A water pump fault diagnosis system

Abstract

The application discloses a kind of water supply pump fault diagnosis systems, belong to data analysis field, by using synchronous acquisition strategy acquisition water supply pump real-time operation data, construct operation database, and when new data record, detection and self-adaptive discrimination determine early warning decision;When detecting water supply pump abnormal early warning, based on operation database screening for fault identification fault evaluation set, and to the fault evaluation data carries out information fusion, obtains fault evaluation feature set;Through the preliminary fusion data of each fault evaluation subset, carry out comprehensive time series analysis, obtain key position fault evaluation score to distinguish fault, to the key position judged as fault, by constructing time axis obtains fault source, and combines correlation coefficient matrix and constructs fault propagation path.Solved the synchronization and abnormal processing of multi-source information when collecting, the targeted screening and effective integration when fusing, and the technical problems of distinguishing, tracing source when carrying out fault diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data analysis, more particularly to a water supply pump fault diagnosis system. BACKGROUND

[0002] Traditional water supply pump fault diagnosis methods have many limitations. Early on, it mainly relied on manual inspection, and maintenance personnel relied on experience and simple tools to conduct regular visual inspection, sound identification, and temperature touch of the water supply pump. This method is inefficient and highly subjective, and minor fault hazards are easily missed, making it difficult to discover potential problems in a timely manner. With the development of technology, single-monitoring-parameter-based diagnosis methods have emerged, such as monitoring vibration amplitude and frequency only through vibration sensors, or simply tracking temperature changes at key device locations using temperature sensors. However, the industrial site environment is complex and harsh, with many interference factors, and a single parameter is easily distorted by external influences, making it impossible to fully and accurately reflect the true operating state of the device. For example, in a strong electromagnetic environment, the vibration sensor signal can be severely disturbed, leading to false positives for device faults; or when a device has multiple initial fault hazards, single temperature monitoring cannot detect other fault germination.

[0003] A Chinese patent with the publication number CN112684778B discloses a steam generator water supply system fault diagnosis method based on multi-source information reinforcement learning, which includes monitoring the state of the steam generator, water supply pump, and water supply valve in the steam generator water supply system to obtain multiple state information; clustering the associated state information based on information source types to form feature information clusters; integrating the feature information clusters to establish a feature matrix, and extracting a state feature vector from the feature matrix; inputting the state feature vector into a data analysis model system for fault type diagnosis, and evaluating and regulating the state parameters of the corresponding components.

[0004] Although the prior art addresses multi-source information fault diagnosis for steam generator water supply systems, it uses signal processing, mutual information correlation, nonlinear dimensionality reduction, deep reinforcement learning, and other techniques to address feature extraction, feature fusion, feature integration, and feature learning for multi-source information, providing a new technical solution for autonomous learning and state recognition diagnosis of steam generator water supply system multi-source information. It still cannot accurately handle missing values, outliers, and deviation points during data collection and cannot adaptively warn, cannot effectively filter data based on sensor distribution during information fusion, and cannot accurately trace the fault source and construct the fault propagation path during fault diagnosis. Therefore, in order to overcome these limitations, the present application proposes a water supply pump fault diagnosis system. SUMMARY

[0005] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a water pump fault diagnosis system, which solves the problems of synchronization, integrity and real-time performance of data collection in water pump fault diagnosis, as well as the problem of detecting and processing missing values, abnormal points and deviation points in the collected data. It also solves the problem of screening and fusing fault assessment data based on discrete abnormality scores during information fusion, and solves the problem of obtaining key part fault assessment scores, tracing the source of the fault and constructing the fault propagation path through time series analysis during fault diagnosis.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A water pump fault diagnosis system, comprising:

[0008] By deploying sensors at key locations of the water supply pump and adopting a synchronous acquisition strategy to collect real-time operating data of the water supply pump, an operation database is constructed. When new data records are added to the operation database, the operating data collected by each sensor is tested, and an adaptive judgment strategy is used to determine whether to issue an abnormality warning for the water supply pump.

[0009] When a water pump abnormality warning is detected, the fault assessment set for fault identification is screened based on the operation database, and the fault assessment data is fused. The fault assessment set is classified, summarized and fused to obtain the fault assessment feature set.

[0010] Fault diagnosis of the water supply pump is performed based on the fault assessment feature set. A comprehensive time series analysis is performed on the preliminary fusion data of each fault assessment subset in the fault assessment feature set using a time series analysis strategy. The fault assessment scores of key parts are obtained to identify the faults. For the key parts judged to be faulty, the fault source is obtained by constructing a time axis, and the fault propagation path is constructed in combination with the correlation coefficient matrix.

[0011] Specifically, the synchronous acquisition strategy is used to simultaneously obtain data monitored by different types of sensors at key parts of the water supply pump. By setting a unified time trigger mechanism and sampling frequency to coordinate the work of each sensor, the water supply pump operation data is collected to build a dynamically updated operation database.

[0012] The adaptive judgment strategy is used to perform data detection on the operating data collected by each sensor when new data records are added, including missing value detection, sensor anomaly detection, and deviation point detection. Adaptive judgment is performed based on the detection results to determine the early warning decision.

[0013] Specifically, the steps of the adaptive decision strategy include:

[0014] When running a new data record in the database, determine whether there are missing values ​​in the new data record;

[0015] If there are missing values, configure the missing evaluation interval and obtain the sensor operation data with missing values ​​within the missing evaluation interval before the time point of the new data recording to form the missing filling data;

[0016] Configure the fluctuation threshold and use the standard deviation of the missing value filling data as its fluctuation degree. If the fluctuation degree is less than the fluctuation threshold, the mean filling method is used, otherwise the spline interpolation method is used;

[0017] If there are no missing values ​​in the newly added data records, the newly added data records are judged as outliers;

[0018] Set the standard value range for the operating data collected by each sensor, and determine whether there is sensor operating data that exceeds the standard value range in the newly added data records, that is, sensor abnormal points.

[0019] Specifically, the adaptive decision strategy also includes:

[0020] If there is a sensor abnormal point, perform sensor abnormal point monitoring;

[0021] Configure the sensor monitoring interval to monitor the sensor operation data of the sensor abnormal point within the sensor monitoring interval after the time point of the newly added data recording to form abnormal monitoring data;

[0022] Configure anomaly monitoring thresholds and count the number of sensor anomalies in the anomaly monitoring data. If the number is greater than the anomaly monitoring threshold, a sensor fault warning is issued. Otherwise, the sensor anomalies in the newly added data records are treated as missing values ​​and filled in.

[0023] Setting deviation assessment thresholds If there is no sensor abnormal point, then based on the historical operation data collected by each sensor, set the deviation judgment parameters of each sensor, including the historical mean and historical standard deviation, and add data records that deviate from the historical mean by more than The sensor operation data with a value greater than the historical standard deviation is marked as a deviation point;

[0024] Configure the deviation point monitoring interval to monitor the data records within the deviation point monitoring interval after the time point of the newly added data record to form a deviation point monitoring data set;

[0025] Configure the deviation abnormality threshold and count the number of deviation points in the deviation point monitoring data set. If the deviation abnormality threshold is greater than the deviation abnormality threshold, a water pump abnormality warning is issued; otherwise, no action is taken.

[0026] Specifically, the information fusion module includes a data screening unit and a feature fusion unit;

[0027] The multi-level screening strategy is configured in the data screening unit, and is used to obtain deviating abnormal sensors when the water supply pump is abnormally warned, and to calculate a discrete abnormal score according to position coordinates of the deviating abnormal sensors, and to construct a corresponding fault evaluation set according to a size of the discrete abnormal score.

[0028] The association fusion strategy is configured in the feature fusion unit, and is used to group the fault evaluation set according to key parts to which the sensors belong to obtain fault evaluation subsets of the key parts, and to perform data fusion on the fault evaluation subsets according to sensor types by using a maximum deviation method to construct a fault evaluation feature set.

[0029] Specifically, the steps of the multi-level screening strategy include:

[0030] When the water supply pump is abnormally warned, a sensor in which a deviating point in the deviating point monitoring data set is located is obtained, and is marked as a deviating abnormal sensor;

[0031] According to position coordinates of the deviating abnormal sensor, distances between each deviating abnormal sensor and the remaining deviating abnormal sensors are calculated, and a discrete abnormal score is obtained in combination with a volume of the water supply pump;

[0032] A discrete abnormal threshold value is configured, if the discrete abnormal score is less than the discrete abnormal threshold value, center coordinates of a region in which all the deviating abnormal sensors are located are obtained, distances from all the deviating abnormal sensors to the center coordinates are calculated, a maximum distance is selected as a spherical radius, and a running deviating region of the water supply pump is divided with the center coordinates as a spherical center;

[0033] A fault evaluation interval is configured All the sensors in the running deviating region are marked as deviating abnormal sensors, and running data within the fault evaluation interval before a time point at which the water supply pump is abnormally warned is obtained to construct a fault evaluation set.

[0034] Specifically, the steps of the multi-level screening strategy further include:

[0035] If the discrete abnormal score is greater than or equal to the discrete abnormal threshold value, it indicates that the deviating abnormal sensors are relatively dispersed in the water supply pump, and the deviating region is not concentrated, and then the fault evaluation interval is adjusted according to a number of the deviating abnormal sensors, that is:

[0036] If a ratio of the number of the deviating abnormal sensors to a total number of the sensors is greater than a set proportion threshold value, the fault evaluation interval is increased, and an increasing degree of the fault evaluation interval is related to a difference between the ratio of the number of the deviating abnormal sensors and the proportion threshold value;

[0037] If the ratio of the number of deviating abnormal sensors to the total number of sensors is less than or equal to the proportion threshold, the fault evaluation interval is reduced, the degree of reduction of the fault evaluation interval is related to the difference between the proportion threshold and the ratio of the number of deviating abnormal sensors, and the fault evaluation interval is regulated by a maximum function to be not less than a minimum fault evaluation interval;

[0038] According to the adjusted fault evaluation interval, the running data of all sensors in the feed water pump within the time interval before the abnormal early warning time point and the time interval is adjusted The fault evaluation set is constructed.

[0039] Specifically, the specific steps of the association fusion strategy include:

[0040] According to the key parts to which the sensors belong, the fault evaluation set is grouped to obtain a fault evaluation subset of each key part;

[0041] Set the sensor collection mean value, use the maximum deviation method, and according to the sensor type, preliminarily fuse the fault evaluation subset, and for the same sensor type, select the running data deviating from the sensor collection mean value the most at each collection time point as the preliminary fusion data of the fault evaluation subset;

[0042] Based on the preliminary fusion data of each fault evaluation subset, a fault evaluation feature set for fault diagnosis of the feed water pump is constructed.

[0043] Specifically, the time series analysis strategy is used for comprehensive time series analysis of the preliminary fusion data of each fault evaluation subset in the fault evaluation feature set to obtain a fault evaluation score of the key part, and the key part is fault judged according to the fault evaluation score;

[0044] The association analysis strategy is used to analyze the key parts judged to be faulty to trace the fault source, identify the fault starting time point, determine the fault source, and obtain the fault propagation relationship according to the correlation coefficient matrix of each pair of potential fault parts, and construct the fault propagation path.

[0045] Specifically, the steps of the time series analysis strategy include:

[0046] In the fault evaluation subset, for the preliminary fusion data of each sensor type, the fluctuation level is obtained through the standard deviation of the preliminary fusion data, and a time series prediction model is selected, including an autoregressive moving average model and a weighted moving average model;

[0047] Configure a prediction threshold, divide the preliminary fusion data of each sensor type in the fault evaluation subset into a training set and a prediction set according to the prediction threshold, train and fit the time series prediction model through the training set, obtain the fitted time series prediction model, and predict the prediction fusion data in the future prediction threshold range according to the training set.

[0048] calculate the prediction relative error of each prediction fusion data and the prediction of the preliminary fusion data of the prediction set, configure an error threshold, if the prediction relative error of the prediction fusion data and the prediction of the preliminary fusion data of the prediction set is greater than the error threshold, mark it as a prediction anomaly point, otherwise do not perform any processing;

[0049] count the number of prediction anomaly points in each sensor type in the fault evaluation subset, and record the time point of the prediction anomaly point, combine the number of prediction anomaly points, the prediction relative error of the prediction anomaly point and the duration length of the prediction anomaly point to obtain the fault evaluation score of the key part;

[0050] configure a score threshold, if the fault evaluation score of the key part where the fault evaluation subset is located is greater than the score threshold, mark the key part as a potential fault part, otherwise do not perform any processing.

[0051] The beneficial effects of the present application are:

[0052] 1. The synchronous acquisition strategy ensures that the multi-source data is consistent in the time dimension, providing complete original data for comprehensive analysis. At the same time, the adaptive judgment strategy can effectively detect and process missing values, abnormal points and deviation points, ensuring data quality, early warning of equipment failure, enhancing system adaptability and reliability, and ensuring stable and efficient operation of the water supply pump.

[0053] 2. According to the sensor distribution, the data is reasonably selected, the analysis range is narrowed, and the fault diagnosis efficiency is improved. The correlation fusion strategy of the feature fusion unit groups and fuses the data according to the key parts and sensor types, and sorts out the data correlation, providing strong support for accurate fault diagnosis and fault location.

[0054] 3. By performing time series analysis on the preliminary fusion data in the fault evaluation feature set, the fault evaluation score is obtained to distinguish faults. With the help of constructing a time axis, tracing the fault source from the starting time point, and constructing a fault propagation path according to the correlation coefficient matrix, the fault can be analyzed comprehensively and deeply, and the fault root cause and propagation path can be accurately found, so that targeted measures can be taken in time to ensure the normal operation of the water supply pump. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 It is a structure schematic view of a water supply pump fault diagnosis system of the present application;

[0056] Figure 2 It is a flowchart of the specific steps of the adaptive judgment strategy of the present application;

[0057] Figure 3 It is a flowchart of the deviation point judgment of the newly added data record of the present application;

[0058] Figure 4 A flowchart of the specific steps of the multi-stage screening strategy of the present invention;

[0059] Figure 5 Flowchart of the specific steps of the timing analysis strategy of the present invention. DETAILED DESCRIPTION

[0060] See also Figure 1 ,This embodiment introduces a water pump fault diagnosis system, including a data acquisition module, an information fusion module and a fault diagnosis module;

[0061] The data acquisition module is used to deploy sensors at key locations of the water supply pump, collect real-time operating data of the water supply pump through a synchronous acquisition strategy, build an operating database, and when new data records are added to the operating database, detect the operating data collected by each sensor, and use an adaptive judgment strategy to determine whether to issue an abnormality warning for the water supply pump; key locations include: pump body, impeller, water inlet, water outlet, motor winding and pump casing, and sensors include: acceleration sensor, pressure sensor, temperature sensor and sound sensor.

[0062] Preferably, the data acquisition module includes a sensor acquisition unit and a data preprocessing unit;

[0063] The sensor acquisition unit is equipped with a synchronous acquisition strategy, which is used to simultaneously acquire data monitored by different types of sensors at key parts of the water supply pump, ensuring the consistency of each data in the time dimension. By setting a unified time trigger mechanism and sampling frequency to coordinate the work of each sensor, the water supply pump operation data is collected to build a dynamically updated operation database; it realizes the synchronous acquisition of real-time operation data of the water supply pump, such as acceleration, pressure, temperature, and sound, to provide complete raw data support for the subsequent comprehensive and accurate analysis of the water supply pump operation status;

[0064] The data preprocessing unit is equipped with an adaptive judgment strategy. The adaptive judgment strategy is used to perform data detection on the operating data collected by each sensor when new data records are added, including missing value detection, sensor abnormal point detection, and deviation point detection. Adaptive judgment is made based on the detection results to determine the early warning decision; it effectively guarantees data quality, warns of equipment failures in advance, enhances adaptability and reliability, and ensures stable and efficient operation of the water pump.

[0065] See also Figure 2 Preferably, the specific steps of the adaptive decision strategy include:

[0066] When a new data record is added to the running database, the new data record is recorded as ,in, It is Sensors at time The collected operating data, is the number of sensors in the feedwater pump;

[0067] Determine new data records Whether there are missing values, if so, configure the missing evaluation interval , obtain the sensor operation data with missing values ​​within the missing evaluation interval before the newly added data recording time point to form the missing filling data, that is, ,in, The sensor where the missing value is located at time point Collected operating data; by configuring the missing evaluation interval, locate and collect historical sensor operating data related to missing values, preparing for the subsequent selection of appropriate filling methods.

[0068] Configure the fluctuation threshold and perform fluctuation analysis on the missing-filled data. The standard deviation of the missing-filled data is used as its fluctuation degree. If the fluctuation degree is less than the fluctuation threshold, it means that the fluctuation range of the data collected by the sensor where the missing-filled data is located is small. In this case, the mean filling method is used, that is, the average value of the missing-filled data is used to fill the missing values. For missing-filled data with a small fluctuation range, the mean filling method can use the relatively stable mean of the data to fill the missing values, so that the filled data is consistent with the overall fluctuation characteristics of the original data, maintaining the stability of the data to the greatest extent, and reducing the interference of the filling operation on the statistical characteristics of the data.

[0069] If the degree of fluctuation is greater than or equal to the fluctuation threshold, it means that the data collected by the sensor where the missing-filled data is located has a large fluctuation range. In this case, the spline interpolation method is used, that is, a smooth curve is constructed to fit the missing-filled data, and the missing value filling value is obtained through the curve; for missing-filled data with a large fluctuation range, the spline interpolation method obtains the missing value filling value by constructing a smooth curve to fit the data, and performs reasonable filling based on the original large fluctuation trend of the data, so that the filled value is more in line with the dynamic change law of the data, avoiding jumps or unreasonable fluctuations in the data that are inconsistent with the actual situation due to simple filling.

[0070] If you add new data records If there are no missing values, the newly added data records will be judged as outliers. The standard value range of the operating data collected by each sensor will be set to determine whether there is sensor operating data that exceeds the standard value range in the newly added data records, that is, sensor anomalies. If there are sensor anomalies, sensor anomaly monitoring will be performed; based on the set standard value range of each sensor operating data, sensor anomalies that exceed the normal range in the newly added data records will be screened out, and the data parts that may have problems will be quickly locked.

[0071] Configuring sensor monitoring intervals , monitor the sensor operation data of the sensor abnormal point within the sensor monitoring interval after the new data recording time point to form the abnormal monitoring data, that is, ,in, The sensor where the abnormal point is located at the time point The collected operating data provides more sufficient and detailed data support for accurately judging whether the sensor is faulty, avoiding hasty judgment of sensor failure due to a single abnormal data, and improving the accuracy of fault judgment.

[0072] Configure the abnormal monitoring threshold and count the number of sensor abnormal points in the abnormal monitoring data. If it is greater than the abnormal monitoring threshold, a sensor fault warning will be issued, indicating that there are many operating data that exceed the standard value range of the sensor and there is a sensor equipment failure, so as to remind maintenance personnel to check, repair or replace the sensor; otherwise, the sensor abnormal points in the newly added data records will be treated as missing values ​​and missing values ​​will be filled; so that the data collection anomalies caused by accidental sensor equipment can be integrated into the overall data through filling, ensuring the integrity of the data and the consistency of the subsequent analysis process.

[0073] See also Figure 3 , set the deviation assessment threshold , The larger the value, the more tolerant it is to fluctuations in operating data. If there is no sensor abnormality, the deviation point judgment is performed on the newly added data records. Based on the historical operating data collected by each sensor, the deviation judgment parameters of each sensor are set, including the historical mean and historical standard deviation. The newly added data records that deviate from the historical mean by more than Sensor operation data with a deviation of times the historical standard deviation is marked as a deviation point; deviation judgment parameters are set based on the historical operation data collected by each sensor to mark the deviation points in the newly added data records, which can mine data that deviates significantly from the historical normal state;

[0074] Configuring the deviation point monitoring interval , monitor the data records within the deviation point monitoring interval after the time point of the newly added data record to form the deviation point monitoring data set, that is, ,in, It's at the time Newly added data records of operating data; in order to more accurately judge whether the deviation is an accidental fluctuation or actually reflects the existence of substantial problems such as operating deviation in the water supply pump, and enhance the reliability of the judgment of the overall operating stability of the water supply pump.

[0075] Configure a deviation anomaly threshold and count the number of deviation points in the deviation point monitoring data set. If the deviation anomaly threshold is greater than the threshold, a water pump anomaly warning is issued, indicating that a large number of sensors have collected deviation points within the deviation point monitoring interval, indicating that there is an operational deviation in the water pump and fault diagnosis is required. Otherwise, no action is taken. This avoids frequent false alarms due to normal small fluctuations in data, reduces unnecessary resource waste and interference with normal work processes, and ensures the accuracy and efficiency of the entire warning mechanism.

[0076] The information fusion module is used to screen the fault assessment set for fault identification based on the operating database when an abnormal warning of the water pump is detected, and to perform information fusion on the fault assessment data. The fault assessment set is classified, summarized and fused to obtain a fault assessment feature set. The module also screens out data related to fault identification from a large amount of data, and extracts a fault assessment feature set through a series of operations such as information fusion and classification and summary processing, which provides a key, integrated feature basis for subsequent accurate fault diagnosis and helps to accurately locate the fault situation.

[0077] Preferably, the information fusion module includes a data screening unit and a feature fusion unit;

[0078] A multi-level screening strategy is configured in the data screening unit. The multi-level screening strategy is used to obtain the deviation abnormal sensor when the water pump abnormality warning is triggered, and calculate the distance based on its position coordinates to obtain a discrete abnormality score. According to the size of the discrete abnormality score, a corresponding fault assessment set is constructed; after the water pump abnormality warning is triggered, the deviation abnormal sensor is locked, and the discrete abnormality score is obtained by calculating the distance with the help of the position coordinates. According to the size of the score, a fault assessment set is constructed. This method can reasonably screen out the most relevant and reference-worthy data according to the sensor distribution and other conditions, narrow the analysis scope, make subsequent fault assessments more targeted, and improve fault diagnosis efficiency.

[0079] The feature fusion unit is configured with a correlation fusion strategy. This strategy groups the fault assessment set by the key sensor location to obtain fault assessment subsets for each key location. The maximum deviation method is then used to fuse the data of each fault assessment subset based on sensor type to construct a fault assessment feature set. Grouping sensors by key location spatially distinguishes data from different locations. The maximum deviation method is then used to fuse the data of each subset based on sensor type, effectively clarifying the correlations between data from different locations and integrating and refining the dispersed data. This fused data not only reflects the characteristics of each key location but also comprehensively reflects the key information of each sensor type, providing strong data support for more accurate fault diagnosis and fault location.

[0080] See also Figure 4Preferably, the specific steps of the multi-stage screening strategy include:

[0081] When a water pump anomaly warning is issued, the sensor where the deviation point is located in the deviation point monitoring data set is obtained and marked as the deviation abnormal sensor. This locks in the key focus for subsequent in-depth analysis of the fault cause and location of the fault area, narrowing the scope of investigation.

[0082] According to the position coordinates of the deviation abnormal sensor, the distance between each deviation abnormal sensor and the other deviation abnormal sensors is calculated to evaluate the deviation degree and obtain the discrete abnormality score, that is:

[0083]

[0084] in, is the discrete anomaly score, It is and The distance from the abnormal sensor, It is the volume of the water supply pump; it intuitively reflects whether the deviation abnormal sensors are relatively concentrated or dispersed as a whole, providing a data basis for further judging the specific characteristics of the water supply pump operation deviation, and helping to more accurately grasp the scope and degree of the possible fault.

[0085] Configure a discrete anomaly threshold. If the discrete anomaly score is less than the discrete anomaly threshold, it indicates that the deviation sensors are relatively concentrated and the area where the water supply pump deviates is relatively concentrated. Then obtain the center coordinates of the area where all deviation sensors are located, and calculate the distance from all deviation sensors to the center coordinates. Select the maximum distance as the radius of the sphere, and divide the water supply pump's operation deviation area with the center coordinates as the center of the sphere. Further narrow the troubleshooting scope to the designated operation deviation area, making subsequent data collection, analysis, and other work more targeted and avoiding interference from irrelevant data.

[0086] Configuring the Fault Evaluation Interval , obtain all sensors in the operation deviation area, mark them as deviation abnormal sensors, and construct the fault evaluation set based on the operation data in the fault evaluation interval before the abnormal warning time point of the water pump, that is, ,in, It's time The data record of the deviation abnormal sensor, that is, ,in, It is the first The deviation abnormal sensor at time point The collected operating data, is the number of deviation abnormal sensors in the operation deviation area;

[0087] If the discrete anomaly score is greater than or equal to the discrete anomaly threshold, it indicates that the deviation abnormal sensors are relatively dispersed in the water pump and the deviation area is not concentrated. Then, the fault assessment interval is adjusted according to the number of deviation abnormal sensors, that is:

[0088]

[0089] in, is the adjusted fault assessment interval, is the number of sensors that deviate from the anomaly, is the number of sensors in the feedwater pump, is the specific gravity threshold. The larger the specific gravity threshold, the greater the acceptance of the sensor deviation from the abnormality. is the minimum fault assessment interval, is the weighting coefficient used to control the magnitude of the adjustment, obtained through experiments. It is a maximum function; it dynamically expands or reduces the time range of data collection according to the number of abnormal sensors.

[0090] Based on the adjusted fault assessment interval , obtain all sensors in the water supply pump, before the abnormal warning time point of the water supply pump The operating data in is used to construct the fault evaluation set, i.e. ,in, It's time Data records of all sensors in the water pump, namely ,in, It is the first The deviation abnormal sensor at time point Collected operating data.

[0091] Preferably, the specific steps of associating the fusion strategy include:

[0092] The fault assessment set is grouped according to the key location to which the sensor belongs, obtaining a fault assessment subset for each key location. For example, the fault assessment data corresponding to all sensors in the pump body are grouped together, those in the impeller area are grouped together, and so on, ultimately obtaining a fault assessment subset corresponding to each key location. Spatially distinguishing data from different locations facilitates subsequent targeted fusion and analysis based on the characteristics of each location. This also aligns with the structural characteristics of the water feed pump, helping to clarify the correlations between data from different locations and providing clearer clues for fault location and diagnosis.

[0093] Set the sensor collection mean, use the maximum deviation method, and evaluate each fault subset according to the sensor type. Perform data fusion, where It is a fault evaluation subset the a sensor type in the operation data collected by the the a sensor type in the a sensor type in the the the number of sensor types in the fault evaluation subset, and for the same sensor type, the operation data with the maximum deviation from the average value of sensor collection at each collection time point is selected as the preliminary fusion data of the fault evaluation subset, that is:

[0094]

[0095] wherein, a fault evaluation subset a sensor type in the the preliminary fusion data of the a sensor type in the the operation data of the a sensor type in the a sensor type in the the operation data of the a sensor type in the the operation data of the a sensor type in the a sensor type in the the operation data of the a sensor type in the the average value of sensor collection of the a sensor type in the

[0096] obtaining the preliminary fusion data of each fault evaluation subset , that is wherein, a fault evaluation subset a sensor type in the the preliminary fusion data of the

[0097] The fault diagnosis module is used for fault diagnosis of the water supply pump according to the fault evaluation feature set, and the key part fault evaluation score is obtained through comprehensive time sequence analysis on the preliminary fusion data of each fault evaluation subset in the fault evaluation feature set to judge the fault. For the key part judged as the fault, the fault source is obtained through the construction of the time axis, and the fault propagation path is constructed in combination with the correlation coefficient matrix.

[0098] ​Preferably, the fault diagnosis module includes a local diagnosis unit and a global diagnosis unit;

[0099] A time series analysis strategy is configured within the local diagnosis unit. This strategy is used to perform a comprehensive time series analysis on the preliminary fusion data of each fault assessment subset in the fault assessment feature set to obtain the fault assessment scores of key parts and to identify faults in key parts based on the fault assessment scores.

[0100] A correlation analysis strategy is configured within the global diagnostic unit. This strategy is used to conduct correlation analysis on key fault locations to trace the source of the fault. By identifying the fault start time and source, the global diagnostic unit obtains the fault propagation relationship and constructs the fault propagation path based on the correlation coefficient matrix of each pair of potential fault locations.

[0101] See also Figure 5 , preferably, the specific steps of the timing analysis strategy include:

[0102] In the fault assessment subset, preliminary fusion data for each sensor type ,in, It is The fault assessment subset is in The sensor type in The preliminary fusion data of the time points, It is used to indicate the number of preliminary fused data, obtain its fluctuation level through the standard deviation of the preliminary fused data, and select the time series prediction model, including the autoregressive moving average model and the weighted moving average model;

[0103] Configure the stability threshold. If the standard deviation of the initial fusion data is greater than the stability threshold, it indicates that the data collected by this sensor type is highly volatile, contains obvious trends, periodic changes, or is subject to large random interference. For example, during the start-up or shutdown phase of a water pump, or in the early stages of a failure, the data of the acceleration sensor and pressure sensor may fluctuate significantly. At this time, the data changes greatly, the differences between adjacent data points are obvious, and there may be complex change patterns. In this case, the autoregressive moving average model is selected; otherwise, the data fluctuations are relatively small and relatively stable, so the weighted moving average model is selected.

[0104] For the autoregressive moving average model, the autoregressive order and moving average order are obtained through autocorrelation analysis and partial autocorrelation analysis. For the weighted moving average model, the moving average window width is set. The larger the window width, the more obvious the smoothing effect, but the slower the response to data changes; the smaller the window width, the faster it can reflect recent data changes, but the smoothing effect will be weakened.

[0105] Configuring prediction thresholds , the preliminary fusion data of each sensor type in the fault assessment subset is combined according to the prediction threshold Divide into training set and prediction set ,Right now , ;

[0106] Through the training set, the time series prediction model is trained and fitted, the fitted time series prediction model is obtained, and the future prediction threshold range is predicted according to the training set. , ]’s predicted fusion data;

[0107] Calculate the relative error of the prediction of each predicted fusion data and the preliminary fusion data of the prediction set, that is:

[0108]

[0109] in, It is The fault assessment subset is in sensor types at time points The relative error of the prediction, It is The fault assessment subset is in sensor types at time points The prediction fusion data of

[0110] Configure the error threshold. If the relative error between the predicted fusion data and the preliminary fusion data of the prediction set is greater than the error threshold, it will be marked as a prediction anomaly. Otherwise, no processing will be performed.

[0111] The number of predicted abnormal points for each sensor type in the fault assessment subset is counted, and the time point of the predicted abnormal point is recorded. The fault assessment score of the key parts is obtained by combining the number of predicted abnormal points, the relative error of the predicted abnormal point, and the duration of the predicted abnormal point, namely:

[0112]

[0113] in, It is The fault assessment scores of the key parts of the fault assessment subsets are: It is In the fault assessment subset The number of predicted anomalies in each sensor type, It is The fault assessment subset is in Sensor type The relative error of the prediction of the outlier point, It is The fault assessment subset is in The time point where the initial predicted abnormal point is located in each sensor type, 、 and are all non-negative weight coefficients;

[0114] Configure a scoring threshold. If the fault assessment score of the key part of the fault assessment subset is greater than the scoring threshold, the key part is marked as a potential fault part. Otherwise, no processing is performed.

[0115] Preferably, the specific steps of the association analysis strategy include:

[0116] Construct a timeline and mark the predicted abnormal points of each potential fault location on the timeline to intuitively display the sequence of events related to faults in different locations;

[0117] For each potential fault location, analyze the timeline variation characteristics of its preliminary fused data and find the earliest data point with obvious anomalies as the fault start time for that location. For example, the time point when a predicted anomaly first appears in a potential fault location is the fault start time, or the time point when the largest number of predicted anomalies first appear simultaneously in the sensors at the potential fault location is the fault start time.

[0118] Compare the fault onset time points determined for each potential fault location, analyze the order of fault onset times for different locations, and identify the key location where the earliest fault signs appear as the source of fault propagation;

[0119] For each pair of potential fault locations, the correlation coefficient matrix of the preliminary fusion data from the fault start time to the current time is calculated. If the correlation coefficients are high, it means that the data changes of the two locations have a strong correlation and there is a fault propagation relationship.

[0120] The largest correlation coefficient in the correlation coefficient matrix of each pair of potential fault locations is selected as the decision parameter of the fault propagation path. According to the fault starting time point of each potential fault location, starting from the source of fault propagation, the potential fault locations with fault propagation relationship are sorted according to the size of the decision parameter to construct the fault propagation path.

[0121] Working principle and its effect:

[0122] A water pump fault diagnosis system deploys acceleration sensors, pressure sensors, temperature sensors, and sound sensors in key locations, including the pump body, impeller, water inlet, water outlet, motor windings, and pump casing. A synchronized data collection strategy, along with a unified time trigger mechanism and sampling frequency, ensures consistent data collection across time from each sensor. This dynamically updated operational database provides complete raw data support for subsequent analysis, ensuring data consistency and comprehensiveness.

[0123] The data acquisition module's adaptive judgment strategy detects missing values, sensor anomalies, and deviations when new data records are added to the running database. Missing values ​​are processed using mean imputation or spline interpolation based on fluctuation thresholds to ensure data stability and conformity to dynamic changes, thereby guaranteeing data quality. For sensor anomalies and deviations, the module accurately determines sensor failures and water pump operation deviations by setting monitoring intervals, anomaly monitoring thresholds, and deviation assessment thresholds. This provides early warning of equipment failures, reduces false alarms, avoids hasty fault judgments based on single abnormal data, and enhances judgment accuracy and reliability.

[0124] When a water pump anomaly warning is detected, the information fusion module activates. Its multi-level screening strategy identifies sensors with abnormal deviations, calculates discrete anomaly scores, and constructs a fault assessment set based on these scores, narrowing the analysis scope and improving fault diagnosis efficiency. The correlation fusion strategy groups sensors by key locations, fuses data using the maximum deviation method, constructs a fault assessment feature set, and analyzes data associations, providing strong data support for precise fault location.

[0125] The fault diagnosis module works based on the fault assessment feature set. The time series analysis strategy selects an autoregressive moving average model or a weighted moving average model based on the standard deviation of the preliminary fusion data for each sensor type. After training fitting, prediction, error calculation, and statistical prediction of the number of abnormal points, the weight coefficient is combined to obtain the fault assessment score of the key parts and accurately identify the fault. The correlation analysis strategy constructs a timeline, compares the fault starting time points of each potential fault part, and finds the key part where the earliest fault signs appear as the source of the fault; by calculating the correlation coefficient matrix of the preliminary fusion data between each potential fault part, the fault propagation path is constructed, and the fault source and propagation path are accurately located, providing clear guidance for repair and maintenance, ensuring the stable and efficient operation of the water supply pump.

[0126] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the concept of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A water pump fault diagnosis system, characterized in that: include: By deploying sensors at key locations of the water supply pump and adopting a synchronous acquisition strategy to collect real-time operating data of the water supply pump, an operation database is constructed. When new data records are added to the operation database, the operating data collected by each sensor is tested, and an adaptive judgment strategy is used to determine whether to issue an abnormality warning for the water supply pump. When a water pump abnormality warning is detected, the fault assessment set for fault identification is screened based on the operation database, and the fault assessment data is fused. The fault assessment set is classified, summarized and fused to obtain the fault assessment feature set. Fault diagnosis of the water supply pump is performed based on the fault assessment feature set. A comprehensive time series analysis is performed on the preliminary fused data of each fault assessment subset in the fault assessment feature set using a time series analysis strategy. Fault assessment scores for key parts are obtained to identify the fault. For key parts identified as faulty, a time axis is constructed using a correlation analysis strategy to obtain the fault source, and the fault propagation path is constructed using a correlation coefficient matrix. A water pump fault diagnosis system further includes an information fusion module, wherein the information fusion module includes a data screening unit and a feature fusion unit; The data screening unit is configured with a multi-level screening strategy, which is used to obtain the deviation abnormality sensor when the water pump is abnormally warned, and calculate the distance based on its position coordinates to obtain a discrete abnormality score, and construct a corresponding fault assessment set based on the size of the discrete abnormality score; The feature fusion unit is configured with an associated fusion strategy, which is used to group the fault assessment set according to the key parts of the sensor to obtain fault assessment subsets of each key part, and use the maximum deviation method to fuse the data of each fault assessment subset according to the sensor type to construct a fault assessment feature set; The steps of the multi-stage screening strategy include: When the water pump is abnormally warned, the sensor where the deviation point is located in the deviation point monitoring data set is obtained and marked as the deviation abnormal sensor; According to the position coordinates of the deviation abnormal sensor, the distance between each deviation abnormal sensor and the other deviation abnormal sensors is calculated, and the discrete abnormality score is obtained by combining it with the volume of the water pump; Configure a discrete anomaly threshold. If the discrete anomaly score is less than the discrete anomaly threshold, obtain the center coordinates of the area where all deviation anomaly sensors are located, calculate the distance from all deviation anomaly sensors to the center coordinates, select the maximum distance as the radius of a sphere, and divide the water pump's operation deviation area with the center coordinates as the center of the sphere. Configuring the Fault Evaluation Interval , obtain all sensors in the operation deviation area, mark them as deviation abnormal sensors, and construct a fault assessment set based on the operation data in the fault assessment interval before the abnormal warning time point of the water supply pump; The specific steps of the association fusion strategy include: The fault assessment set is grouped according to the key parts to which the sensor belongs, and the fault assessment subset of each key part is obtained; Set the sensor acquisition mean and use the maximum deviation method to perform preliminary data fusion on the fault assessment subset based on the sensor type. For the same sensor type, select the operating data with the largest deviation from the sensor acquisition mean at each acquisition time point as the preliminary fusion data for the fault assessment subset. Based on the preliminary fusion data of each fault assessment subset, a fault assessment feature set for feedwater pump fault diagnosis is constructed; The steps of the timing analysis strategy include: In the fault assessment subset, for each sensor type's preliminary fusion data, the standard deviation of the preliminary fusion data is used to obtain its fluctuation level, and a time series prediction model is selected, including the autoregressive moving average model and the weighted moving average model. Configure a prediction threshold and divide the preliminary fusion data of each sensor type in the fault assessment subset into a training set and a prediction set based on the prediction threshold. Use the training set to train and fit the time series prediction model, obtain the fitted time series prediction model, and predict future prediction fusion data within the prediction threshold range based on the training set. Calculate the prediction relative error between each predicted fusion data and the preliminary fusion data of the prediction set, configure the error threshold, and if the prediction relative error between the predicted fusion data and the preliminary fusion data of the prediction set is greater than the error threshold, mark it as a prediction anomaly, otherwise no processing is performed; Count the number of predicted abnormal points for each sensor type in the fault assessment subset and record the time point of the predicted abnormal points. Combine the number of predicted abnormal points, the relative error of the predicted abnormal points, and the duration of the predicted abnormal points to obtain the fault assessment score of the key parts. Configure a scoring threshold. If the fault assessment score of the key part of the fault assessment subset is greater than the scoring threshold, the key part is marked as a potential fault part. Otherwise, no processing is performed.

2. A water pump fault diagnosis system according to claim 1, characterized in that: The synchronous acquisition strategy is used to simultaneously acquire data monitored by different types of sensors at key parts of the water supply pump. By setting a unified time trigger mechanism and sampling frequency to coordinate the work of each sensor, the water supply pump operation data is collected to build a dynamically updated operation database. The adaptive judgment strategy is used to perform data detection on the operating data collected by each sensor when new data records are added, including missing value detection, sensor abnormal point detection, and deviation point detection, and to perform adaptive judgment based on the detection results to determine the early warning decision.

3. A water pump fault diagnosis system according to claim 2, characterized in that: The steps of the adaptive decision strategy include: When running a new data record in the database, determine whether there are missing values ​​in the new data record; If there are missing values, configure the missing evaluation interval and obtain the sensor operation data with missing values ​​within the missing evaluation interval before the time point of the new data recording to form the missing filling data; Configure the fluctuation threshold and use the standard deviation of the missing value filling data as its fluctuation degree. If the fluctuation degree is less than the fluctuation threshold, the mean filling method is used, otherwise the spline interpolation method is used; If there are no missing values ​​in the newly added data records, the newly added data records are judged as outliers; Set the standard value range for the operating data collected by each sensor, and determine whether there is sensor operating data that exceeds the standard value range in the newly added data records, that is, sensor abnormal points.

4. A water pump fault diagnosis system according to claim 3, characterized in that: The adaptive decision strategy also includes: If there is a sensor abnormal point, perform sensor abnormal point monitoring; Configure the sensor monitoring interval to monitor the sensor operation data of the sensor abnormal point within the sensor monitoring interval after the new data recording time point to form abnormal monitoring data; Configure anomaly monitoring thresholds and count the number of sensor anomalies in the anomaly monitoring data. If the number is greater than the anomaly monitoring threshold, a sensor fault warning is issued. Otherwise, the sensor anomalies in the newly added data records are treated as missing values ​​and filled in. Setting deviation assessment thresholds If there is no sensor abnormal point, then based on the historical operation data collected by each sensor, set the deviation judgment parameters of each sensor, including the historical mean and historical standard deviation, and add data records that deviate from the historical mean by more than The sensor operation data with a value greater than the historical standard deviation is marked as a deviation point; Configure the deviation point monitoring interval to monitor the data records within the deviation point monitoring interval after the time point of the newly added data record to form a deviation point monitoring data set; Configure the deviation abnormality threshold and count the number of deviation points in the deviation point monitoring data set. If the deviation abnormality threshold is greater than the deviation abnormality threshold, a water pump abnormality warning is issued; otherwise, no action is taken.

5. A water pump fault diagnosis system according to claim 1, characterized in that: The steps of the multi-stage screening strategy also include: If the discrete anomaly score is greater than or equal to the discrete anomaly threshold, the fault assessment interval is adjusted according to the number of sensors that deviate from the anomaly, that is: If the ratio of the number of abnormal deviation sensors to the total number of sensors is greater than the set proportion threshold, the fault assessment interval is increased. The degree of increase in the fault assessment interval is related to the difference between the ratio of the number of abnormal deviation sensors and the proportion threshold. If the ratio of the number of abnormally deviated sensors to the total number of sensors is less than or equal to the weight threshold, the fault assessment interval is reduced. The degree of reduction in the fault assessment interval is related to the difference between the weight threshold and the ratio of the number of abnormally deviated sensors. The maximum value function is used to control the fault assessment interval to be no less than the minimum fault assessment interval. According to the adjusted fault assessment interval, the operating data of all sensors in the water supply pump before the abnormal warning time point and within the adjusted fault assessment interval are obtained to construct a fault assessment set.

6. A water pump fault diagnosis system according to claim 1, characterized in that: The time series analysis strategy is used to perform a comprehensive time series analysis on the preliminary fusion data of each fault assessment subset in the fault assessment feature set to obtain the fault assessment score of the key parts, and to perform fault discrimination on the key parts according to the fault assessment score; The correlation analysis strategy is used to perform correlation analysis on key parts judged to be faulty to trace the source of the fault. By identifying the fault starting time point and the source of the fault for the key parts judged to be faulty, the fault propagation relationship is obtained based on the correlation coefficient matrix of each pair of potential fault parts, and the fault propagation path is constructed.

Images