 1 range. In the feature correlation calculation stage, the standardized temperature feature matrix and vibration feature matrix were transformed by Z scores. The temperature features included four statistical values: mean 36 degrees Celsius, variance 6.25, kurtosis 2.8, and skewness 0.15. The vibration features included four eigenvalues: main frequency 35 Hz, frequency bandwidth 5 Hz, amplitude 0.5 mm / s2, and waveform factor 1.5. A 4 by 4 dimensional correlation coefficient matrix was constructed. At the same time, the lagged correlation coefficient was calculated in a 10-minute step within the time window of 0 to 4 hours, and 24 sets of lagged correlation values ​​were obtained.In the relevant feature screening stage, the correlation coefficient between the temperature mean and the vibration amplitude is 0.87, the correlation coefficient between the temperature variance and the main vibration frequency after 2 hours is 0.83, and the correlation coefficient between the temperature kurtosis and the frequency bandwidth is 0.86. A total of 8 groups of feature combinations with correlation coefficients greater than 0.8 were screened out. By calculating the difference in correlation coefficients between feature combinations, 4 groups of repeated features were eliminated, and four independent feature combinations of temperature mean-vibration amplitude, temperature variance-main vibration frequency, temperature kurtosis-frequency bandwidth, and temperature skewness-waveform factor were obtained. In the nonlinear correlation verification stage, the mutual information entropy of the four feature combinations was calculated. The normalized mutual information entropy of the temperature mean-vibration amplitude combination was 0.85, the temperature variance-main vibration frequency combination was 0.82, the temperature kurtosis-frequency bandwidth combination was 0.75, and the temperature skewness-waveform factor combination was 0.72. The feature combinations were sorted according to the information gain ratio, and the feature correlation results were updated and calculated every 1 hour. During the update process, the number of feature combinations was fixed to 4 groups.

[0028] S103, constructing a fault diagnosis model for the oil pressure pump for a subset of temperature and vibration related features, and establishing a mapping relationship between the temperature and vibration features of the oil pressure pump and the fault type by training temperature and vibration feature samples of preset fault types.

[0029] The temperature mean, vibration amplitude, temperature variance and main vibration frequency feature combination in the oil pressure pump label data are obtained, and a training data set is constructed according to the normal operating status, bearing wear status, seal leakage status and impeller wear status ratio markers; for the training data set, the maximum and minimum method is used to normalize the feature data, and the feature vector is converted to a three-dimensional feature space through principal component transformation to obtain feature components whose cumulative contribution rate meets the requirements; according to the four operating status sample points in the three-dimensional feature space, a classification tree is constructed using a random forest algorithm, and the classifier parameters are optimized through grid search to obtain the optimal number of decision trees and learning rate parameters; for the classifier, a test data set is used for verification, and if the proportion of the test data set to the total samples is the set value, the classifier performance evaluation result is obtained; the Mahalanobis distance between the new feature vector and the centroid of each category is calculated by the classifier, and if the Mahalanobis distance confidence is higher than the threshold, it is determined that the new feature vector corresponds to the fault type label, and if the Mahalanobis distance confidence is lower than the threshold, it is determined to be in a pending state.

[0030] Specifically, read the four feature combinations of temperature mean, vibration amplitude, temperature variance, and vibration main frequency from the oil pressure pump label data, and mark 10,000 feature samples according to the normal operation state accounting for 40%, bearing wear state accounting for 25%, seal leakage state accounting for 20%, and impeller wear state accounting for 15%. Build a training data set and remove abnormal samples that exceed the range of plus or minus three times the standard deviation of the temperature mean and the range of plus or minus four times the root mean square of the vibration amplitude. The four sets of feature data in the marked training data set are normalized to the interval of zero to one according to the maximum and minimum value method, and the Pearson correlation coefficient between the feature combinations is calculated. Based on the principal component transformation, the four-dimensional feature vector is converted to the three-dimensional feature space and the feature components with a cumulative contribution rate of more than 95% are retained. The distribution map of the four types of operating state sample points is drawn in the three-dimensional feature space and the Mahalanobis distance between classes is calculated. Based on random forest, a classification tree is constructed for the four operating status sample points in the three-dimensional feature space. Five-fold cross validation is used to determine the optimal number of decision trees, 500, with a lower limit of 20 leaf node samples, a maximum tree depth of 8 layers, and a feature sampling ratio of 0.8. The learning rate parameter is optimized to 0.05 between 0.01 and 1 through grid search. The trained classifier is verified using a test data set that accounts for 20% of the total samples. The newly collected temperature and vibration feature vectors are input, and the Mahalanobis distance between the feature points and the centroids of the four categories is calculated in the three-dimensional feature space. The fault type label corresponding to the closest category is determined based on the confidence threshold of 0.85, and the prediction results with confidence levels below the threshold are marked as pending. During the processing of the oil pump fault diagnosis data, four sets of characteristic data were read, including the temperature mean reference value of 36 degrees Celsius, the vibration amplitude reference value of 0.5 mm per square second, the temperature variance reference value of 4 square degrees Celsius, and the vibration main frequency reference value of 35 Hz. 10,000 sets of complete samples were extracted from the oil pump operation history database. Among them, the temperature mean fluctuation range of 4,000 sets of samples in normal operation was between 35 and 37 degrees Celsius, the vibration amplitude of 2,500 sets of samples in bearing wear state exceeded 0.8 mm per square second, the temperature variance of 2,000 sets of samples in seal leakage state was greater than 6 square degrees Celsius, and the vibration main frequency of 1,500 sets of samples in impeller wear state deviated from the reference value by more than 5 Hz. In the feature preprocessing stage, the four sets of feature data were normalized, the temperature mean was mapped to the interval of 0.3 to 0.7, the vibration amplitude was mapped to the interval of 0.2 to 0.8, the temperature variance was mapped to the interval of 0.4 to 0.6, and the vibration main frequency was mapped to the interval of 0.25 to 0.75. The calculated feature correlation coefficient matrix showed that the correlation coefficient between the temperature mean and the vibration amplitude was 0.85, and the correlation coefficient between the temperature variance and the vibration main frequency was 0.78. After the principal component conversion, the contribution rates of the three feature dimensions were 65%, 20%, and 10%, respectively.In the classifier training stage, the random forest setting includes 500 decision trees, and the maximum depth of each tree is limited to 8 layers to prevent overfitting. The leaf node contains at least 20 samples to ensure generalization ability. The feature sampling ratio is 0.8 to improve the stability of the model. By searching the parameters every 0.01 between the learning rate of 0.01 and 1, it is finally determined that the accuracy of the five-fold cross validation reaches the highest value of 92% when the learning rate is 0.05. In the fault diagnosis prediction stage, 2000 groups of test samples are input into the trained classifier, and the Mahalanobis distance from the sample point to the center of the four categories in the three-dimensional feature space is calculated. The center coordinates of the normal state category (0.5, 0.5, 0.5), the center coordinates of the bearing wear category (0.3, 0.7, 0.4), the center coordinates of the seal leakage category (0.7, 0.4, 0.6), and the center coordinates of the impeller wear category (0.4, 0.6, 0.3) are output when the confidence of the closest distance category exceeds 0.85, otherwise it is marked as pending.

[0031] According to the definition of preset fault types, the fault types are divided into bearing fault, pump body fault, seal fault and motor fault, and a fault type label is assigned to each temperature and vibration feature sample.

[0032] The operation specification and physical characteristics of the oil pressure pump are obtained, and four preset fault libraries of bearing fault, pump body fault, seal fault and motor fault are constructed according to the operation specification. The fault library is used to divide the samples of temperature data and vibration data; the temperature and vibration data are preprocessed by a fifth-order Butterworth low-pass filter, and the preprocessed data after noise removal is obtained by four-layer decomposition of the db4 wavelet basis function; the temperature characteristic value and the vibration characteristic value are calculated in the sliding time window according to the preprocessed data, and the characteristic data mapped to the interval of zero to one is obtained by maximum and minimum value normalization; a multi-level threshold marking rule is established for the characteristic data, and the marking rule includes a bearing fault temperature mean classification threshold, a vibration amplitude classification threshold, a pump body fault temperature variance classification threshold, a vibration frequency deviation classification threshold, a seal fault temperature change rate classification threshold, and a motor fault vibration frequency classification threshold; if the temperature feature meets a single fault criterion, the initial label is marked, and if the vibration feature meets the corresponding criterion, the label is confirmed. When the two feature criteria are inconsistent, the label with a high severity level is selected, and when multiple features meet different fault criteria at the same time, the combined fault label is marked, and when the criterion boundary conditions are not met, the normal state label is marked.

[0033] Specifically, according to the operating specifications and physical characteristics of the oil pressure pump, a preset fault library including bearing fault, pump body fault, seal fault and motor fault was constructed. The temperature data with a collection frequency of 1000 Hz and the vibration data with a collection frequency of 2000 Hz were sampled for 60 seconds. The samples were divided according to the number of fault types in the ratio of 3000:2500:2500:2000 in 10,000 groups of original data. The divided temperature and vibration data were preprocessed using a fifth-order Butterworth low-pass filter, and the four-layer decomposition of the db4 wavelet basis function was selected to remove data noise. The four time domain eigenvalues ​​of temperature mean, variance, kurtosis and skewness and the four frequency domain eigenvalues ​​of vibration signal amplitude, frequency, waveform factor and peak factor were calculated within a 30-second sliding time window, and mapped to the interval of zero to one through maximum and minimum value normalization. Based on the four types of fault judgment criteria, a multi-level threshold marking rule is established. The bearing fault temperature mean classification threshold is 45 / 50 / 55 degrees Celsius, corresponding to mild / moderate / severe, the vibration amplitude classification threshold is 0.8 / 1.0 / 1.2 mm / s2, the pump body fault temperature variance classification threshold is 2 / 3 / 4 degrees Celsius, the vibration frequency deviation classification threshold is 2 / 3 / 4 Hz, the seal fault temperature change rate classification threshold is 5 / 10 / 15 degrees Celsius per minute, and the motor fault vibration frequency classification threshold is 60 / 80 / 100 Hz. The feature samples are automatically marked. If the temperature feature meets a single fault judgment criterion, the initial label is marked. If the vibration feature meets the corresponding criterion, the label is confirmed. When the two feature criteria are inconsistent, the label with a higher severity level is preferred. When multiple features meet different fault judgment criteria at the same time, a combined fault label is marked. If the boundary conditions of the judgment criteria are not met, a normal state label is marked. The sample label is updated every 60 seconds. During the construction of the preset fault label system for the oil pressure pump, the temperature sensor sampling frequency was set to 1000 Hz with a data collection accuracy of 0.1 degrees Celsius, and the vibration sensor sampling frequency was set to 2000 Hz with a data collection accuracy of 0.01 mm per square second. 3000 groups of bearing fault data among the 10,000 groups of original data were marked in real time. Among them, the average temperature of 1500 groups with mild faults was 45-50 degrees Celsius, the average temperature of 1000 groups with moderate faults was 50-55 degrees Celsius, and the average temperature of 500 groups with severe faults exceeded 55 degrees Celsius. At the same time, the vibration amplitude was graded according to the three-level thresholds of 0.8 / 1.0 / 1.2 mm per square second. In the data preprocessing stage, the cutoff frequency of the fifth-order Butterworth low-pass filter was set to 100 Hz to filter out high-frequency noise. The wavelet transform used the four-layer decomposition of the db4 basis function to remove the baseline drift. The temperature data extracted four statistical values ​​within a 30-second time window, including the mean of 36 degrees Celsius, the variance of 4 square degrees Celsius, the kurtosis of 2.8, and the skewness of 0.15. The vibration data extracted four eigenvalues, including the amplitude of 0.5 mm per square second, the main frequency of 35 Hz, the waveform factor of 1.5, and the peak factor of 3.0, and mapped them to the interval of zero to one through normalization.Among the 2500 groups of samples of pump body failure, the three levels of light, medium and severe are divided according to the temperature variance of 2 / 3 / 4 degrees Celsius and the frequency deviation of 2 / 3 / 4 Hz. The 2500 groups of samples of seal failure are graded according to the temperature change rate of 5 / 10 / 15 degrees Celsius per minute and the vibration amplitude of 0.2 / 0.4 / 0.6 mm per square second. The 2000 groups of samples of motor failure are graded based on the vibration frequency of 60 / 80 / 100 Hz and the temperature average of 40 / 45 / 50 degrees Celsius. The highest level criterion is used for combined failure. The label allocation priority from high to low is motor failure, bearing failure, pump body failure, and seal failure. When the temperature feature meets a single fault criterion, the initial label is recorded. When the vibration feature meets the corresponding criterion at the same time, the fault label is confirmed. When the two features conflict, the higher severity level label is selected. When the feature meets multiple fault criteria, the combined fault is marked. The label update cycle is 60 seconds, and the original label is kept during the update process until the new feature meets other criteria.

[0034] S104. Based on the oil pump fault diagnosis model, typical faults of the oil pump such as bearing wear and poor gear meshing are classified and identified, and the fault diagnosis results of the oil pump temperature and vibration are determined through the mapping relationship between the oil pump temperature and vibration characteristics and the fault type.

[0035] The vibration data is fast Fourier transformed by using Hanning window, and the envelope demodulation signal is obtained by Hilbert transform, and the bearing inner and outer ring fault frequency and gear meshing frequency characteristics are obtained from the envelope demodulation signal; a reference threshold is set according to the bearing inner and outer ring fault frequency characteristics and the gear meshing frequency characteristics, and the reference threshold includes the bearing inner ring fault frequency amplitude, the outer ring fault frequency amplitude and the gear meshing frequency amplitude; according to the percentage of the vibration amplitude exceeding the reference threshold, the fault level is divided into mild fault, moderate fault and severe fault, and the mild fault corresponds to an excess percentage of 20%, the moderate fault corresponds to an excess percentage of 40%, and the severe fault corresponds to an excess percentage of 60%; the bearing fault samples and gear fault samples are trained by using gradient boosting tree, and the optimal model parameters are determined by ten-fold cross validation, and the optimal model parameters include the number of trees, learning rate, feature sampling rate and the minimum number of samples of leaf nodes; the feature weight is determined according to the temperature feature and the vibration feature, and the online prediction is performed for the new collected data feature. If the confidence of the prediction result exceeds the preset threshold and the prediction results of the preset number of consecutive time windows are consistent, the fault type and severity are output.

[0036] Specifically, according to the oil pump speed of 1500 rpm, the characteristic frequency of the inner ring fault of the bearing is 85 Hz, the characteristic frequency of the outer ring fault is 65 Hz, and the gear meshing base frequency is 110 Hz. The vibration data with a sampling frequency of 2000 Hz is fast Fourier transformed using a 2048-point Hanning window with a frequency resolution of 1 Hz. The vibration signal is envelope demodulated by Hilbert transform to extract the modulation frequency sideband characteristics. The amplitude of 0.6 mm / s2 at the bearing inner ring fault frequency of 85 Hz, the amplitude of 0.5 mm / s2 at the outer ring fault frequency of 65 Hz, and the amplitude of 0.4 mm / s2 at the gear meshing frequency of 110 Hz in the envelope spectrum are set as the reference threshold. The temperature mean of 40 degrees Celsius and the temperature variance of 2 degrees Celsius are recorded as the reference value in the 120-second observation window. According to the vibration amplitude exceeding the reference value by 20%, 40%, and 60%, the three levels of faults are mild, moderate, and severe. Based on the gradient boosting tree, 5000 groups of bearing fault samples and 5000 groups of gear fault samples were trained. Each type of fault included 2000 mild groups, 2000 moderate groups, and 1000 severe groups. The optimal parameters were determined by ten-fold cross validation, with 800 trees, a learning rate of 0.05, a feature sampling rate of 0.8, and a minimum number of leaf node samples of 30. The optimal model was selected based on the accuracy of the validation set. The feature weights were determined based on the change trend of the temperature and vibration features within a 120-second time window. The weights of the temperature mean and vibration amplitude were 0.3, the temperature variance and frequency feature weights were 0.2, the waveform factor and peak factor weights were 0.15, and the temperature kurtosis and skewness weights were 0.15. The features of the newly collected data were predicted online. When the confidence of the prediction results exceeded 0.9 and the prediction results of three consecutive time windows were consistent, the fault type and severity were output. During the implementation of the oil pump fault diagnosis, the sampling parameters are set according to the speed of 1500 revolutions per minute, the temperature sensor sampling frequency is 1000 Hz with an accuracy of 0.1 degrees Celsius, the vibration sensor sampling frequency is 2000 Hz with an accuracy of 0.01 mm per square second, and the inner ring characteristic frequency is 85 Hz and the outer ring characteristic frequency is 65 Hz based on the bearing outer diameter of 90 mm, inner diameter of 45 mm, rolling element diameter of 12 mm, and 12 rolling elements. The gear module is 4 mm and the number of teeth is 28, and the meshing fundamental frequency is 110 Hz. The vibration signal is subjected to a 2048-point Hanning window for fast Fourier transform to obtain a spectrum diagram, and then the envelope spectrum is extracted through the Hilbert transform. It can be seen from the envelope spectrum that the inner ring fault sideband interval is 25 Hz, the outer ring fault sideband interval is 18 Hz, and the gear meshing sideband interval is 15 Hz.In the characteristic threshold judgment stage, based on 10,000 sets of historical data, the vibration amplitude of the bearing inner ring frequency under normal operating conditions is 0.6 mm / s2, the vibration amplitude of the outer ring frequency is 0.5 mm / s2, and the vibration amplitude of the gear meshing frequency is 0.4 mm / s2. The temperature reference mean is 40 degrees Celsius, and the temperature reference variance is 2 degrees Celsius. Within the 120-second observation window, if the vibration amplitude exceeds the reference value by 20% and the temperature mean exceeds the reference value by 10%, it is judged as a minor fault; if the vibration amplitude exceeds 40% and the temperature mean exceeds 20%, it is judged as a moderate fault; if the vibration amplitude exceeds 60% and the temperature mean exceeds 30%, it is judged as a severe fault. During the gradient boosting tree training stage, 10,000 groups of labeled samples were used to construct the training set, including 2,000 groups of mild bearing faults with an average temperature of 44 degrees Celsius and a vibration amplitude of 0.72 mm / s2, 2,000 groups of moderate bearing faults with an average temperature of 48 degrees Celsius and a vibration amplitude of 0.84 mm / s2, and 1,000 groups of severe bearing faults with an average temperature of 52 degrees Celsius and a vibration amplitude of 0.96 mm / s2. The gear fault samples were distributed in the same proportion, and the optimal parameter combination was selected through ten-fold cross validation: the number of trees was 800, the learning rate was 0.05, the feature sampling rate was 0.8, and the minimum number of samples for leaf nodes was 30. In the fault diagnosis and prediction stage, a feature vector is constructed based on the temperature mean and vibration amplitude with a weight of 0.3, the temperature variance and frequency characteristics with a weight of 0.2, the waveform factor and peak factor with a weight of 0.15, and the temperature kurtosis and skewness with a weight of 0.15. A sliding prediction is performed on the newly collected data features within a 120-second time window. When the prediction results of three consecutive windows are consistent and the confidence level exceeds 0.9, the fault diagnosis results are output, including three dimensional information: fault type, fault location, and severity.

[0037] For the identification of bearing wear and poor gear meshing of the oil pressure pump, the vibration amplitude of the bearing is monitored and the vibration amplitude threshold is preset to determine the bearing wear fault; the gear meshing vibration signal is monitored and the signal higher than the preset amplitude is detected to determine the poor gear meshing.

[0038] Vibration data is obtained from a vibration sensor of the oil pressure pump, and a vibration signal with a preset sampling frequency and a preset accuracy is obtained through the oil pressure pump vibration sensor; time window segmentation processing is performed according to the vibration signal, the segmented data is filtered by a Butterworth high-pass filter, and the filtered data is windowed by a Hanning window function to obtain a preprocessed signal; a Hilbert transform is performed on the preprocessed signal, an envelope demodulation signal is obtained through the transform, and modulation characteristics are obtained from the envelope demodulation signal; a spectrum diagram is calculated according to the modulation characteristics, and the bearing characteristic frequency and frequency multiple vibration amplitude, the gear meshing frequency and the sideband frequency vibration amplitude are extracted from the spectrum diagram to obtain a characteristic parameter sequence; if the bearing characteristic frequency vibration amplitude exceeds a preset reference threshold and the frequency multiple vibration amplitude does not exceed the standard, it is determined that the bearing is in a light wear state; if the gear meshing frequency vibration amplitude exceeds the preset reference threshold and the sideband frequency vibration amplitude does not exceed the standard, it is determined that the gear is in a light meshing state; if the characteristic parameter sequence exceeds the preset number of determinations continuously, the corresponding fault level is output.

[0039] Specifically, vibration data is collected from the oil pump vibration sensor, the sampling frequency is set to 20,000 Hz, the data collection accuracy is 0.001 mm / s2, the collected data is segmented according to a 10-second time window and a 50% data overlap rate is used, the low-frequency drift below 5 Hz is removed by a fifth-order Butterworth high-pass filter, a 4096-point Hanning window is used to perform windowing on the segmented data, and the signal is envelope demodulated based on the Hilbert transform to obtain the modulation characteristics. The vibration amplitudes at the bearing characteristic frequency of 85 Hz and its double frequency of 170 Hz and triple frequency of 255 Hz, the gear meshing frequency of 110 Hz and its sideband frequencies of 95 Hz and 125 Hz are extracted from the spectrum diagram, the spectrum kurtosis value and the frequency band energy ratio are calculated, and the base thresholds are set for the bearing base frequency amplitude of 0.5 mm / s2 and the double frequency amplitude of 0.3 mm / s2, and the base thresholds are set for the gear meshing base frequency amplitude of 0.4 mm / s2 and the sideband amplitude of 0.2 mm / s2. According to the vibration characteristics of bearings, fault judgment criteria are established based on the three parameters of fundamental frequency amplitude, frequency multiple amplitude, and spectrum kurtosis. When the fundamental frequency amplitude exceeds the reference threshold by 20% and the frequency multiple amplitude does not exceed the standard, it is judged as mild wear. When the fundamental frequency exceeds 40% and the frequency multiple exceeds 20%, it is judged as moderate wear. When the fundamental frequency exceeds 60% and the frequency multiple exceeds 40% or the spectrum kurtosis is greater than 3.5, it is judged as severe wear. When the calculation result is lower than the threshold, it is judged as normal state. According to the vibration characteristics of gears, fault judgment criteria are established based on the three parameters of meshing frequency amplitude, sideband amplitude, and frequency band energy ratio. When the meshing frequency amplitude exceeds the reference threshold by 20% and the sideband amplitude does not exceed the standard, it is judged as mild meshing mismatch. When the meshing frequency exceeds 40% and the sideband exceeds 20%, it is judged as moderate meshing mismatch. When the meshing frequency exceeds 60% and the sideband exceeds 40% or the frequency band energy ratio is greater than 0.6, it is judged as severe meshing mismatch. When the judgment results are consistent in three consecutive time windows, the final fault state is output. In the specific implementation process of oil pump fault diagnosis, the vibration sensor sampling frequency is set to 20,000 Hz to collect data, and the data accuracy reaches 0.001 mm per square second. The collected original signal is filtered out by a fifth-order Butterworth high-pass filter to remove low-frequency drift components below 5 Hz. The filtered signal is segmented according to a 10-second time window, and the overlap rate of adjacent windows is set to 50%. Each data segment is weighted by a 4096-point Hanning window, and the spectrum characteristics are analyzed in the frequency range of 0-10,000 Hz. The frequency resolution reaches 2.44 Hz, and the envelope spectrum is obtained through Hilbert transform to extract the modulation frequency characteristics.In the bearing fault diagnosis example, when the bearing is operating normally, the vibration amplitude at the fundamental frequency of 85 Hz is 0.5 mm / s2, the vibration amplitude at the second harmonic frequency of 170 Hz is 0.3 mm / s2, the vibration amplitude at the third harmonic frequency of 255 Hz is 0.2 mm / s2, and the spectrum kurtosis value is 2.8. When it is detected that the fundamental frequency amplitude rises to 0.6 mm / s2 and the harmonic frequency amplitude remains unchanged, it is judged as light wear. When the fundamental frequency amplitude reaches 0.7 mm / s2 and the second harmonic frequency rises to 0.36 mm / s2, it is judged as moderate wear. When the fundamental frequency amplitude exceeds 0.8 mm / s2, the second harmonic frequency exceeds 0.42 mm / s2, or the spectrum kurtosis value exceeds 3.5, it is judged as severe wear. In the gear meshing fault diagnosis example, the fundamental frequency amplitude at 110 Hz is 0.4 mm / s2 during normal meshing, the amplitude at the sideband frequencies of 95 Hz and 125 Hz is 0.2 mm / s2, and the frequency band energy accounts for 0.4. When the meshing frequency amplitude rises to 0.48 mm / s2 but the sideband amplitude does not change much, it is judged as mild meshing mismatch. When the meshing frequency amplitude reaches 0.56 mm / s2 and the sideband amplitude rises to 0.24 mm / s2, it is judged as moderate meshing mismatch. When the meshing frequency amplitude exceeds 0.64 mm / s2 and the sideband amplitude exceeds 0.28 mm / s2, or the frequency band energy accounts for more than 0.6, it is judged as severe meshing mismatch. In actual applications, the data acquisition device completes data segment processing every 10 seconds, obtains spectrum feature parameters through real-time calculation, and outputs the final fault diagnosis result only when the judgment results of three consecutive time windows are consistent, thereby avoiding misjudgment caused by instantaneous interference and achieving stable and reliable fault monitoring.

[0040] S105. Based on the fault diagnosis results of the oil pressure pump temperature and vibration, a comprehensive decision is made on multiple diagnosis results, and corresponding weight coefficients are assigned according to the preset confidence levels of different diagnosis results to generate a final health status assessment result of the oil pressure pump.

[0041] The temperature diagnosis result and vibration diagnosis result of the oil pressure pump are obtained, and a first confidence value is assigned according to the detection accuracy of the temperature diagnosis result, and a second confidence value is assigned according to the detection accuracy of the vibration diagnosis result; for the temperature diagnosis result and the vibration diagnosis result, a triangular membership function is used for fuzzy mapping, wherein the temperature characteristic membership parameter includes a normal interval and an abnormal interval, and the vibration characteristic membership parameter includes a normal interval and an abnormal interval; according to the fuzzy mapping result and the diagnosis weight, the membership scores of the healthy state, the warning state, the fault state, and the dangerous state are obtained by weighted averaging, and the state level is determined according to the maximum membership principle; if the state evaluation value is in a preset interval and remains stable continuously, it is determined to be a healthy state; if the state evaluation value is lower than the preset interval or the temperature diagnosis result and the vibration diagnosis result are abnormal at the same time, it is determined to be a fault state; for the state level, an exponential weighted average is used in a sliding time window to obtain a state evaluation mean and standard deviation, and if the lower limit of the mean confidence interval is higher than a preset threshold and the standard deviation is less than the preset threshold, the final healthy state is output, otherwise it is marked as a state to be observed.

[0042] Specifically, the temperature and vibration diagnosis results of the oil pressure pump are read, and the confidence level of temperature diagnosis is 0.8 and the confidence level of vibration diagnosis is 0.9 based on the detection accuracy and data reliability. The initial weights of the four diagnostic results of abnormal temperature, bearing wear, seal leakage, and motor failure are set to 0.15, 0.35, 0.3, and 0.2, respectively. A 4x4-dimensional diagnostic result data matrix is ​​constructed, and the matrix data is normalized according to the maximum and minimum values. The triangular membership function is used to perform fuzzy mapping on the diagnostic data. The temperature feature membership parameter is set to the normal range of 35-38 degrees Celsius and the abnormal range of 38-45 degrees Celsius. The vibration feature membership parameter is set to the normal range of 0.3-0.5 mm / s2 and the abnormal range of 0.5-0.8 mm / s2. The membership scores of the four states of health, warning, failure, and danger are calculated based on the weighted average, and the state level is determined using the maximum membership principle. The status assessment benchmark is established based on the statistical results of historical data. The oil pump is considered healthy if its evaluation value is between 0.9 and 1.0 and remains stable for 30 consecutive seconds. It is considered to be in a warning state if its evaluation value is between 0.7 and 0.9 or its fluctuation exceeds 0.1. It is considered to be in a fault state if its evaluation value is between 0.5 and 0.7 and the temperature and vibration are abnormal at the same time. It is considered to be in a dangerous state if its evaluation value is lower than 0.5 or a dangerous sign is detected. The status assessment results are dynamically updated within a 60-second sliding time window. The mean and standard deviation of the status assessment are calculated based on the exponential weighted average. If the lower limit of the mean confidence interval is higher than 0.85, the result is considered reliable, and if the standard deviation is less than 0.1, the result is considered stable. The final healthy state is output when both conditions are met. Otherwise, it is marked as a state to be observed and the original assessment result is maintained. During the implementation of the oil pump status assessment, 60 sets of temperature and vibration diagnostic data are collected every minute. The temperature sensor detection accuracy of 0.1 degrees Celsius corresponds to a confidence level of 0.8, and the vibration sensor detection accuracy of 0.01 millimeters per square second corresponds to a confidence level of 0.9. The actual operation data shows that the frequency of abnormal temperature is 15%, the frequency of bearing wear is 35%, the frequency of seal leakage is 30%, and the frequency of motor failure is 20%. Based on these statistical features, the corresponding diagnostic weights are set, and the diagnostic data is standardized to the range of 0-1. In the fuzzy evaluation stage, the temperature data has a membership that increases from 0 to 1 and then decreases to 0 in the range of 35-38 degrees Celsius, and changes in the opposite direction in the range of 38-45 degrees Celsius. The vibration data has a membership that increases from 0 to 1 and then decreases to 0 in the range of 0.3-0.5 millimeters per square second, and changes in the opposite direction in the range of 0.5-0.8 millimeters per square second. The calculated membership is 0.95 for the healthy state, 0.85 for the warning state, 0.65 for the fault state, and 0.45 for the dangerous state.During the status assessment process, the status assessment value of the oil pump is maintained at 0.95 with a fluctuation range of 0.02 when it is operating normally. When a slight abnormality occurs, the assessment value drops to 0.85 and the fluctuation range increases to 0.05. When a fault occurs, the assessment value drops to 0.65 and the temperature rises to 42 degrees Celsius and the vibration amplitude reaches 0.6 mm / s2. When a serious fault occurs, the assessment value is lower than 0.5 and is accompanied by a sharp increase in temperature or a sharp increase in vibration. In the dynamic monitoring stage, a 60-second sliding time window is used to calculate the status assessment results. The mean assessment value is 0.92 and the standard deviation is 0.03 during normal operation, the mean assessment value is 0.83 and the standard deviation is 0.08 for slight abnormalities, and the mean assessment value is 0.61 and the standard deviation is 0.12 for faults. The reliability of the results is determined by the lower limit of the confidence interval of 0.85 and the standard deviation threshold of 0.1, realizing the real-time assessment of the health status of the oil pump.

[0043] With respect to the assignment of weight coefficients, the historical occurrence frequency, severity, and impact of fault types of different diagnostic results are obtained, the weights of fault types of different diagnostic results are set, the preset confidences of different diagnostic results and the weights of fault types are weighted and fused, the comprehensive confidence is calculated, and the comprehensive confidence is compared with the pre-set health status threshold to achieve health status classification, providing a basis for the final health status assessment result.

[0044] Fault record information is obtained from the oil pressure pump operation database, and the fault record information includes four types of fault data, namely, bearing fault, seal leakage, gear wear, and motor fault; three index data, namely, fault frequency, repair time, and loss ratio, are calculated according to the fault record information; the index data is standardized by the principal component conversion method, and a weight coefficient matrix is ​​obtained by calculating the index information entropy, and the weight coefficient matrix corresponds to the three indicators, namely, fault frequency, repair time, and loss ratio; a multi-source information fusion weight matrix is ​​constructed according to the weight coefficient matrix, temperature diagnosis confidence, and vibration diagnosis confidence, and the multi-source information fusion weight matrix is ​​updated according to a preset time period; an exponential decay function is used to calculate the time weight of the multi-source information fusion weight matrix, and a state evaluation score is obtained after removing abnormal fluctuation data points by a sliding variance method; the health level of the oil pressure pump is divided according to the state evaluation score, and if the state evaluation score is higher than the first threshold, it is determined to be completely healthy; if the state evaluation score is between the second threshold and the first threshold, it is determined to be slightly abnormal; if the state evaluation score is between the third threshold and the second threshold, it is determined to be a fault warning; if the state evaluation score is lower than the third threshold, it is determined to be a fault state.

[0045] Specifically, the fault records within 90 days were read from the historical operation database of the oil pump. Based on the continuous monitoring data of the vibration temperature sensor with a sampling frequency of 1000 Hz, the number of occurrences of four types of faults, namely bearing fault, seal leakage, gear wear, and motor fault, was counted. The repair time from triggering the alarm to resuming normal operation for each type of fault was recorded, and the percentage of production loss caused by a single fault was calculated. The fault type frequency distribution was obtained as 0.3, 0.25, 0.25, and 0.2, the average repair time was 8 hours, 6 hours, 4 hours, and 6 hours, and the production loss ratio was 0.4, 0.3, 0.2, and 0.3. Based on the principal component transformation, the data of the three indicators of fault frequency, repair time, and loss ratio were standardized, and the information entropy of the indicators was calculated to obtain weight coefficients of 0.4, 0.3, and 0.3. The weighted sum of the weight indicators of the four types of faults was obtained to obtain comprehensive scores of 0.35, 0.25, 0.2, and 0.2. Combined with the temperature diagnosis confidence of 0.85 and the vibration diagnosis confidence of 0.9, a multi-source information fusion weight matrix was constructed to achieve the weight coefficient update every 24 hours. The time weight is calculated using an exponential decay function, with a half-life of 30 days. The time decay coefficient is assigned to the diagnosis results within 90 days. The weight matrix and the time weight coefficient are integrated to calculate the status assessment score. The sliding variance method is used to eliminate abnormal fluctuation data points. The health status change trend curve of the oil pump is obtained through least squares fitting, and the status mean is 0.85 and the standard deviation is 0.08. The status level is divided according to the health status mean and three times the standard deviation principle of the oil pump. The status assessment score above 0.93 is considered completely healthy, the 0.85 to 0.93 range is considered slightly abnormal, the 0.77 to 0.85 range is considered a fault warning, and the score below 0.77 is considered a fault state. The status assessment result is calculated using a 60-minute sliding time window. If the judgment results of three consecutive windows are consistent, the health status is updated. During the implementation of the weight coefficient assignment for the oil pressure pump, the data acquisition unit continuously collected temperature and vibration data at a frequency of 1000 Hz. Within 90 days, 45 bearing failures were recorded, with an average repair time of 8 hours and a production loss of 40%, 38 seal leakages were recorded, with an average repair time of 6 hours and a production loss of 30%, 37 gear wears were recorded, with an average repair time of 4 hours and a production loss of 20%, and 30 motor failures were recorded, with an average repair time of 6 hours and a production loss of 30%. Based on these original data, the fault frequency distributions of 0.3, 0.25, 0.25, and 0.2 were calculated, and the three-dimensional indicators were mapped to a unified scoring space through principal component transformation. In the weight calculation stage, the information entropy is calculated for the standardized indicator data. The frequency index entropy value is 0.95, with an information volume of 0.4; the repair index entropy value is 0.92, with an information volume of 0.3; the loss index entropy value is 0.92, with an information volume of 0.3. The weighted scores of the four types of faults are 0.35, 0.25, 0.2, and 0.2, respectively. Combined with the temperature sensor detection accuracy of 0.1 degrees Celsius corresponding to a confidence level of 0.85 and the vibration sensor detection accuracy of 0.01 millimeters per square second corresponding to a confidence level of 0.9, an 8-dimensional weight fusion matrix is ​​constructed, and the weight coefficients are dynamically updated every 24 hours based on the newly added data.In the time series data processing, the exponential decay function is used to calculate the time weight. The weight of the data 30 days ago decays to 0.5, the weight of the data 60 days ago decays to 0.25, and the weight of the data 90 days ago decays to 0.125. The corresponding time weight is assigned to 10,000 groups of diagnosis results in the last 90 days. The sliding variance is used to remove abnormal points with fluctuations exceeding three times the standard deviation. The health status change curve is fitted based on the least squares method, and the state mean is 0.85 and the standard deviation is 0.08. In the health assessment stage, when the oil pressure pump operation status score is above 0.93 and remains stable for three consecutive 60-minute windows, it is judged to be completely healthy. The score fluctuates in the range of 0.85-0.93, which is considered to be slightly abnormal. The score drops to the range of 0.77-0.85, which is considered to be a fault warning. The score is lower than 0.77 or drops sharply by more than 0.15 in a short period of time, which is considered to be a fault state. The health status assessment results of the oil pressure pump are updated through real-time calculation.

[0046] S106. Based on the health status assessment result of the oil pressure pump, combined with the pre-set health threshold and alarm rules, determine whether the oil pressure pump is in an abnormal working state. If the health index of the oil pressure pump is lower than the preset threshold or the corresponding alarm rule is triggered, a fault warning signal is issued to prompt maintenance or inspection of the oil pressure pump.

[0047] The operation status data of the oil pressure pump is obtained, and the temperature parameter characteristic values ​​and the vibration parameter characteristic values ​​are extracted according to the operation status data. The temperature mean, temperature standard deviation, vibration amplitude mean, and vibration standard deviation are calculated by the parameter characteristic values; the health status benchmark threshold is set according to the temperature mean, temperature standard deviation, vibration amplitude mean, and vibration standard deviation by adopting the principle of three times the standard deviation, and the temperature parameter and vibration parameter three-level alarm threshold corresponding to the bearing component, the sealing component, and the gear component are determined according to the health status benchmark threshold; the real-time state evaluation value of the oil pressure pump is received, and it is determined whether the state evaluation value is lower than the health status benchmark threshold. If the duration of the state evaluation value reaches a preset duration, a corresponding level warning signal is triggered; for the warning signal, the XGBoost algorithm is used to train the historical operation data, and a warning rule library is constructed through the temperature parameter characteristic values ​​and the vibration parameter characteristic values. The warning rule library outputs the proportion of the warning types of bearing failure, seal leakage, gear wear, and motor failure; the oil pressure pump state data corresponding to the warning signal is recorded, and the warning threshold is updated by using the exponential sliding average, and the warning accuracy is determined according to the warning record. If the warning accuracy is lower than the preset threshold, a rule optimization instruction is triggered.

[0048] Specifically, the normal state parameter characteristics were extracted from the 30-day operation data of the oil pressure pump, and the temperature mean was calculated to be 38 degrees Celsius with a standard deviation of 2 degrees Celsius, and the vibration amplitude was 0.4 mm / s2 with a standard deviation of 0.1 mm / s2. Based on the principle of three times the standard deviation, the health status benchmark threshold of 0.9 was set, and the third level alarm thresholds were set for bearing temperature of 42 / 45 / 48 degrees Celsius, vibration amplitude of 0.6 / 0.7 / 0.8 mm / s2, sealing temperature of 34 / 36 / 38 degrees Celsius, vibration amplitude of 0.3 / 0.4 / 0.5 mm / s2, gear temperature of 40 / 43 / 46 degrees Celsius, and vibration amplitude of 0.5 / 0.6 / 0.7 mm / s2. The warning is triggered based on the comparison between the state evaluation value of the oil pump and the threshold. When the state evaluation value is between 0.8 and 0.9 for 30 seconds, the attention warning is triggered. When the evaluation value is between 0.7 and 0.8 for 20 seconds, the warning warning is triggered. When the evaluation value is lower than 0.7 for 10 seconds, the danger warning is triggered. When a single parameter exceeds the corresponding time of the third-level alarm threshold, the corresponding level of warning is triggered. The false alarm rate is calculated through the latest 5,000 warning records, and the warning rules with a false alarm rate exceeding 5% are corrected. XGBoost is used to train the warning rules for confirmed fault cases in the historical operation data of the oil pump. The warning rule library is constructed based on six features: temperature mean, temperature change rate, vibration amplitude, vibration frequency, speed, and pressure. The rule input is the feature real-time value, 10-second mean, and 60-second mean. The rule output is the proportion of warning types. The warning types include bearing failure, seal leakage, gear wear, and motor failure. The characteristic values, warning types, duration, and actual fault types of the oil pressure pump status data that trigger the warning signal are recorded. The exponential sliding average with a decay coefficient of 0.1 is used to update the warning threshold. The repair time and cost are recorded to establish a warning loss evaluation index. The threshold parameters and rule structure are optimized based on the most recent 10,000 warning records. When the warning accuracy is lower than 0.8 or the loss index exceeds expectations, rule optimization is automatically triggered. During the specific implementation of the oil pump fault warning, the normal operating parameter distribution was statistically analyzed through the operation data collected for 30 consecutive days. The temperature data had an average of 38 degrees Celsius and a standard deviation of 2 degrees Celsius. The vibration amplitude had an average of 0.4 mm / s2 and a standard deviation of 0.1 mm / s2. On this basis, the health status benchmark threshold of 0.9 was set, and three levels of warning thresholds were formulated for different components: bearing temperature of 42 / 45 / 48 degrees Celsius corresponded to attention / warning / danger levels, with a vibration amplitude of 0.6 / 0.7 / 0.8 mm / s2; sealing temperature of 34 / 36 / 38 degrees Celsius, with a vibration amplitude of 0.3 / 0.4 / 0.5 mm / s2; gear temperature of 40 / 43 / 46 degrees Celsius, with a vibration amplitude of 0.5 / 0.6 / 0.7 mm / s2.During the early warning trigger judgment stage, when the oil pressure pump status assessment value is in the range of 0.8-0.9 for 30 seconds, the temperature rises to 42 degrees Celsius or the vibration amplitude reaches 0.6 mm per square second, triggering a concern-level alert; the status assessment value drops to the range of 0.7-0.8 for 20 seconds, the temperature reaches 45 degrees Celsius or the vibration amplitude reaches 0.7 mm per square second, triggering a warning-level alert; the status assessment value is lower than 0.7 for 10 seconds, the temperature exceeds 48 degrees Celsius or the vibration amplitude exceeds 0.8 mm per square second, triggering a danger-level alert. The false alarm rate calculated based on the most recent 5,000 warning records is 4.8%, which confirms the effectiveness of the early warning rules. The warning rule training uses an XGBoost model with 300 decision trees. The input features include a temperature mean of 38 degrees Celsius, a temperature change rate of 0.5 degrees Celsius per minute, a vibration amplitude of 0.4 millimeters per square second, a vibration frequency of 35 Hz, a rotation speed of 1500 revolutions per minute, and a pressure of 2.5 MPa. For each feature, the real-time value, 10-second mean, and 60-second mean are calculated, totaling 18-dimensional inputs. The output is the probability of four types of fault warnings, with an accuracy rate of 85%. In the early warning rule optimization phase, the 18-dimensional feature data, warning level, duration, and actual fault type are recorded after each warning is triggered. The exponential sliding average is calculated using a 0.1 attenuation coefficient to update the warning threshold. The latest 10,000 warning records show an accuracy rate of 82%, an average repair time of 6 hours, and an average maintenance cost of 5,000 yuan. Based on these indicators, the warning rules and threshold parameters are continuously optimized, and the rule update is automatically triggered when the accuracy rate is lower than 80%.

[0049] The present invention provides a real-time monitoring system for auxiliary equipment of a hydropower station based on the Internet of Things, which mainly includes:

[0050] The data acquisition and preprocessing module is used to collect the temperature data and vibration data of the oil pressure pump, preprocess the collected temperature data and vibration data, remove the abnormal values ​​and noise interference, obtain the time series data of the oil pressure pump temperature and vibration, and perform feature extraction. Through time domain analysis and frequency domain analysis, the statistical characteristics and spectrum characteristics of the oil pressure pump temperature and vibration signals are extracted to obtain the characteristic vector of the oil pressure pump temperature and vibration characteristics;

[0051] A correlation analysis module is used to perform correlation analysis on the characteristic vectors of the temperature and vibration characteristics of the oil pressure pump according to the historical working data of the oil pressure pump, calculate the correlation coefficient matrix between different characteristics, and obtain the temperature and vibration related feature subset by screening out the features in the matrix whose correlation between the temperature and vibration characteristics is higher than the preset threshold;

[0052] The fault diagnosis modeling module is used to build a fault diagnosis model for the oil pump based on the temperature and vibration related feature subsets. By training the temperature and vibration feature samples of the preset fault types, the mapping relationship between the temperature and vibration features of the oil pump and the fault type is established.

[0053] The typical fault classification and identification module is used to classify and identify the typical faults of bearing wear and gear meshing failure of the oil pump based on the oil pump fault diagnosis model, and determine the fault diagnosis results of the oil pump temperature and vibration through the mapping relationship between the oil pump temperature and vibration characteristics and the fault type;

[0054] The comprehensive decision-making module is used to make a comprehensive decision on multiple diagnosis results based on the fault diagnosis results of the oil pump temperature and vibration, assign corresponding weight coefficients according to the preset confidence of different diagnosis results, and generate the final health status assessment result of the oil pump;

[0055] The health status assessment and early warning module is used to determine whether the oil pressure pump is in an abnormal working state based on the health status assessment results of the oil pressure pump, combined with the pre-set health thresholds and alarm rules. If the health index of the oil pressure pump is lower than the preset threshold or the corresponding alarm rules are triggered, a fault warning signal is issued to prompt maintenance or inspection of the oil pressure pump.

[0056] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A real-time monitoring method for auxiliary equipment of a hydropower station based on the Internet of Things, characterized in that: The method comprises: Collect the temperature data and vibration data of the oil pump, pre-process the collected temperature data and vibration data, remove the abnormal values ​​and noise interference, obtain the time series data of the oil pump temperature and vibration, and perform feature extraction. Through time domain analysis and frequency domain analysis, extract the statistical characteristics and spectrum characteristics of the oil pump temperature and vibration signals, and obtain the characteristic vector of the oil pump temperature and vibration characteristics; According to the historical working data of the oil pump, the correlation analysis is performed on the characteristic vectors of the temperature and vibration characteristics of the oil pump, and the correlation coefficient matrix between different characteristics is calculated. By screening out the features in the matrix whose correlation between temperature and vibration characteristics is higher than the preset threshold, the temperature and vibration related feature subset is obtained; A fault diagnosis model for the oil pump is constructed for the subset of temperature and vibration related features. By training the temperature and vibration feature samples of the preset fault types, a mapping relationship between the temperature and vibration features of the oil pump and the fault type is established. Based on the fault diagnosis model of the oil pump, the typical faults of the oil pump such as bearing wear and gear meshing failure are classified and identified. The fault diagnosis results of the oil pump temperature and vibration are determined through the mapping relationship between the oil pump temperature and vibration characteristics and the fault type. According to the fault diagnosis results of the oil pump temperature and vibration, a comprehensive decision is made on multiple diagnosis results, and corresponding weight coefficients are assigned according to the preset confidence of different diagnosis results to generate the final health status assessment result of the oil pump; Based on the health status assessment results of the oil pressure pump, combined with the pre-set health thresholds and alarm rules, it is determined whether the oil pressure pump is in an abnormal working state. If the health index of the oil pressure pump is lower than the preset threshold or the corresponding alarm rules are triggered, a fault warning signal is issued to prompt maintenance or inspection of the oil pressure pump.

2. The method according to claim 1, characterized in that The temperature data and vibration data of the oil pressure pump are collected, and the collected temperature data and vibration data are preprocessed to remove abnormal values ​​and noise interference therein, so as to obtain time series data of the temperature and vibration of the oil pressure pump, and feature extraction is performed, and statistical features and spectrum features of the temperature and vibration signals of the oil pressure pump are extracted through time domain analysis and frequency domain analysis to obtain feature vectors of the temperature and vibration characteristics of the oil pressure pump, including: According to the Nyquist sampling theorem, the temperature data and vibration data of the oil pressure pump are collected and timestamped, wherein the collection frequency of the temperature data is to collect a first preset number of data points per second, and the collection frequency of the vibration data is to collect a second preset number of data points per second; The temperature data is filtered using a Butterworth low-pass filter, and outliers are removed from the filtered temperature data using the triple standard deviation principle to obtain processed temperature data; The vibration data is subjected to four-layer wavelet decomposition using the db four-wavelet basis function, and the processed vibration data is obtained by removing the high-frequency noise components; Performing segmented calculation on the processed temperature data according to a preset time window to obtain a time domain eigenvalue, and performing fast Fourier transform on the processed vibration data to obtain a frequency domain eigenvalue; The time domain eigenvalues ​​and the frequency domain eigenvalues ​​are normalized to obtain normalized eigenvalues, the Pearson correlation coefficient between the normalized eigenvalues ​​is calculated, and a feature combination with a correlation coefficient greater than a preset threshold is selected. The principal component transformation is used to reduce the dimension of the feature combination to obtain a temperature vibration eigenvector of the oil pressure pump.

3. The method according to claim 1, characterized in that According to the historical working data of the oil pressure pump, the characteristic vectors of the temperature and vibration characteristics of the oil pressure pump are subjected to correlation analysis, and the correlation coefficient matrix between different characteristics is calculated. By screening out the characteristics in the matrix whose correlation between the temperature and vibration characteristics is higher than a preset threshold, a subset of temperature and vibration related characteristics is obtained, including: Obtain the temperature feature vector and vibration feature vector from the oil pressure pump history database, align the feature vectors according to the millisecond timestamp, and obtain the standardized feature matrix through the sliding time window; For the standardized feature matrix, four time domain features of mean, variance, kurtosis and skewness in the temperature matrix and four frequency domain features of main frequency, frequency bandwidth, amplitude and waveform factor in the vibration matrix are extracted, and the standardized feature values ​​are obtained by Z-score standardization; According to the standardized characteristic values, a cross-correlation coefficient matrix between the temperature characteristics and the vibration characteristics is calculated to obtain the lag correlation between the characteristics within the time delay range; If the absolute value of the correlation coefficient is greater than the preset first threshold or the absolute value of the lagged correlation coefficient is greater than the preset second threshold, the corresponding feature combination is added to the candidate feature subset; For the feature combinations in the candidate feature subsets, the nonlinear correlation is verified by mutual information entropy, and the temperature vibration related feature subsets are obtained by arranging them in descending order according to the information gain ratio.

4. The method according to claim 1, characterized in that: The oil pump fault diagnosis model is constructed for the temperature and vibration related feature subsets, and the mapping relationship between the oil pump temperature and vibration features and the fault type is established by training the temperature and vibration feature samples of the preset fault type, including: The temperature mean, vibration amplitude, temperature variance and main vibration frequency feature combination in the oil pressure pump label data are obtained, and a training data set is constructed according to the normal operation status, bearing wear status, seal leakage status and impeller wear status ratio mark; For the training data set, the maximum and minimum method is used to normalize the feature data, and the feature vector is converted into a three-dimensional feature space through principal component transformation to obtain the feature components whose cumulative contribution rate meets the requirements; According to the four operating state sample points in the three-dimensional feature space, a classification tree is constructed using a random forest algorithm, and classifier parameters are optimized through grid search to obtain the optimal number of decision trees and learning rate parameters; For the classifier, a test data set is used for verification. If the ratio of the test data set to the total samples is a set value, the classifier performance evaluation result is obtained; The classifier is used to calculate the Mahalanobis distance between the new feature vector and the centroid of each category. If the confidence of the Mahalanobis distance is higher than a threshold, it is determined that the new feature vector corresponds to a fault type label. If the confidence of the Mahalanobis distance is lower than the threshold, it is determined to be in a pending state. It also includes: the definition of preset fault types, classifying the fault types into bearing fault, pump body fault, seal fault and motor fault, and assigning a fault type label to each temperature and vibration feature sample.

5. The method according to claim 1, characterized in that Based on the fault diagnosis model of the oil pump, the typical faults of bearing wear and gear meshing failure of the oil pump are classified and identified, and the fault diagnosis results of the oil pump temperature and vibration are determined through the mapping relationship between the oil pump temperature and vibration characteristics and the fault type, including: The vibration data is subjected to a fast Fourier transform using a Hanning window, and an envelope demodulation signal is obtained through a Hilbert transform, and the fault frequency of the inner and outer rings of the bearing and the gear meshing frequency characteristics are obtained from the envelope demodulation signal; A reference threshold is set according to the bearing inner and outer ring fault frequency characteristics and the gear meshing frequency characteristics, wherein the reference threshold includes the bearing inner ring fault frequency amplitude, the outer ring fault frequency amplitude and the gear meshing frequency amplitude; According to the percentage of the vibration amplitude exceeding the reference threshold, the fault is divided into mild fault, moderate fault and severe fault levels, wherein the mild fault corresponds to a 20% excess percentage, the moderate fault corresponds to a 40% excess percentage, and the severe fault corresponds to a 60% excess percentage; The bearing fault samples and gear fault samples were trained using a gradient boosting tree, and the optimal model parameters were determined through ten-fold cross validation. The optimal model parameters included the number of trees, the learning rate, the feature sampling rate, and the minimum number of samples for leaf nodes. Determine feature weights based on temperature and vibration features, and perform online predictions on newly collected data features. If the confidence level of the prediction results exceeds a preset threshold and the prediction results for a preset number of consecutive time windows are consistent, output the fault type and severity. It also includes: identification of bearing wear and gear meshing problems of the oil pressure pump, by monitoring the vibration amplitude of the bearing and presetting the vibration amplitude threshold to determine the bearing wear fault; By monitoring the gear meshing vibration signal and detecting the signal higher than the preset amplitude, it can be judged that the gear meshing is poor.

6. The method according to claim 5, characterized in that The identification of bearing wear and gear meshing failure of the oil pressure pump includes: monitoring the vibration amplitude of the bearing, presetting the vibration amplitude threshold, so as to judge the bearing wear fault; and monitoring the gear meshing vibration signal, detecting the signal higher than the preset amplitude, so as to judge the gear meshing failure, including: Acquire vibration data from a vibration sensor of the oil pressure pump, and obtain a vibration signal with a preset sampling frequency and a preset accuracy through the vibration sensor of the oil pressure pump; Performing time window segmentation processing according to the vibration signal, filtering the segmented data using a Butterworth high-pass filter, and performing windowing processing on the filtered data using a Hanning window function to obtain a preprocessed signal; Performing Hilbert transform on the preprocessed signal, obtaining an envelope demodulation signal through the transform, and acquiring modulation features from the envelope demodulation signal; Calculate a frequency spectrum according to the modulation characteristics, extract the bearing characteristic frequency and frequency multiple vibration amplitude, the gear meshing frequency and sideband frequency vibration amplitude from the frequency spectrum, and obtain a characteristic parameter sequence; If the bearing characteristic frequency vibration amplitude exceeds the preset reference threshold and the frequency multiplication vibration amplitude does not exceed the standard, the bearing is judged to be in a state of slight wear; If the gear meshing frequency vibration amplitude exceeds the preset reference threshold and the sideband frequency vibration amplitude does not exceed the standard, it is determined that the gear is in a slightly poor meshing state; If the characteristic parameter sequence exceeds the preset number of determination times continuously, the corresponding fault level is output.

7. The method according to claim 1, characterized in that According to the fault diagnosis results of the oil pump temperature and vibration, a comprehensive decision is made on multiple diagnosis results, and corresponding weight coefficients are assigned according to the preset confidence of different diagnosis results to generate the final health status assessment result of the oil pump, including: Acquire a temperature diagnosis result and a vibration diagnosis result of the oil pressure pump, assign a first confidence value according to the detection accuracy of the temperature diagnosis result, and assign a second confidence value according to the detection accuracy of the vibration diagnosis result; For the temperature diagnosis result and the vibration diagnosis result, a triangular membership function is used for fuzzy mapping, wherein the temperature characteristic membership parameter includes a normal interval and an abnormal interval, and the vibration characteristic membership parameter includes a normal interval and an abnormal interval; According to the fuzzy mapping results and diagnostic weights, the membership scores of the healthy state, warning state, fault state and dangerous state are obtained by weighted average, and the state level is determined according to the maximum membership principle; If the status evaluation value is within the preset range and remains stable continuously, it is judged to be in a healthy state; If the state evaluation value is lower than a preset interval or the temperature diagnosis result and the vibration diagnosis result are both abnormal, it is determined to be a fault state; For the state level, exponential weighted averaging is used within the sliding time window to obtain the state assessment mean and standard deviation. If the lower limit of the mean confidence interval is higher than the preset threshold and the standard deviation is lower than the preset threshold, the final health state is output, otherwise it is marked as a state to be observed; It also includes: assigning weight coefficients, obtaining the historical frequency, severity, and impact of fault types with different diagnostic results, setting weights for fault types with different diagnostic results, weighted fusion of preset confidence levels and fault type weights of different diagnostic results, calculating the comprehensive confidence level, comparing the comprehensive confidence level with a preset health status threshold, achieving health status classification, and providing a basis for the final health status assessment result.

8. The method according to claim 7, characterized in that The assignment of weight coefficients, obtaining the historical occurrence frequency, severity, and impact of fault types of different diagnostic results, setting weights of fault types of different diagnostic results, integrating the preset confidence of different diagnostic results and the weights of fault types for weighted fusion, calculating the comprehensive confidence, comparing the comprehensive confidence with the preset health status threshold, realizing health status classification, and providing a basis for the final health status assessment result, include: Acquire fault record information from the oil pressure pump operation database, wherein the fault record information includes four types of fault data: bearing fault, seal leakage, gear wear, and motor fault; Calculate three indicators of fault frequency, repair time and loss ratio according to the fault record information; The index data is standardized by using the principal component conversion method, and a weight coefficient matrix is ​​obtained by calculating the index information entropy. The weight coefficient matrix corresponds to the three indicators of fault frequency, repair time, and loss ratio; A multi-source information fusion weight matrix is ​​constructed according to the weight coefficient matrix, the temperature diagnosis confidence, and the vibration diagnosis confidence, and the multi-source information fusion weight matrix is ​​updated according to a preset time period; An exponential decay function is used to calculate the time weight of the multi-source information fusion weight matrix, and a state assessment score is obtained after removing abnormal fluctuation data points through a sliding variance method; The health level of the oil pressure pump is divided according to the status assessment score. If the status assessment score is higher than the first threshold, it is determined to be completely healthy. If the status assessment score is between the second threshold and the first threshold, it is determined to be slightly abnormal. If the status assessment score is between the third threshold and the second threshold, it is determined to be a fault warning. If the status assessment score is lower than the third threshold, it is determined to be in a fault state.

9. The method according to claim 1, characterized in that: The health status evaluation result of the oil pressure pump is combined with the preset health threshold and alarm rules to determine whether the oil pressure pump is in an abnormal working state. If the health index of the oil pressure pump is lower than the preset threshold or the corresponding alarm rule is triggered, a fault warning signal is issued to prompt maintenance or overhaul of the oil pressure pump, including: Acquire the operating status data of the oil pressure pump, extract the temperature parameter characteristic value and the vibration parameter characteristic value according to the operating status data, and calculate the temperature mean value, temperature standard deviation, vibration amplitude mean value and vibration standard deviation through the parameter characteristic value; According to the temperature mean, temperature standard deviation, vibration amplitude mean and vibration standard deviation, a health status reference threshold is set using the triple standard deviation principle, and three-level alarm thresholds of temperature parameters and vibration parameters corresponding to bearing components, sealing components and gear components are determined based on the health status reference threshold; Receive a real-time status evaluation value of the oil pressure pump, determine whether the status evaluation value is lower than the health status reference threshold, and trigger a corresponding level warning signal if the duration of the status evaluation value reaches a preset duration; For the warning signal, the XGBoost algorithm is used to train the historical operation data, and a warning rule base is constructed through the characteristic values ​​of temperature parameters and vibration parameters. The warning rule base outputs the proportion of bearing failure, seal leakage, gear wear and motor failure warning types; The oil pressure pump status data corresponding to the warning signal is recorded, the warning threshold is updated by exponential sliding average, the warning accuracy is determined according to the warning record, and if the warning accuracy is lower than the preset threshold, the rule optimization instruction is triggered.

10. The real-time monitoring system for auxiliary equipment of hydropower station based on Internet of Things is characterized by: The system comprises: The data acquisition and preprocessing module is used to collect the temperature data and vibration data of the oil pressure pump, preprocess the collected temperature data and vibration data, remove the abnormal values ​​and noise interference, obtain the time series data of the oil pressure pump temperature and vibration, and perform feature extraction. Through time domain analysis and frequency domain analysis, the statistical characteristics and spectrum characteristics of the oil pressure pump temperature and vibration signals are extracted to obtain the characteristic vector of the oil pressure pump temperature and vibration characteristics; A correlation analysis module is used to perform correlation analysis on the characteristic vectors of the temperature and vibration characteristics of the oil pressure pump according to the historical working data of the oil pressure pump, calculate the correlation coefficient matrix between different characteristics, and obtain the temperature and vibration related feature subset by screening out the features in the matrix whose correlation between the temperature and vibration characteristics is higher than the preset threshold; The fault diagnosis modeling module is used to build a fault diagnosis model for the oil pump based on the temperature and vibration related feature subsets. By training the temperature and vibration feature samples of the preset fault types, the mapping relationship between the temperature and vibration features of the oil pump and the fault type is established. The typical fault classification and identification module is used to classify and identify the typical faults of bearing wear and gear meshing failure of the oil pump based on the oil pump fault diagnosis model, and determine the fault diagnosis results of the oil pump temperature and vibration through the mapping relationship between the oil pump temperature and vibration characteristics and the fault type; The comprehensive decision-making module is used to make a comprehensive decision on multiple diagnosis results based on the fault diagnosis results of the oil pump temperature and vibration, assign corresponding weight coefficients according to the preset confidence of different diagnosis results, and generate the final health status assessment result of the oil pump; The health status assessment and early warning module is used to determine whether the oil pressure pump is in an abnormal working state based on the health status assessment results of the oil pressure pump, combined with the pre-set health thresholds and alarm rules. If the health index of the oil pressure pump is lower than the preset threshold or the corresponding alarm rules are triggered, a fault warning signal is issued to prompt maintenance or inspection of the oil pressure pump.

Images