An online diagnosis method and system for vibration fault of energy storage hydroelectric generating unit

Abstract

The present application relates to the technical field of mechanical vibration, and discloses a kind of energy storage hydropower station unit vibration fault online diagnosis method and system, the method includes: acquisition unit under multiple operating conditions Original vibration data set;Eliminate the time-space deviation of different source signals by adjusting time frame and space reference system, obtain standardized vibration data matrix;Anti-interference fusion is carried out to standardized vibration data matrix, and dimension uniform operation is carried out to the data after anti-interference fusion, obtain multi-source anti-noise data cube;Multi-dimensional vibration fault feature coupling analysis is carried out to multi-source anti-noise data cube, obtain vibration fault feature topology set;Vibration fault feature topology set is matched with historical fault case base Feature mapping, obtain fault diagnosis result sequence;Based on real-time feedback data, fault diagnosis result sequence is dynamically corrected, and final fault diagnosis report is obtained;The present application can improve the accuracy of unit vibration fault online diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical vibration, in particular to a kind of energy storage hydropower unit vibration fault online diagnosis method and system. BACKGROUND

[0002] The prior art has obvious deficiencies in the collection and preprocessing of unit vibration data in the online diagnosis of energy storage hydropower unit vibration faults. Only single working condition or limited source vibration data is collected, multiple operating conditions are not covered, and effective anti-power frequency interference filtering processing and working condition grouping marking are lacking, resulting in a large number of interference signals mixed in the original data, the data under different working conditions cannot be clearly distinguished, and at the same time, time synchronization alignment is not performed, the time sequence correlation between data is chaotic, and it is difficult to form a data set that can accurately reflect the vibration state of the unit, which poses a hidden danger for subsequent diagnosis and analysis.

[0003] At the same time, the prior art has defects in data standardization, feature extraction and diagnosis result optimization. On the one hand, the time frame and spatial reference system are not adjusted to eliminate signal space deviation, and anti-interference fusion and dimensionless are not performed, the data format is chaotic, the noise interference is serious, and it is difficult to construct a high-quality analysis data basis; on the other hand, the extraction of vibration features lacks multi-dimensional coupling analysis, only single-dimensional identification features are extracted, it is difficult to accurately construct the fault feature topology relationship, and after fault matching, the diagnosis result is not dynamically corrected combined with real-time feedback data, which is easily affected by the limitations of historical cases, resulting in a large deviation between the diagnosis result and the actual fault state of the unit, and it is difficult to meet the high-precision and real-time fault diagnosis requirements. SUMMARY

[0004] The present application provides a kind of energy storage hydropower unit vibration fault online diagnosis method and system to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present application provides a kind of energy storage hydropower unit vibration fault online diagnosis method, which comprises:

[0006] S1, collecting the original vibration data set of the unit under multiple operating conditions;

[0007] S2, eliminate the space-time deviation of different source signals in the original vibration data set by adjusting the time frame and spatial reference system, and obtain the standardized vibration data matrix of the unit;

[0008] S3, anti-interference fusion is carried out on the standardized vibration data matrix, and dimensionless operation is carried out on the data after anti-interference fusion, to obtain the multi-source anti-noise data cube of the unit;

[0009] S4, multi-dimensional vibration fault feature coupling analysis is carried out on the multi-source anti-noise data cube, to obtain the vibration fault feature topology set of the multi-source anti-noise data cube;

[0010] S5, feature mapping matching the vibration fault feature topology set with a historical fault case library to obtain a fault diagnosis result sequence of the vibration fault feature topology set;

[0011] S6, dynamically correcting the fault diagnosis result sequence based on real-time feedback data to obtain a final fault diagnosis report of the unit.

[0012] In a preferred embodiment, the collected original vibration data set of the unit under multiple operating conditions includes:

[0013] Collecting multi-source vibration data of the unit;

[0014] Anti-power frequency interference filtering processing the multi-source vibration data to obtain pre-processed vibration data of the unit;

[0015] Grouping and marking the pre-processed vibration data according to the type of operating conditions to obtain a condition marked vibration data group of the unit;

[0016] Time synchronization alignment of the condition marked vibration data group to obtain the original vibration data set of the unit.

[0017] In a preferred embodiment, the elimination of the time and space deviations of different source signals in the original vibration data set by adjusting the time frame and the spatial reference system to obtain the standardized vibration data matrix of the unit includes:

[0018] Time frame alignment of the original vibration data set based on a preset time reference sequence to obtain a time synchronized signal set of the unit;

[0019] Establishing a unified spatial reference system based on the structural topology relationship of the unit, mapping the time synchronized signal set to the unified spatial reference system to obtain a spatial aligned signal set of the unit;

[0020] Amplitude normalization processing the spatial aligned signal set to obtain a standardized vibration data matrix of the unit.

[0021] In a preferred embodiment, the anti-interference fusion of the standardized vibration data matrix and the dimension unification operation on the anti-interference fused data to obtain a multi-source anti-noise data cube of the unit includes:

[0022] Fusion of multi-source signals in the standardized vibration data matrix to obtain an initial fusion data matrix of the unit;

[0023] Wavelet threshold denoising the initial fusion data matrix to obtain anti-interference fusion data of the unit;

[0024] dimensionally normalizing the anti-interference fusion data to obtain a normalized feature data set of the unit;

[0025] structuring and integrating the normalized feature data set according to feature dimensions to obtain a multi-source anti-noise data cube of the unit.

[0026] In a preferred embodiment, the multi-dimensional vibration fault feature coupling analysis on the multi-source anti-noise data cube obtains a vibration fault feature topology set of the multi-source anti-noise data cube, including:

[0027] extracting time-domain vibration features and frequency-domain vibration features in the multi-source anti-noise data cube, and performing feature integration on the time-domain vibration features and the frequency-domain vibration features to obtain a vibration feature vector set of the unit;

[0028] performing time-frequency domain correlation transformation on the vibration feature vector set to obtain a feature coupling relationship matrix of the multi-source anti-noise data cube;

[0029] performing topology structure analysis based on the feature coupling relationship matrix to obtain a vibration fault feature topology graph of the unit;

[0030] extracting topology connection relationships and node attributes in the vibration fault feature topology graph and collecting them as a vibration fault feature topology set of the multi-source anti-noise data cube.

[0031] In a preferred embodiment, the feature mapping matching of the vibration fault feature topology set with a historical fault case library obtains a fault diagnosis result sequence of the vibration fault feature topology set, including:

[0032] obtaining a fault feature topology structure in the historical fault case library;

[0033] extracting feature vectors of corresponding nodes in the vibration fault feature topology set and the fault feature topology structure;

[0034] based on the feature vectors, calculating a topology similarity of the vibration fault feature topology set and the fault feature topology structure;

[0035] according to a comparison result of the topology similarity and a preset threshold, screening out a matched fault feature topology structure to constitute a candidate fault set;

[0036] performing confidence evaluation on fault types in the candidate fault set to obtain a fault type confidence sequence of the candidate fault set;

[0037] generating the fault diagnosis result sequence according to the fault type confidence sequence.

[0038] In a preferred embodiment, the calculation formula of the topological similarity is as follows: ;

[0039] In the formula, is the topological similarity, is the total number of nodes in the feature topological structure, is the preset weight coefficient of the th node, is the node index, is the feature vector of the th node in the vibration fault feature topology set, is the feature vector of the th node in the fault feature topology structure, is a function of the similarity between two feature vectors.

[0040] In a preferred embodiment, the dynamic correction of the fault diagnosis result sequence based on real-time feedback data to obtain the final fault diagnosis report of the unit includes:

[0041] Obtaining real-time vibration signals in the operation process of the unit, and extracting time domain features and frequency domain features of the real-time vibration signals to construct real-time feature vectors of the unit;

[0042] Comparing the real-time feature vectors with a preset historical fault feature library to obtain a feature deviation value of the real-time feature vectors;

[0043] According to the feature deviation value, dynamically adjusting the fault type confidence in the fault diagnosis result sequence to obtain a corrected fault diagnosis result sequence;

[0044] Based on the corrected fault diagnosis result sequence, generating a final fault diagnosis report of the unit.

[0045] In a preferred embodiment, the dynamic adjustment of the fault type confidence in the fault diagnosis result sequence according to the feature deviation value to obtain a corrected fault diagnosis result sequence includes:

[0046] Obtaining an initial confidence of a fault type in the fault diagnosis result sequence and a corresponding feature deviation value;

[0047] According to the feature deviation value, determining a corresponding confidence adjustment strategy;

[0048] Based on the confidence adjustment strategy, dynamically correcting the initial confidence to obtain a target confidence of the unit;

[0049] According to the target confidence, the fault types in the fault diagnosis result sequence are reordered to obtain a revised fault diagnosis result sequence.

[0050] To solve the above problems, the application further provides an online vibration fault diagnosis system for energy storage hydropower units, comprising:

[0051] A multi-source vibration data acquisition module is configured to acquire original vibration data sets of the unit under multiple operating conditions.

[0052] A space-time normalization processing module is configured to eliminate the space-time deviation of different source signals in the original vibration data sets by adjusting the time frame and the space reference system, thereby obtaining a normalized vibration data matrix of the unit.

[0053] A multi-source anti-noise data cube establishing module is configured to perform anti-interference fusion on the normalized vibration data matrix, and perform dimension unification operation on the anti-interference fused data, thereby obtaining a multi-source anti-noise data cube of the unit.

[0054] A feature coupling analysis module is configured to perform multi-dimensional vibration fault feature coupling analysis on the multi-source anti-noise data cube, thereby obtaining a vibration fault feature topology set of the multi-source anti-noise data cube.

[0055] An intelligent diagnosis matching module is configured to perform feature mapping matching between the vibration fault feature topology set and a historical fault case library, thereby obtaining a fault diagnosis result sequence of the vibration fault feature topology set.

[0056] A dynamic optimization evaluation module is configured to perform dynamic correction on the fault diagnosis result sequence based on real-time feedback data, thereby obtaining a final fault diagnosis report of the unit.

[0057] Compared with the prior art, the application has the following beneficial effects:

[0058] 1. The application can obtain comprehensive and low-interference basic data by acquiring original vibration data sets of the unit under multiple operating conditions, performing anti-power frequency interference filtering, operating condition grouping marking and time synchronization alignment processing, eliminating signal space-time deviation by adjusting the time frame and the space reference system, and constructing a multi-source anti-noise data cube by combining anti-interference fusion and dimension unification operation, thereby accurately integrating multi-dimensional vibration data, effectively reducing the influence of noise and format difference on analysis, providing high-quality data support for subsequent fault feature extraction, and significantly improving the accuracy of vibration fault feature recognition.

[0059] 2.The method can construct a complete vibration fault feature topology set by performing multi-dimensional vibration fault feature coupling analysis on a multi-source anti-noise data cube, realize accurate mapping matching with a historical fault case library by combining topology similarity calculation, generate a fault diagnosis result sequence with confidence ranking, dynamically correct the diagnosis result based on real-time feedback data, timely adjust the fault type confidence and optimize the result ranking, ensure that the final diagnosis report is highly consistent with the actual fault state of the unit, and greatly improve the accuracy and real-time performance of fault diagnosis, thereby providing a reliable basis for efficient disposal of vibration faults of the energy storage hydropower station unit. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 FIG. 1 shows a flowchart of an online vibration fault diagnosis method for an energy storage hydropower station unit according to an embodiment of the present application.

[0061] Figure 2 FIG. 2 shows a function block diagram of an online vibration fault diagnosis system for an energy storage hydropower station unit according to an embodiment of the present application.

[0062] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0063] It should be understood that the specific embodiments described herein are merely intended to explain the present application and are not intended to limit the present application.

[0064] An online vibration fault diagnosis method for an energy storage hydropower station unit is provided in the embodiments of the present application. The execution subject of the online vibration fault diagnosis method for the energy storage hydropower station unit includes but is not limited to at least one of electronic devices such as a server and a terminal that can be configured to execute the method provided in the embodiments of the present application. In other words, the online vibration fault diagnosis method for the energy storage hydropower station unit can be executed by software or hardware installed in a terminal device or a server device. The server includes but is not limited to a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be a stand-alone server, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms, etc.

[0065] Referring to Figure 1 FIG. 1 shows a flowchart of an online vibration fault diagnosis method for an energy storage hydropower station unit according to an embodiment of the present application. In the embodiment, the online vibration fault diagnosis method for the energy storage hydropower station unit includes:

[0066] S1, collecting a raw vibration data set of the generator set under various operating conditions;

[0067] In the embodiment of the present application, the raw vibration data set of the generator set under various operating conditions comprises:

[0068] Collecting multi-source vibration data of the generator set;

[0069] Anti-power frequency interference filtering processing is performed on the multi-source vibration data to obtain pre-processed vibration data of the generator set;

[0070] The pre-processed vibration data is grouped and marked according to the type of operating condition to obtain an operating condition marked vibration data group of the generator set;

[0071] Time synchronization alignment is performed on the operating condition marked vibration data group to obtain a raw vibration data set of the generator set.

[0072] Specifically, collecting multi-source vibration data of the generator set means simultaneously collecting vibration information of each key part of the generator set through vibration sensors installed at different positions of the generator set during operation of the generator set. These sensors correspond to different components such as bearings, rotors, and casings of the generator set, and the collected information collectively constitutes multi-source vibration data.

[0073] Further, anti-power frequency interference filtering processing is performed on the multi-source vibration data to obtain pre-processed vibration data of the generator set. Specifically, a notch filtering method is used, which can specifically filter out interference signals caused by power system frequency in the multi-source vibration data. By setting a filter frequency point equal to the power frequency, the interference signals at this frequency point and nearby are removed from the multi-source vibration data. After such processing, the remaining vibration data is the pre-processed vibration data.

[0074] Further, the pre-processed vibration data is grouped and marked according to the type of operating condition to obtain an operating condition marked vibration data group of the generator set. When operating, first determine various operating conditions of the generator set, such as start-up condition, stable operation condition, and shutdown condition.

[0075] Further, according to the operating condition information recorded during operation of the generator set, all data belonging to the same operating condition in the pre-processed vibration data are grouped together, and the corresponding operating condition type is marked on each group of data, thereby forming each group of data with operating condition marking, which is the operating condition marked vibration data group.

[0076] Further, time synchronization alignment is performed on the operating condition marked vibration data group to obtain a raw vibration data set of the generator set. The method is to select a unified time reference, which can be a specific time during operation of the generator set.

[0077] Further, then, the acquisition time of all data in each working condition marked vibration data set is compared and adjusted with the unified time reference, so that the vibration data collected at different positions under the same working condition is consistent in time, and all the working condition marked vibration data sets after time synchronization alignment collectively form the original vibration data set.

[0078] Overall, the multi-source vibration data of the unit is collected, covering key parts such as bearings, rotors, and casings, avoiding one-sided information, providing complete basic data for comprehensive analysis of vibration state and identification of potential faults, and not missing abnormal signals of key components;

[0079] Overall, the multi-source vibration data is subjected to anti-power frequency interference filtering, the power frequency interference is removed through notch filtering, the irrelevant interference influence is reduced, the preprocessed vibration data is fitted to the actual operation state of the unit, the real fault characteristics are avoided to be covered, and the subsequent analysis accuracy is improved;

[0080] Overall, the preprocessed vibration data is grouped and marked according to the working condition type, the working condition data such as starting, stable operation, and shutdown is clearly distinguished, the analysis deviation caused by mixed data is avoided, the vibration anomaly of a specific working condition can be checked in subsequent targeted manner, and the vibration characteristic difference of the unit in different stages is matched;

[0081] Overall, the working condition marked vibration data set is subjected to time synchronization alignment, the time sequence is adjusted through a unified time reference, the consistency of the data time dimension is ensured, the vibration correlation confusion caused by the acquisition time difference is eliminated, and a time sequence unified data basis is provided for analyzing the vibration correlation of each part and locating the fault source.

[0082] S2, eliminate the time and space deviation of different source signals in the original vibration data set by adjusting the time frame and the space reference system, and obtain a standardized vibration data matrix of the unit;

[0083] In the embodiment of the application, the standardized vibration data matrix of the unit obtained by adjusting the time frame and the space reference system to eliminate the time and space deviation of different source signals in the original vibration data set comprises:

[0084] Align the time frame of the original vibration data set based on a preset time reference sequence, and obtain a time-synchronized signal set of the unit;

[0085] A unified space reference system is established based on the structure topology relationship of the unit, the time-synchronized signal set is mapped to the unified space reference system, and a space-aligned signal set of the unit is obtained;

[0086] The space-aligned signal set is subjected to amplitude normalization processing, and a standardized vibration data matrix of the unit is obtained.

[0087] Specifically, I will follow the requirements and elaborate on the specific implementation process for the three steps of eliminating the spatiotemporal bias of the original vibration data set to obtain the standardized vibration data matrix. Ensure that each step is clear in method, product is clear and no uncertain expression.

[0088] Further, based on the preset time reference sequence, the original vibration data set is time frame aligned to obtain the time synchronization signal set of the unit. First, determine the preset time reference sequence, which is a set of continuous time points preset according to the standard time interval during normal operation of the unit. Then, extract the collection time corresponding to each data point in the original vibration data set one by one, and match each collection time with the time points in the preset time reference sequence.

[0089] Further, if the collection time of a certain data point coincides with a certain time point in the reference sequence, the data point is directly retained; if the collection time is between two reference time points, the vibration value of the corresponding reference time point is calculated according to the vibration values of the adjacent data points before and after the data point through linear interpolation, to supplement the missing reference time point data. After the matching and supplement of all data points in the original vibration data set with the preset time reference sequence are completed, all the obtained data points together constitute the time synchronization signal set.

[0090] Further, based on the structural topology relationship of the unit, a unified spatial reference system is established, and the time synchronization signal set is mapped to the unified spatial reference system to obtain the spatial alignment signal set of the unit. First, sort out the structural topology relationship of the unit, and clearly define the position distribution of each component of the unit and the connection relationship between the components. Take the geometric center of the unit as the origin, and set three mutually perpendicular spatial coordinate axes corresponding to the length direction, width direction and height direction of the unit, thereby establishing a unified spatial reference system.

[0091] Further, determine the position of each data point in the time synchronization signal set corresponding to the unit component, and associate the vibration information of the data point with the corresponding spatial coordinates according to the coordinates of the component in the unified spatial reference system, that is, mark the vibration value of each data point on the spatial coordinate position of the component to which it belongs. After associating all data points with the coordinates of the unified spatial reference system, the data set formed by integrating the spatial coordinates and corresponding vibration values is the spatial alignment signal set.

[0092] Further, the spatial alignment signal set is subjected to amplitude normalization processing to obtain the standardized vibration data matrix of the unit. First, traverse the vibration amplitudes of all data points in the spatial alignment signal set to find the maximum and minimum vibration amplitudes. Then, for the vibration amplitude of each data point, use a fixed calculation method to subtract the minimum vibration amplitude from the vibration amplitude of the data point, and then divide the obtained difference by the difference between the maximum and minimum vibration amplitudes.

[0093] Further, by such calculation, the vibration amplitude of each data point is converted into the range of 0 to 1, and after all the vibration amplitudes of the data points in the space alignment signal set are converted, the data points after amplitude normalization are sequentially arranged into a matrix form according to the coordinate order in the unified space reference system, and the matrix is a standardized vibration data matrix.

[0094] In summary, the original vibration data set is aligned based on the preset time reference sequence to obtain a time synchronization signal set, the data is supplemented by matching a unified time point, the time collection difference is eliminated, the vibration data time dimension is ensured to be consistent, the correlation distortion caused by time sequence confusion is avoided, and a foundation is laid for mining time correlation;

[0095] In summary, a unified space reference system is established according to the structure topology of the unit, the time synchronization signal set is mapped into the unified space reference system to obtain a space alignment signal set, the data corresponding to the component position is clear, the spatial deviation caused by the sensor installation difference is eliminated, and the vibration propagation and fault position correlation are analyzed from the spatial angle;

[0096] In summary, the amplitude of the space alignment signal set is normalized to obtain a standardized vibration data matrix, the amplitude is unified to a fixed range, the order of magnitude difference is eliminated, the influence on the fairness of feature extraction and fault judgment is avoided, a unified format data is formed, and the subsequent diagnosis efficiency and accuracy are improved.

[0097] S3, anti-interference fusion is performed on the standardized vibration data matrix, and dimension unification operation is performed on the data after anti-interference fusion to obtain a multi-source anti-noise data cube of the unit;

[0098] In the embodiment of the application, the anti-interference fusion is performed on the standardized vibration data matrix, and the dimension unification operation is performed on the data after anti-interference fusion to obtain a multi-source anti-noise data cube of the unit, including:

[0099] The multi-source signals in the standardized vibration data matrix are fused to obtain an initial fusion data matrix of the unit;

[0100] The initial fusion data matrix is subjected to wavelet threshold denoising to obtain anti-interference fusion data of the unit;

[0101] The anti-interference fusion data is subjected to dimension normalization processing to obtain a normalized feature data set of the unit;

[0102] The normalized feature data set is structured and integrated according to the feature dimension to obtain a multi-source anti-noise data cube of the unit.

[0103] Specifically, the multi-source signals in the standardized vibration data matrix are fused to obtain an initial fusion data matrix of the unit, specifically, vibration signals from different sensors in the standardized vibration data matrix are combined according to their corresponding positions in a unified spatial reference system.

[0104] Further, in the case that there are multiple sensors collecting vibration signals at the same spatial position, the arithmetic mean of these signals is taken as the vibration signal value at the position, and in the case that there is only one sensor collecting signals at the position, the signal value is directly retained, and in this way, all multi-source signals are integrated into a matrix containing vibration information at all spatial positions, which is the initial fusion data matrix.

[0105] Further, the initial fusion data matrix is subjected to wavelet threshold denoising to obtain anti-interference fusion data of the unit. First, the initial fusion data matrix is decomposed into sub-signals of different frequency components, which contain vibration information of different frequencies, including both effective vibration signals of the unit in normal operation and various noise signals.

[0106] Further, a fixed threshold is then set, and components with amplitudes below the threshold in each sub-signal obtained after decomposition are regarded as noise and their amplitudes are set to zero, while components with amplitudes above the threshold are retained, and then all processed sub-signals are recombined to form a new data set, which is the anti-interference fusion data.

[0107] Further, the anti-interference fusion data is subjected to dimensionless normalization processing to obtain a normalized feature data set of the unit. First, the range of values of all data in the anti-interference fusion data is determined, and the maximum and minimum values are found, and then for each data, the data is subtracted from the minimum value.

[0108] Further, the result obtained is divided by the difference between the maximum value and the minimum value. Through such processing, the value range of all data is uniformly converted to between 0 and 1, ensuring that data of different features have the same dimension. All data obtained after such dimensionless normalization processing collectively constitute the normalized feature data set.

[0109] Further, the normalized feature data set is structured and integrated according to the feature dimensions to obtain a multi-source anti-noise data cube of the unit. First, the feature dimensions included in the normalized feature data set are determined, which include the time dimension, the spatial dimension, and the vibration feature dimension.

[0110] Further, the time dimension corresponds to the time point of data collection, the space dimension corresponds to the coordinate position in the unified space reference system, and the vibration feature dimension corresponds to the processed vibration amplitude value, then the data in the normalized feature data set is rearranged according to the three feature dimensions, so that each data point corresponds to specific values in the three dimensions, and finally a three-dimensional data structure is formed, which is a multi-source anti-noise data cube.

[0111] In summary, the multi-source signals fused with the standardized vibration data matrix form an initial fusion data matrix, the signals of sensors at different parts of the unit are integrated, the limitations of a single signal perspective are compensated for, a data matrix reflecting the overall vibration state is formed, comprehensive support is provided for subsequent fault feature extraction, and fault feature omission is reduced.

[0112] In summary, the anti-interference fusion data is obtained by carrying out wavelet threshold denoising on the initial fusion data matrix, the data is decomposed into sub-signals of different frequencies, low-amplitude noise is removed, effective vibration signals are retained, environmental and electromagnetic interference is removed, the data is fitted to the actual vibration of the unit, noise is avoided from masking fault features, and data reliability is improved.

[0113] In summary, the normalized feature data set is obtained by carrying out dimensionless normalization on the anti-interference fusion data, the order of magnitude deviation caused by the range and unit difference of the sensor is eliminated, the data value range is unified, the weights of each feature data are ensured to be equal, the dimension problem is avoided from dominating the analysis result, the feature analysis is ensured to be fair and accurate, and a foundation is laid for structured integration.

[0114] In summary, the multi-source anti-noise data cube is obtained by integrating the normalized feature data set according to the feature dimension, a three-dimensional structure is constructed in the time, space and vibration feature dimensions, the dimensional correlation is clearly presented, multi-dimensional coupling analysis of fault features is facilitated, the fault time, position and performance are quickly located, and the diagnosis efficiency and accuracy are improved.

[0115] S4, multi-dimensional vibration fault feature coupling analysis is performed on the multi-source anti-noise data cube to obtain a vibration fault feature topology set of the multi-source anti-noise data cube.

[0116] In the embodiment of the application, the multi-dimensional vibration fault feature coupling analysis is performed on the multi-source anti-noise data cube to obtain the vibration fault feature topology set of the multi-source anti-noise data cube, which includes:

[0117] The time domain vibration features and the frequency domain vibration features in the multi-source anti-noise data cube are extracted, and the time domain vibration features and the frequency domain vibration features are integrated to obtain a vibration feature vector set of the unit.

[0118] The vibration feature vector set is subjected to time-frequency domain correlation transformation to obtain a feature coupling relationship matrix of the multi-source anti-noise data cube.

[0119] performing topology analysis based on the feature coupling relationship matrix to obtain a vibration fault feature topology graph of the unit;

[0120] extracting a topology connection relationship and a node attribute in the vibration fault feature topology graph and collecting them as a vibration fault feature topology set of the multi-source anti-noise data cube.

[0121] Specifically, time-domain vibration features and frequency-domain vibration features in the multi-source anti-noise data cube are extracted, and the time-domain vibration features and the frequency-domain vibration features are integrated to obtain a vibration feature vector set of the unit. Specifically, vibration amplitude variation law related information corresponding to each time point is extracted from time dimension data of the multi-source anti-noise data cube as a time-domain vibration feature.

[0122] For example, the maximum value, the minimum value and the change trend of the vibration amplitude in each time period; meanwhile, vibration energy distribution related information corresponding to different frequency bands is extracted from frequency dimension data of the multi-source anti-noise data cube as a frequency-domain vibration feature.

[0123] For example, the strength of the vibration energy in a specific frequency band, and then the time-domain vibration features and the frequency-domain vibration features corresponding to each spatial position at the same time point are arranged in sequence to form a vector containing time-domain and frequency-domain information of the position. All such vectors corresponding to different time points of all spatial positions jointly constitute the vibration feature vector set.

[0124] Further, time-frequency domain correlation transformation is performed on the vibration feature vector set to obtain a feature coupling relationship matrix of the multi-source anti-noise data cube. When operating, any two different feature vectors in the vibration feature vector set are selected, one of which is mainly a time-domain feature and the other of which is mainly a frequency-domain feature. The corresponding relationship of the two vectors at different time points and frequency bands is analyzed to determine whether the change of the time-domain feature will cause the corresponding change of the frequency-domain feature and the correlation degree of the change.

[0125] Further, a fixed correlation value is assigned to this correlation degree. In the same way, the correlation values between all pairs of feature vectors in the vibration feature vector set are calculated in sequence. These correlation values are filled into the matrix according to the arrangement order of the feature vectors in the form of row and column correspondence. The rows and columns correspond to different feature vectors respectively, and each element in the matrix is the correlation value of the corresponding two feature vectors. The matrix formed in this way is the feature coupling relationship matrix.

[0126] Further, topology analysis is performed based on the feature coupling relationship matrix to obtain a vibration fault feature topology graph of the unit. First, each feature vector in the feature coupling relationship matrix is regarded as a node, and the name of the node is consistent with the identification of the corresponding feature vector.

[0127] Furthermore, the correlation values ​​of each element in the matrix are then examined. If the correlation value between two eigenvectors is greater than the set correlation threshold, then a correlation is considered to exist between the two nodes. The two nodes are then connected by a line segment. The thickness of the line segment can be set according to the magnitude of the correlation value; the larger the correlation value, the thicker the line segment.

[0128] Furthermore, if the correlation value is less than or equal to the correlation threshold, then the two nodes are not connected. In this way, all nodes with correlation relationships are connected with line segments, and the resulting graph containing nodes and connecting line segments is the vibration fault feature topology map.

[0129] Furthermore, the topological connections and node attributes in the vibration fault feature topology map are extracted and aggregated into a vibration fault feature topology set of a multi-source noise-resistant data cube. First, the connection between all nodes in the vibration fault feature topology map is recorded one by one, including which nodes each node is connected to and the degree of correlation corresponding to the thickness of the connecting line segments. This information together constitutes the topological connection relationship.

[0130] Furthermore, the attribute information corresponding to each node is extracted simultaneously, including the type of feature vector represented by the node, the spatial location and time range corresponding to the feature vector. This information together constitutes the node attributes. Then, all topological connections and node attributes are organized into the same data set to ensure that each connection can correspond to a specific node attribute. This data set is the vibration fault feature topology set of the multi-source noise-resistant data cube.

[0131] In summary, the time-domain and frequency-domain vibration features of the multi-source noise-resistant data cube are extracted and integrated into a vibration feature vector set. At the same time, the amplitude variation law in the time dimension and the energy distribution in the frequency dimension are obtained, avoiding the limitations of a single dimension, making the vibration features more comprehensive, providing a complete feature foundation for subsequent analysis, and reducing the risk of missing fault features.

[0132] In summary, the time-frequency domain correlation transformation of the vibration feature vector set yields the feature coupling relationship matrix, which quantitatively presents the coupling relationship of all feature vectors, clearly shows the intrinsic connection of time-frequency domain features, breaks the isolation of analysis, provides quantitative basis for topology analysis, and helps to uncover hidden fault correlation features.

[0133] In summary, topological analysis based on the feature coupling relationship matrix yields a topological map of vibration fault features. Feature vectors are transformed into nodes, and coupling relationships are transformed into connecting line segments. The thickness of the line segments reflects the degree of correlation, making the abstract relationships visual and facilitating the identification of key nodes and associated paths, thus providing intuitive support for locating fault features.

[0134] In summary, the connection relationships and node attributes of the extracted vibration fault feature topology map are aggregated into a vibration fault feature topology set, forming a structured feature set. This retains key information and facilitates subsequent matching with historical fault case databases, providing standardized and reusable data for diagnosis and improving matching efficiency and accuracy.

[0135] S5. Perform feature mapping matching between the vibration fault feature topology set and the historical fault case library to obtain the fault diagnosis result sequence of the vibration fault feature topology set;

[0136] In this embodiment of the invention, the step of performing feature mapping matching between the vibration fault feature topology set and the historical fault case database to obtain the fault diagnosis result sequence of the vibration fault feature topology set includes:

[0137] Obtain the fault feature topology structure from the historical fault case library;

[0138] Extract the feature vectors of the corresponding nodes in the vibration fault feature topology set and the fault feature topology structure;

[0139] Based on the feature vectors, the topological similarity between the vibration fault feature topology set and the fault feature topology structure is calculated;

[0140] Based on the comparison results between the topological similarity and the preset threshold, matching fault feature topological structures are selected to form a candidate fault set.

[0141] The confidence level of the fault types in the candidate fault set is evaluated to obtain the fault type confidence sequence of the candidate fault set;

[0142] The fault diagnosis result sequence is generated based on the fault type confidence sequence.

[0143] The formula for calculating the topological similarity is as follows:

[0144]

[0145] In the formula, The topological similarity is... This represents the total number of nodes in the characteristic topology. For the first The preset weight coefficients for each node, For node indexing, For the vibration fault feature topology set, the first The feature vector of each node The fault feature topology is the first The feature vector of each node This is a function of the similarity between two feature vectors.

[0146] Specifically, to obtain the fault feature topology structure in the historical fault case library, it is necessary to first determine the storage path and data format of the case library. The case library stores the fault feature topology structure corresponding to various faults in the past of the unit. Each structure contains node information related to the fault type and the connection relationship between nodes.

[0147] Furthermore, by calling the case library data reading interface, all complete fault feature topologies are extracted according to the fault time sequence or type classification. At the same time, the integrity of each structure is verified to ensure that no nodes are missing or the connection relationship is damaged, and finally the fault feature topologies in the historical fault case library are obtained.

[0148] Furthermore, feature vectors of corresponding nodes in the vibration fault feature topology set and the fault feature topology structure are extracted. First, the attribute information of the nodes in the two sets is compared, such as the feature type represented by the node, the corresponding spatial location and time range. The corresponding nodes with completely consistent attributes are found, and the feature vectors of the time domain and frequency domain feature combination of the corresponding nodes are extracted from the vibration fault feature topology set.

[0149] Furthermore, feature vectors of corresponding nodes are extracted from the fault feature topology, ensuring that the vector dimension and data type are consistent. All fault feature topologies in the case library are processed in the same way to obtain feature vectors of all corresponding node pairs.

[0150] Furthermore, based on the feature vectors, the topological similarity between the vibration fault feature topology set and the fault feature topology structure is calculated. First, the feature vector similarity of each pair of corresponding nodes is calculated. The feature values ​​of the same dimension are compared, and the number of dimensions with differences within the allowable range is counted. The difference is divided by the total number of dimensions to obtain the similarity of a single node pair. Then, the arithmetic mean of the similarities of all corresponding node pairs is calculated.

[0151] Furthermore, the number of identical connections between the two is counted simultaneously, and the connection similarity is obtained by dividing it by the total number of connections in the vibration fault feature topology set. The two similarities are then weighted and summed in equal proportions, and the result is the topological similarity between the two. Finally, the topological similarity with all other structures in the case library is calculated.

[0152] Furthermore, based on the comparison results of topological similarity and preset threshold, matching structures are selected to form a candidate fault set. A fixed value of the preset threshold is determined. The topological similarity of each fault feature topological structure is compared with the threshold. If the similarity is greater than the threshold, it is determined to be a match, marked as a matching structure and associated with its corresponding fault type. If it is less than or equal to the threshold, it is excluded. All matching structures and corresponding fault types are collected to form a candidate fault set.

[0153] Furthermore, the confidence of the fault types in the candidate fault set is evaluated to obtain a fault type confidence sequence. The frequency of each fault type in the candidate set is counted and divided by the frequency of occurrence of the total number of fault types. The average topological similarity of the matching structure corresponding to each fault type is then calculated. The frequency of occurrence and the average topological similarity are weighted by a fixed ratio to calculate the confidence of each fault type. The fault types and their corresponding confidence are arranged from high to low to form a fault type confidence sequence.

[0154] Furthermore, a fault diagnosis result sequence is generated based on the fault type confidence sequence. First, it is verified that there are no duplicate fault types in the sequence and that the confidence is calculated accurately. Fault types with high confidence levels and their confidence levels are retained. If all confidence levels are low, the few with the highest confidence levels are retained. Diagnostic content based on historical cases is added to each retained fault type. The results are then sorted from high to low confidence levels into an ordered list containing fault type, confidence level, and diagnostic content, which is the fault diagnosis result sequence.

[0155] Specifically, the total number of nodes in the characteristic topology is obtained by counting each node in the vibration fault characteristic topology set or the fault characteristic topology, and the two numbers are consistent.

[0156] Furthermore, the first The preset weight coefficients of each node are pre-set based on the degree of influence of the node on the fault judgment in historical fault diagnosis data. Key nodes have a larger weight and are stored in the historical fault case library configuration file for use.

[0157] Furthermore, the node index is a unique identifier assigned to each node according to the order of nodes in the feature topology, used for quick node location.

[0158] Furthermore, the vibration fault feature topology set of the first The feature vector of each node is formed by extracting the time-domain and frequency-domain vibration features of the corresponding node from the topological set and combining them in a fixed order.

[0159] Furthermore, in the fault feature topology, the first The feature vector of each node is formed by extracting the time-domain and frequency-domain vibration features of the corresponding index node from the structure of the historical case library and combining them in the same order.

[0160] Furthermore, a function for calculating the similarity between two feature vectors is used to determine the degree of directional consistency by comparing the distribution of feature values ​​in each dimension of the vectors. The closer the directions are, the greater the result. The calculation logic is embedded in the diagnostic system algorithm module.

[0161] Furthermore, the formula quantifies the similarity between the vibration fault feature topology set and the fault feature topology, introduces node weights to highlight the influence of key nodes, and makes the results more in line with the actual fault judgment needs, accurately reflecting the matching of the two fault features.

[0162] Furthermore, the formula shows that the closer the corresponding node feature vectors are in direction, the larger the similarity function result, the more significant the influence of key nodes, the more the total numerator increases, and the greater the topological similarity; conversely, the similarity decreases as the direction of the feature vectors of the two nodes changes accordingly.

[0163] In summary, by acquiring the fault characteristic topology structure of the historical fault case library and calling up the complete topology information of past unit faults, a reliable historical reference is provided for subsequent matching, avoiding the problem of directionless diagnosis caused by lack of data support, and ensuring that the diagnosis is traceable.

[0164] In summary, the corresponding node feature vectors of the vibration fault feature topology set and the fault feature topology structure are extracted. By comparing node attributes, matching nodes are found, eliminating node bias. This provides a vector basis with unified dimensions and data types for topology similarity calculation, ensuring the accuracy of the calculation.

[0165] In summary, calculating topological similarity based on feature vectors quantifies the degree of similarity between the node features and connection relationships of two entities, transforming abstract matching into measurable values, avoiding subjective biases in empirical judgments, and providing a precise quantitative standard for selecting matching structures.

[0166] In summary, by selecting matching structures based on topological similarity and preset thresholds to form a candidate fault set, the matching criteria are clarified, low-association historical structures are eliminated, irrelevant interference is reduced, the diagnostic scope is narrowed, and the efficiency of subsequent confidence assessment is improved.

[0167] In summary, confidence scores are obtained by evaluating the failure types in the candidate failure set. The confidence scores are then calculated by combining the frequency of occurrence with the average topological similarity, thus quantifying the degree of credibility, avoiding single-factor bias, and providing a priority basis for the diagnostic result sequence.

[0168] In summary, this topological similarity calculation formula introduces the first... The preset weight coefficients of each node highlight the impact of key nodes on fault judgment. Key nodes are given higher weights because they play a greater role in historical fault diagnosis. Their contribution to the similarity results is amplified during calculation to avoid the bias caused by equal weights, so that the results are closer to actual needs and the sensitivity of core fault feature matching is improved.

[0169] In summary, the formula uses a sine function to calculate the first digit of the two topologies. The similarity of feature vectors of individual nodes accurately measures the degree of consistency in vector direction. The closer the direction, the larger the function result. It precisely quantifies the matching situation of a single node, provides a reliable basis for the overall similarity calculation, and reduces the deviation caused by inaccurate feature vector comparison.

[0170] In summary, the formula calculates the overall topological similarity by summing the products of the weight coefficients of all nodes and their corresponding vector similarities in the numerator, and then dividing by the sum of the weight coefficients of all nodes. This reasonable calculation considers both the matching situation of all nodes and the differences in node importance, avoiding the excessive influence of a single node. It provides an accurate quantitative standard for screening matching structures and constructing candidate fault sets, reducing false or false faults.

[0171] In summary, a diagnostic result sequence is generated based on the confidence sequence of the fault type, and the fault information is sorted and organized according to the confidence level to form a clear and prioritized result, which facilitates focusing on the core fault.

[0172] S6. Based on real-time feedback data, the fault diagnosis result sequence is dynamically corrected to obtain the final fault diagnosis report of the unit.

[0173] In this embodiment of the invention, the step of dynamically correcting the fault diagnosis result sequence based on real-time feedback data to obtain the final fault diagnosis report of the unit includes:

[0174] The real-time vibration signal during the operation of the unit is acquired, and the time-domain and frequency-domain features of the real-time vibration signal are extracted to construct the real-time feature vector of the unit.

[0175] The real-time feature vector is compared with a preset historical fault feature database to obtain the feature deviation value of the real-time feature vector;

[0176] The confidence level of the fault type in the fault diagnosis result sequence is dynamically adjusted according to the characteristic deviation value to obtain the corrected fault diagnosis result sequence.

[0177] The final fault diagnosis report of the unit is generated based on the corrected fault diagnosis result sequence.

[0178] The step of dynamically adjusting the confidence level of the fault type in the fault diagnosis result sequence based on the characteristic deviation value to obtain the corrected fault diagnosis result sequence includes:

[0179] Obtain the initial confidence level and corresponding feature deviation value of the fault type in the fault diagnosis result sequence;

[0180] Determine the corresponding confidence adjustment strategy based on the aforementioned feature deviation value;

[0181] The initial confidence level is dynamically corrected based on the confidence level adjustment strategy to obtain the target confidence level of the unit;

[0182] The fault types in the fault diagnosis result sequence are reordered according to the target confidence level to obtain the corrected fault diagnosis result sequence.

[0183] Specifically, real-time vibration signals during unit operation are acquired, and the time-domain and frequency-domain features of the real-time vibration signals are extracted to construct the unit's real-time feature vector.

[0184] Furthermore, by installing vibration sensors at various key parts of the unit, vibration signals during the current operation of the unit are continuously collected. These signals cover the vibration status of different components of the unit. From the collected real-time vibration signals, the maximum value, minimum value and trend of vibration amplitude at each time point are extracted as time-domain features.

[0185] Furthermore, the intensity distribution of vibration energy in different frequency bands is extracted as frequency domain features. These time domain features and frequency domain features are combined and arranged in a fixed order to form a vector that reflects the current vibration state of the unit. This vector is the real-time feature vector of the unit.

[0186] Furthermore, the real-time feature vector is compared with the preset historical fault feature library to obtain the feature deviation value of the real-time feature vector. The preset historical fault feature library stores the feature vectors corresponding to the occurrence of various faults. Each feature vector contains the time domain and frequency domain features of a specific fault. During the comparison, historical fault feature vectors with the same dimension as the real-time feature vector and the same unit part are selected from the library, and the feature values ​​of the same position of the two are compared one by one.

[0187] Furthermore, the difference between the feature values ​​at each corresponding location is calculated, and the absolute values ​​of all the differences are summed and averaged. The result is the feature deviation value of the real-time feature vector, which reflects the degree of difference between the real-time features and the historical fault features.

[0188] Furthermore, the confidence level of the fault type in the fault diagnosis result sequence is dynamically adjusted based on the feature deviation value to obtain the corrected fault diagnosis result sequence. When the feature deviation value is small, it indicates that the real-time features are not significantly different from the historical fault features. In this case, the confidence level of the corresponding fault type is appropriately increased, and the increase is inversely proportional to the deviation value.

[0189] Furthermore, when the feature deviation value is large, it indicates a large difference. The confidence level of the corresponding fault type is appropriately reduced, and the reduction is proportional to the deviation value. After adjusting the confidence level of all fault types in this way, the resulting new sequence is the corrected fault diagnosis result sequence.

[0190] Furthermore, based on the corrected fault diagnosis result sequence, the final fault diagnosis report of the unit is generated. First, the corrected sequence is sorted from high to low confidence, and the top three fault types with the highest confidence are selected as the main diagnostic results. For each fault type, a detailed description is generated by combining the typical manifestations, possible causes and handling measures of the fault in the preset historical fault feature library.

[0191] Furthermore, this information is organized into a document according to confidence level from high to low. The document also includes basic information such as the extraction time of the real-time feature vector and the feature deviation value. The final complete document is the final fault diagnosis report of the unit.

[0192] Specifically, to obtain the initial confidence level and corresponding feature deviation value of the fault type in the fault diagnosis result sequence, it is necessary to first extract each fault type and its corresponding initial confidence level from the generated fault diagnosis result sequence. The initial confidence level is a value determined after the confidence evaluation of the candidate fault set, and is directly associated with the fault type and stored in the sequence.

[0193] Furthermore, based on the correspondence between fault types and historical fault characteristics, the feature deviation values ​​corresponding to the historical fault characteristics matched by each fault type are selected from the calculated feature deviation value data. This ensures that each fault type can be matched with a unique initial confidence level and a unique feature deviation value, forming a one-to-one correspondence of "fault type - initial confidence level - feature deviation value", providing a data basis for subsequent adjustments.

[0194] Furthermore, based on the feature deviation value, the corresponding confidence adjustment strategy is determined. First, three feature deviation value ranges are preset in the system, namely the low deviation range, the medium deviation range, and the high deviation range. Each range corresponds to a fixed confidence adjustment method.

[0195] Furthermore, the low deviation range corresponds to the strategy of "increasing confidence," with the increase percentage being a fixed value, meaning the initial confidence increases by that percentage; the medium deviation range corresponds to the strategy of "keeping confidence unchanged," meaning the initial confidence is not adjusted; and the high deviation range corresponds to the strategy of "decreasing confidence," with the decrease percentage also being a fixed value, meaning the initial confidence decreases by that percentage. Once the characteristic deviation value of a certain fault type is obtained, it is determined which preset range the deviation value falls into, thereby determining the confidence adjustment strategy applicable to that fault type.

[0196] Furthermore, the initial confidence level is dynamically corrected based on the confidence level adjustment strategy to obtain the target confidence level of the unit. For each type of fault, the correction operation is performed according to the determined confidence level adjustment strategy: if the strategy is "increase confidence level", the initial confidence level of the fault type is multiplied by a value added to the preset increase ratio, and the result is the corrected confidence level.

[0197] Furthermore, if the strategy is "keep the confidence level unchanged", the initial confidence level is directly used as the corrected result; if the strategy is "reduce the confidence level", the initial confidence level is multiplied by a value that is subtracted from the preset reduction ratio, and the result is the corrected confidence level; the confidence level obtained after each fault type is corrected by the corresponding strategy is the target confidence level for that fault type.

[0198] Furthermore, the fault types in the fault diagnosis result sequence are reordered according to the target confidence level to obtain the corrected fault diagnosis result sequence. First, all fault types and their corresponding target confidence levels are summarized to form a list of "fault type - target confidence level". Then, the fault types in the list are sorted in descending order of target confidence level.

[0199] Furthermore, if there are fault types with the same target confidence level, they are arranged according to their order in the original fault diagnosis result sequence. After sorting, a new fault type sequence is formed, in which each fault type carries the corresponding target confidence level. This new sequence is the corrected fault diagnosis result sequence.

[0200] In summary, acquiring real-time vibration signals from the generator unit and extracting time-domain and frequency-domain features to construct real-time feature vectors can capture the latest vibration information of the generator unit, avoid diagnostic lag, provide real-time data support for dynamically correcting diagnostic results, and ensure that diagnostic results keep up with changes in the generator unit's status.

[0201] In summary, by comparing real-time feature vectors with a preset historical fault feature database to obtain feature deviation values, the difference between current and historical fault features is quantified, providing an objective basis for judging fault correlation, avoiding subjective bias in experience-based judgments, and providing clear data references for confidence adjustment.

[0202] In summary, the confidence level of the fault type in the fault diagnosis result sequence is adjusted according to the characteristic deviation value. The confidence level is increased when the deviation is small and decreased when the deviation is large. This optimizes the initial diagnosis results, avoids the deviation caused by the limitations of historical cases, and makes the corrected sequence more consistent with the current actual fault state of the unit, thereby improving the accuracy of diagnosis.

[0203] In summary, the final fault diagnosis report is generated based on the corrected sequence. Combined with real-time data and correlation analysis of high-confidence faults, a complete, accurate, and instructive report is formed, providing a basis for staff to understand the fault situation and formulate handling plans, thereby improving the efficiency and accuracy of fault handling.

[0204] In summary, obtaining the initial confidence level and corresponding feature deviation value of the fault type in the fault diagnosis result sequence, clarifying the credibility and feature differences of each fault type, establishing a one-to-one correspondence among the three, provides a clear data basis for confidence level adjustment, avoids adjustment without basis, and ensures traceability and verification.

[0205] In summary, the confidence adjustment strategy is determined based on the characteristic deviation value. The direction and method of adjustment are clarified by the correspondence between the preset deviation range and the strategy, avoiding subjective arbitrariness, making the adjustment have fixed standards, and ensuring the process is standardized and consistent.

[0206] In summary, the target confidence level is obtained by adjusting the initial confidence level based on the adjustment strategy, and the credibility of each fault type is optimized according to the feature differences. This avoids the initial confidence level from being out of touch with the actual fault state of the unit, and makes the target confidence level more in line with the real credibility.

[0207] In summary, the corrected sequence obtained by reordering fault types according to target confidence level prioritizes high-confidence fault types, helping staff quickly focus on core faults, avoid missing key points, improve the practicality and guidance of diagnostic results, and provide clear basis for handling priorities.

[0208] like Figure 2 The diagram shown is a functional block diagram of an online diagnostic system for vibration faults in a hydropower station unit provided in an embodiment of the present invention.

[0209] The online vibration fault diagnosis system 100 for energy storage hydropower station units described in this invention can be installed in an electronic device. Depending on the functions implemented, the online vibration fault diagnosis system 100 for energy storage hydropower station units may include a multi-source vibration data acquisition module 101, a spatiotemporal standardization processing module 102, a multi-source noise-resistant data cube establishment module 103, a feature coupling analysis module 104, an intelligent diagnosis matching module 105, and a dynamic optimization evaluation module 106. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0210] In this embodiment, the functions of each module / unit are as follows:

[0211] The multi-source vibration data acquisition module 101 is used to acquire the original vibration data set of the unit under various operating conditions.

[0212] The spatiotemporal standardization processing module 102 is used to eliminate the spatiotemporal deviation of signals from different sources in the original vibration data set by adjusting the time frame and spatial reference system, so as to obtain the standardized vibration data matrix of the unit.

[0213] The multi-source noise-resistant data cube establishment module 103 is used to perform anti-interference fusion on the standardized vibration data matrix and to perform dimension unification operation on the data after anti-interference fusion to obtain the multi-source noise-resistant data cube of the unit.

[0214] The feature coupling analysis module 104 is used to perform multi-dimensional vibration fault feature coupling analysis on the multi-source noise-resistant data cube to obtain the vibration fault feature topology set of the multi-source noise-resistant data cube.

[0215] The intelligent diagnosis matching module 105 is used to perform feature mapping matching between the vibration fault feature topology set and the historical fault case library to obtain the fault diagnosis result sequence of the vibration fault feature topology set.

[0216] The dynamic optimization evaluation module 106 is used to dynamically correct the fault diagnosis result sequence based on real-time feedback data to obtain the final fault diagnosis report of the unit.

[0217] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0218] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0219] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0220] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0221] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0222] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for online diagnosis of vibration faults in a hydropower station unit, characterized in that, The method includes: S1. Collect raw vibration data sets of the unit under various operating conditions; S2. By adjusting the time frame and spatial reference system, the spatiotemporal deviation of signals from different sources in the original vibration data set is eliminated, and the standardized vibration data matrix of the unit is obtained. S3. Perform anti-interference fusion on the standardized vibration data matrix, and perform dimensional unification on the fused data to obtain the multi-source noise-resistant data cube of the unit, including: The multi-source signals in the standardized vibration data matrix are fused to obtain the initial fused data matrix of the unit. Wavelet threshold denoising is performed on the initial fused data matrix to obtain the anti-interference fused data of the unit; The anti-interference fusion data is subjected to dimension normalization processing to obtain the normalized feature dataset of the unit; The normalized feature dataset is structured and integrated according to the feature dimensions to obtain the multi-source noise-resistant data cube of the unit; S4. Perform multi-dimensional vibration fault feature coupling analysis on the multi-source noise-resistant data cube to obtain the vibration fault feature topology set of the multi-source noise-resistant data cube; S5. Perform feature mapping matching between the vibration fault feature topology set and the historical fault case database to obtain a fault diagnosis result sequence of the vibration fault feature topology set, including: Obtain the fault feature topology structure from the historical fault case library; Extract the feature vectors of the corresponding nodes in the vibration fault feature topology set and the fault feature topology structure; Based on the feature vectors, the topological similarity between the vibration fault feature topology set and the fault feature topology is calculated. The formula for calculating the topological similarity is as follows: ； In the formula, The topological similarity is... This represents the total number of nodes in the characteristic topology. For the first The preset weight coefficients for each node, For node indexing, For the vibration fault feature topology set, the first The feature vector of each node The fault feature topology is the first The feature vector of each node This is a function of the similarity between two feature vectors; Based on the comparison results between the topological similarity and the preset threshold, matching fault feature topological structures are selected to form a candidate fault set. The confidence level of the fault types in the candidate fault set is evaluated to obtain the fault type confidence sequence of the candidate fault set; The fault diagnosis result sequence is generated based on the fault type confidence sequence; S6. Based on real-time feedback data, the fault diagnosis result sequence is dynamically corrected to obtain the final fault diagnosis report of the unit.

2. The online diagnosis method for vibration faults of energy storage hydropower station units as described in claim 1, characterized in that, The raw vibration data sets of the data acquisition unit under various operating conditions include: Collect multi-source vibration data of the unit; The multi-source vibration data is subjected to anti-power frequency interference filtering to obtain the pre-processed vibration data of the unit; The preprocessed vibration data is grouped and labeled according to the operating condition type to obtain the operating condition labeled vibration data group of the unit. The vibration data set marked by the operating conditions is time-synchronized and aligned to obtain the original vibration dataset of the unit.

3. The online diagnosis method for vibration faults of energy storage hydropower station units as described in claim 1, characterized in that, The process of eliminating spatiotemporal biases of signals from different sources in the original vibration dataset by adjusting the time frame and spatial reference frame to obtain the standardized vibration data matrix of the unit includes: The original vibration dataset is time-frame aligned based on a preset time reference sequence to obtain the time synchronization signal set of the unit. A unified spatial reference system is established based on the structural topology of the unit, and the time synchronization signal set is mapped to the unified spatial reference system to obtain the spatial alignment signal set of the unit. The amplitude normalization process is performed on the spatially aligned signal set to obtain the standardized vibration data matrix of the unit.

4. The online diagnosis method for vibration faults of energy storage hydropower station units as described in claim 1, characterized in that, The multi-dimensional vibration fault feature coupling analysis of the multi-source noise-resistant data cube yields a vibration fault feature topology set for the multi-source noise-resistant data cube, including: Extract the time-domain vibration features and frequency-domain vibration features from the multi-source noise-resistant data cube, and integrate the time-domain vibration features and frequency-domain vibration features to obtain the vibration feature vector set of the unit; Perform a time-frequency domain correlation transformation on the vibration feature vector set to obtain the feature coupling relationship matrix of the multi-source noise-resistant data cube; Based on the aforementioned characteristic coupling relationship matrix, a topology analysis is performed to obtain the vibration fault characteristic topology map of the unit. The topological connections and node attributes in the vibration fault feature topology map are extracted and aggregated into the vibration fault feature topology set of the multi-source noise-resistant data cube.

5. The online diagnosis method for vibration faults of energy storage hydropower station units as described in claim 1, characterized in that, The process of dynamically correcting the fault diagnosis result sequence based on real-time feedback data to obtain the final fault diagnosis report of the unit includes: The real-time vibration signal during the operation of the unit is acquired, and the time-domain and frequency-domain features of the real-time vibration signal are extracted to construct the real-time feature vector of the unit. The real-time feature vector is compared with a preset historical fault feature database to obtain the feature deviation value of the real-time feature vector; The confidence level of the fault type in the fault diagnosis result sequence is dynamically adjusted according to the characteristic deviation value to obtain the corrected fault diagnosis result sequence. The final fault diagnosis report of the unit is generated based on the corrected fault diagnosis result sequence.

6. The online diagnosis method for vibration faults of energy storage hydropower station units as described in claim 5, characterized in that, The step of dynamically adjusting the confidence level of the fault type in the fault diagnosis result sequence based on the characteristic deviation value to obtain the corrected fault diagnosis result sequence includes: Obtain the initial confidence level and corresponding feature deviation value of the fault type in the fault diagnosis result sequence; Determine the corresponding confidence adjustment strategy based on the aforementioned feature deviation value; The initial confidence level is dynamically corrected based on the confidence level adjustment strategy to obtain the target confidence level of the unit; The fault types in the fault diagnosis result sequence are reordered according to the target confidence level to obtain the corrected fault diagnosis result sequence.

7. An online diagnostic system for vibration faults in a hydropower station unit, characterized in that, For using the online diagnosis method for vibration faults of a hydropower station unit according to claim 1, the system comprises: The multi-source vibration data acquisition module is used to collect raw vibration datasets of the unit under various operating conditions; The spatiotemporal standardization processing module is used to eliminate the spatiotemporal deviation of signals from different sources in the original vibration data set by adjusting the time frame and spatial reference system, so as to obtain the standardized vibration data matrix of the unit. A multi-source noise-resistant data cube building module is used to perform anti-interference fusion on the standardized vibration data matrix and to perform dimensional unification operations on the fused data to obtain the multi-source noise-resistant data cube of the unit, including: The multi-source signals in the standardized vibration data matrix are fused to obtain the initial fused data matrix of the unit. Wavelet threshold denoising is performed on the initial fused data matrix to obtain the anti-interference fused data of the unit; The anti-interference fusion data is subjected to dimension normalization processing to obtain the normalized feature dataset of the unit; The normalized feature dataset is structured and integrated according to the feature dimensions to obtain the multi-source noise-resistant data cube of the unit; The feature coupling analysis module is used to perform multi-dimensional vibration fault feature coupling analysis on the multi-source noise-resistant data cube to obtain the vibration fault feature topology set of the multi-source noise-resistant data cube. The intelligent diagnostic matching module is used to perform feature mapping matching between the vibration fault feature topology set and the historical fault case library to obtain a fault diagnosis result sequence of the vibration fault feature topology set, including: Obtain the fault feature topology structure from the historical fault case library; Extract the feature vectors of the corresponding nodes in the vibration fault feature topology set and the fault feature topology structure; Based on the feature vectors, the topological similarity between the vibration fault feature topology set and the fault feature topology is calculated. The formula for calculating the topological similarity is as follows: ； In the formula, The topological similarity is... This represents the total number of nodes in the characteristic topology. For the first The preset weight coefficients for each node, For node indexing, For the vibration fault feature topology set, the first The feature vector of each node The fault feature topology is the first The feature vector of each node This is a function of the similarity between two feature vectors; Based on the comparison results between the topological similarity and the preset threshold, matching fault feature topological structures are selected to form a candidate fault set. The confidence level of the fault types in the candidate fault set is evaluated to obtain the fault type confidence sequence of the candidate fault set; The fault diagnosis result sequence is generated based on the fault type confidence sequence; The dynamic optimization and evaluation module is used to dynamically correct the fault diagnosis result sequence based on real-time feedback data to obtain the final fault diagnosis report of the unit.

Images