A hydroelectric generator set bearing bush temperature anomaly identification and fault positioning method

Abstract

This invention discloses a method for identifying and locating abnormal bearing temperatures in hydro-generator units. The method involves acquiring bearing temperature data and unit operating parameters from a database; identifying the current operating condition based on unit speed and active power; using a multi-criteria fusion approach to identify abnormal bearing temperatures; extracting preset interval statistical features from three dimensions—cooling system, lubrication system, and shaft vibration—to form a multi-dimensional feature vector; calculating the mean vector and covariance matrix corresponding to each predefined typical bearing system fault to form a fault standard library; and using the mean vector of each fault type as a reference center and the covariance matrix as a distribution scale, calculating the statistical distance from the multi-dimensional feature vector of the current abnormal window to each fault category using the low-rank Mahalanobis distance method, and determining the category with the smallest distance as the most likely fault type, outputting the fault location result.

Description

Technical Field

[0001] This invention relates to a method for identifying abnormal bearing temperatures and locating faults in hydro-generator sets, belonging to the field of intelligent operation and maintenance and fault diagnosis technology for hydro-generator sets. Background Technology

[0002] (1) Bearing temperature alarm technology based on fixed threshold:

[0003] Existing systems typically compare the bearing temperature with preset static alarm thresholds (such as 65℃ or 70℃): using simple upper and lower limit comparison logic; an alarm signal is triggered when the temperature exceeds the threshold; without involving dynamic trends, inter-bearing relationships, or condition-dependent modeling.

[0004] (2) Anomaly detection techniques based on simple trends:

[0005] The analysis only focuses on a single temperature variable and does not take into account key operational information such as the unit's operating status.

[0006] (1) Monitoring methods based on a single threshold have a high false alarm rate and lack early warning capabilities. Fixed threshold mechanisms have the following problems:

[0007] With changes in load, cooling water temperature, or oil temperature, bearing temperature will fluctuate normally; seasonal temperature differences and cooler switching cause frequent triggering of thresholds; key characteristics such as temperature rise trends and fluctuation patterns cannot be identified; and there is a lack of ability to identify early signs of failure.

[0008] (2) Unable to identify complex anomalies such as trend types and local hotspot types:

[0009] Anomalies in tile temperature manifest as: slow temperature rise, increased local temperature differences between tiles, abnormal fluctuation amplitude, and distribution bias. Existing methods can only identify anomalies that "exceed a threshold," but cannot identify trend changes or structural changes related to the tiles.

[0010] Single criterion, high false alarm / false negative rate: Traditional methods cannot accurately identify comprehensive anomalies with multiple measurement points and multiple time series characteristics, and are easily affected by noise and occasional fluctuations;

[0011] (3) Lack of an engineering-applicable fault location system: Existing technologies can only tell maintenance personnel that "a certain tile is hot", but cannot: indicate the fault type; output confidence level; identify key tiles; or provide quantitative evidence. Therefore, it relies on experienced engineers to make manual judgments, which is inefficient and prone to misjudgment. Summary of the Invention

[0012] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for identifying and locating abnormal bearing temperatures in hydro-generator sets. By using multi-criteria fusion anomaly detection, interval statistical feature construction, and low-rank extended Mahalanobis distance discrimination, the method can achieve real-time automatic identification of anomalies and accurate fault location for various bearing measuring points such as thrust bearings and upper guide bearings.

[0013] To achieve the above objectives / to solve the above technical problems, the present invention is implemented using the following technical solution: A method for identifying and locating abnormal bearing temperatures in a hydro-generator set, comprising: S1: Obtain temperature data of each bearing and unit operating parameters from the database, and perform multi-level preprocessing on the temperature data; S2: Identify the current operating condition based on the unit speed and active power, and use a multi-criteria fusion method to identify abnormalities in the bearing temperature. If an abnormality exists, extract preset interval statistical features from three dimensions—cooling system, lubrication system, and shaft vibration—for the abnormal time window to form a multi-dimensional feature vector. S3: For each type of predefined typical bearing system fault, based on the multidimensional feature vector extracted within the historical fault time period, calculate the mean vector and covariance matrix corresponding to the fault type to form a fault standard library; wherein, the mean vector is obtained by taking the arithmetic mean of all valid historical samples of the fault type in each feature dimension, and is used to characterize the central prototype of the fault type in the multidimensional feature space. S4: For the multidimensional feature vector of the current abnormal window, use the mean vector of each type of fault in step S3 as the reference center and the covariance matrix as the distribution scale, and use the low-rank Mahalanobis distance method to calculate the statistical distance to each fault category respectively. Then, determine the category with the smallest distance as the most likely fault type and output the fault location result.

[0014] Optionally, the multi-level preprocessing includes outlier removal, validity assessment, missing value completion, and resampling to a 1-minute sampling period.

[0015] Optionally, the multiple criteria include a temperature threshold criterion, a temperature rise trend criterion, and a tile consistency criterion. If any criterion is triggered, it is determined that there is a temperature anomaly.

[0016] Optionally, identifying the current operating condition based on the unit speed and active power includes: The operating status of the unit is determined based on the unit speed n and active power P. When 99.5% of n... r ≤n≤100.5%n r When n is in the operating state, the unit is determined to be in operation. r This refers to the rated speed of the unit; Determine the active power P of the unit under operating conditions. When P ≥ P thThen it is considered to be in a power generation state, P≤-P th This indicates a pumping state, where P th This is an empirical threshold.

[0017] Optionally, the temperature threshold criterion includes: The upper limit T of temperature at each measuring point is configured using a relational database. max As a temperature threshold, when T > T max If an error occurs, it will be recorded and displayed.

[0018] Optionally, the temperature rise trend criterion is: The temperature slope is calculated using a sliding time window method, with a time window W = 5 min and a step size S = 1 min. The temperature data T within this time window is evaluated; if T < (T0 + T) max Skip the time window when T0 is 2 / 2, where T0 is the ambient temperature; the formula for calculating the temperature rise slope is: ; In the formula, k is the slope, t represents the current time, and T t Let T be the temperature at time t. t -W represents the temperature at time tW. When the slope k exceeds the threshold, an anomaly is recorded and displayed.

[0019] Optionally, the tile consistency criterion includes: calculating the average temperature T of each tile over the 10 minutes. avg If the average temperature of a tile deviates from that of the lowest tile by more than 3°C, it is considered that there is a local abnormal hot spot and the abnormality is recorded.

[0020] Optionally, the construction of the multidimensional feature vector includes: Average technical water supply flow rate Q avg The total technical water supply flow rate within the window is taken as the average value over time. Q i Let N be the technical water supply flow rate at the ii-th sampling point; N is the total number of sampling points within the abnormal time window, and i is the serial number of the sampling point, i=1,2,…N; Calculate the difference between the inlet and outlet water temperatures ΔT, and the average difference between the inlet and outlet water temperatures: ;T in,avg T represents the average inlet water temperature of the technical water supply within the time window. out,avg This represents the average temperature of the technical water supply outlet within the time window. Average outlet pressure p of the technical water supply pump avg , ;p i The outlet pressure of the technical water supply pump at the i-th sampling point; The pressure difference Δp across the water supply filter screen. ;pin,avg p represents the average pressure before the filter within the time window. out,avg This represents the average pressure behind the filter within the time window. Average oil level H in the oil tank avg , H i Let i be the oil level in the oil tank at the i-th sampling point; Average oil temperature T in the oil tank t-avg , ;T t,i Let i be the oil temperature in the oil tank at the i-th sampling point; X-axis swing peak value X p2p , Xmax and Xmin are the maximum and minimum values ​​of the X-axis swing signal within the time window, respectively. X-axis swing low-frequency component amplitude X l , ; This function represents the operation function for taking the amplitude of a signal. X-axis swing high-frequency component amplitude X h , ; Y-axis swing peak value Y p2p， Corresponding to the X-axis, it reflects the range of swing in the Y-axis. ; The maximum value of the Y-axis swing signal within the abnormal time window. The minimum value of the Y-axis swing signal within the abnormal time window; Y-axis swing low-frequency component amplitude Y l , ; Y-axis swing high-frequency component amplitude Y h , ; This forms an interval-type feature vector.

[0021] Optionally, the formation of the fault standard library includes:

[0022] Fault Category Statistical Feature Model Construction: For each fault category c, its sample set is denoted as , Xc(i) represents the multidimensional feature vector of the i-th historical sample of the c-th type of fault; Built based on the sample set: Class mean vector, Nc represents the total number of valid historical samples for the c-th type of fault. Class covariance matrix: μ cLet Σc be the mean vector of the c-th type of fault, Σc be the covariance matrix of the c-th type of fault, and T denote the matrix transpose operation.

[0023] Optionally, the low-rank Mahalanobis distance fault location includes: The process for handling singularities in the covariance matrix is ​​as follows: Perform SVD decomposition: ; Where U is the eigenvector matrix, Given an eigenvalue matrix, select the first k eigenvalues ​​to form: , The k-th eigenvalue represents the variance of the data along the direction of the k-th principal component. Corresponding eigenvector matrix: ; For the low-rank inverse approximation, we get: ; For corresponding eigenvectors, U k Let denot be the truncated eigenvector matrix composed of the first k eigenvectors, let Λk be the truncated diagonal matrix composed of the first k largest eigenvalues, and let Σc,k be the low-rank approximation matrix of the covariance matrix of the c-th type of fault. After integrating the interval feature vectors X, calculate the following for each class c: Dc is the low-rank Mahalanobis distance from the current sample XX to the center of the cc-th type of fault; X is the multidimensional feature vector of the current sample to be tested; μc represents the mean vector of the c-th type of fault. Obtain the statistical distance from the current sample to the center of each fault class; Final Failure Category: ; Fault confidence level: ; Key feature contribution Sort the output to the most sensitive features. The index for the most likely fault category is determined. Fault confidence, W i Contribution to key features.

[0024] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0025] (1) Improve the accuracy of anomaly identification and reduce the risk of false alarms and missed alarms: This invention expands the single threshold judgment into a multi-dimensional joint discrimination method by constructing a multi-criteria fusion identification mechanism based on temperature threshold criteria, temperature rise trend criteria, and tile consistency criteria. This allows the anomaly identification process to simultaneously consider absolute temperature level, change trend, and structural relationship between tiles. Compared with traditional methods based solely on fixed alarm thresholds, this invention can effectively reduce false alarms caused by changes in normal operating conditions such as load fluctuations, ambient temperature changes, and cooling system switching, while improving the ability to identify slow-heating anomalies and local hotspot anomalies.

[0026] (2) Achieving multi-dimensional feature modeling based on system mechanism correlation: This invention constructs interval statistical feature vectors within the anomaly window, and integrates cooling system parameters, lubrication system parameters, and shaft vibration parameters into the feature space for comprehensive analysis. This allows bearing temperature anomalies to no longer rely solely on a single temperature variable for judgment, but rather on pattern recognition through multi-system correlation features. Through the above technical solution, a statistical correlation can be established between temperature anomalies and changes in cooling capacity, oil film state, and shaft dynamic characteristics, thereby improving the physical consistency and engineering rationality of fault identification.

[0027] (3) Solving the problem of singularity of covariance matrix under small sample conditions and improving algorithm stability: In actual engineering scenarios, the number of fault samples is usually limited and the feature dimension is high, which can easily lead to non-invertibility of the covariance matrix, thus affecting the stability of traditional Mahalanobis distance calculation. This invention uses the singular value decomposition method of covariance matrix to construct a low-rank approximate inverse matrix, which suppresses the influence of noise direction while retaining the main feature direction information, so that Mahalanobis distance calculation still maintains numerical stability and statistical validity under small sample conditions, thereby improving the reliability of fault location results.

[0028] (4) Achieving quantitative localization of multiple types of faults: This invention constructs a fault standard library and establishes corresponding mean vectors and covariance matrix models for each type of fault, enabling the current abnormal feature vector to calculate its statistical distance to the center of each type of fault. By comparing the magnitude of various distances, the most likely fault category can be determined, and the corresponding confidence information can be output, thus upgrading from "abnormal prompts" to "fault category localization". Attached Figure Description

[0029] Figure 1 The diagram shown is a schematic representation of the method flow of the present invention. Detailed Implementation

[0030] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0031] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0032] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0033] like Figure 1 As shown, a method for identifying and locating abnormal bearing temperatures in a hydro-generator unit is disclosed, including:

[0034] S1: Obtain temperature data of each bearing and unit operating parameters from the database, and perform multi-level preprocessing on the temperature data;

[0035] S2: Identify the current operating condition based on the unit speed and active power, and use a multi-criteria fusion method to identify abnormalities in the bearing temperature. If an abnormality exists, extract preset interval statistical features from three dimensions—cooling system, lubrication system, and shaft vibration—for the abnormal time window to form a multi-dimensional feature vector.

[0036] S3: For each type of predefined typical bearing system fault, based on the multidimensional feature vector extracted within the historical fault time period, calculate the mean vector and covariance matrix corresponding to the fault type to form a fault standard library; wherein, the mean vector is obtained by taking the arithmetic mean of all valid historical samples of the fault type in each feature dimension, and is used to characterize the central prototype of the fault type in the multidimensional feature space.

[0037] S4: For the multidimensional feature vector of the current abnormal window, use the mean vector of each type of fault in step S3 as the reference center and the covariance matrix as the distribution scale, and use the low-rank Mahalanobis distance method to calculate the statistical distance to each fault category respectively. Then, determine the category with the smallest distance as the most likely fault type and output the fault location result.

[0038] The following section details a method for identifying abnormal bearing temperatures and locating multi-dimensional faults in hydroelectric generator sets (including pumped storage units). This invention proposes a method for real-time automatic identification and accurate fault location of bearing temperature anomalies at various bearing measuring points, including thrust bearings and upper guide bearings, through multi-criteria fusion anomaly detection, interval statistical feature construction, and low-rank extended Mahalanobis distance discrimination. The technical solution includes the following steps:

[0039] (1) Data acquisition and multi-level preprocessing

[0040] Retrieve unit operating parameters (power, speed), configuration parameters (thresholds for temperature measurement points at each bearing) and real-time data (temperature values ​​for each bearing) from relational databases and real-time databases, and perform the following preprocessing steps:

[0041] 1) Data acquisition: Data is acquired every 5 minutes. Based on the current time t0, the temperature data from the current time to time t1 (t0-10min) is acquired.

[0042] 2) Abnormal point elimination: Eliminate point values ​​that do not conform to engineering physics, such as temperature of the tile temperature measuring point being less than 0℃ or greater than 200℃.

[0043] 3) Missing complement: Determine the temperature value N after removing outliers. If N is less than 2, the temperature data for that channel is considered invalid. Valid temperature data undergoes pre-complementation and resampling, with a resampling period of 1 minute per point.

[0044] (2) Waven anomaly identification based on multi-criteria (threshold, trend, consistency) fusion

[0045] 1) Unit operating condition identification: The unit's operating status is determined based on its speed (n) and active power (P). When 99.5% of n... r ≤n≤100.5%n r When n is in the operating state, the unit is determined to be in operation. r The rated speed of the unit. Determine the active power P of the unit under operating conditions; when P ≥ P0... th Then it is considered to be in a power generation state, P≤-P th This indicates a pumping state, where P th This is an empirical threshold (e.g., 3MW). Different temperature measuring points are selected based on the unit's operating conditions (the temperature measuring points for the same tile differ between power generation and pumping modes in pumped-storage units).

[0046] 2) Temperature threshold determination: Use the upper temperature limit T for each measuring point configured in the relational database. max As a temperature threshold, when T > T max If an error occurs, it will be recorded and displayed.

[0047] 3) Trend Judgment: The temperature slope is calculated using a sliding time window method, with a time window W = 5 min and a step size S = 1 min. First, the temperature data T within this time window is judged. When T < (T0 + T) max Skip the window when T0 is 2 / 2, where T0 is the ambient temperature. The formula for calculating the temperature rise slope is: ; In the formula, k is the slope, and T t Let T be the temperature at time t. t -W represents the temperature at time tW. An anomaly is recorded and displayed when the slope k exceeds the threshold.

[0048] 4) Consistency judgment: Calculate the average temperature T of each tile over the 10 minutes. avg If the average temperature of a tile deviates from that of the lowest tile by more than 3°C, it is considered that there is a local abnormal hot spot and the abnormality is recorded.

[0049] 5) Abnormal Trigger: If any of the above three criteria is triggered, the current window is judged to be in an abnormal period and the fault location process is triggered.

[0050] (3) Extraction of statistical features of intervals To improve the robustness of anomaly identification and fault location, this invention constructs interval statistical feature vectors from three aspects: cooling system, lubrication system, and shaft vibration.

[0051] 1) Average technical water supply flow rate Q avg The total technical water supply flow rate within the window is taken as the average value over time. ;

[0052] This characteristic reflects the actual water supply capacity of the bearing cooling system and is closely related to the bearing shell temperature control capability. When the water supply is insufficient, the shell temperature tends to rise slowly.

[0053] 2) Temperature difference ΔT between inlet and outlet water in the technical water supply

[0054] Calculate the difference between the average inlet and outlet water temperatures: ;

[0055] A large temperature difference usually indicates that the cooling water has a stronger heat absorption capacity, which may be related to higher oil temperature and increased bearing friction power; a small temperature difference may mean that the cooling efficiency has decreased, the flow rate is insufficient, or there is a blockage.

[0056] 3) Average outlet pressure p of the technical water supply pump avg ;

[0057] This indicator reflects changes in water supply pipeline resistance and pump operating stability, and is an important basis for judging whether the cooling system is functioning properly.

[0058] 4) Pressure difference Δp before and after the technical water supply filter ;

[0059] An increase in pressure differential usually indicates filter blockage, which leads to a decrease in effective flow rate and can directly cause an increase in bearing temperature. This is an important characteristic of abnormal cooling.

[0060] 5) Average oil level H in the oil tank avg ;

[0061] Low oil levels can lead to poor oil film formation, increased stress on local bearing plates, and a sudden increase in bearing temperature.

[0062] 6) Average oil temperature T in the oil tank t-avg； ;

[0063] Oil temperature affects oil film viscosity and bearing friction power, and is one of the important driving factors for the increase of bearing temperature.

[0064] 7) Peak-to-peak value of X-axis swing p2p ;

[0065] The peak value of the runout reflects the range of the shaft's swing in the X direction. An increase in the value may indicate shaft eccentricity, uneven oil film, or increased clearance.

[0066] 8) Amplitude of the low-frequency component of the X-axis swing l ;

[0067] This component often corresponds to mechanisms such as hydraulic fluctuations and structural loosening.

[0068] 9) Amplitude of the high-frequency component of the X-axis swing h ; This is usually associated with increased friction and early signs of oil film rupture.

[0069] 10) Peak-to-peak value of Y-axis swing p2p

[0070] Corresponding to the X-axis, it reflects the swing range in the Y-axis.

[0071] ;

[0072] 11) Amplitude of the low-frequency component of the Y-axis swing l

[0073] Similar to the X-axis, it is obtained through low-frequency filtering.

[0074] ;

[0075] 12) Amplitude of high-frequency component of Y-axis swing h ,

[0076] Similar to the X-axis, it is obtained through high-frequency filtering.

[0077] ;

[0078] This forms an interval-type feature vector.

[0079] (4) Construction of the fault standard library To achieve fault mode recognition based on statistical distance, this invention constructs a multi-category fault standard sample library for hydroelectric generator bearing systems. This sample library is organized according to different fault categories, with each category consisting of multiple interval feature samples to characterize the typical features of that type of fault in the cooling system, lubrication system, and shaft dynamics.

[0080] 1) Definition and classification of fault categories

[0081] Based on industry experience and operational mechanisms, this invention constructs a typical fault location for bearing systems:

[0082] 1. Technical water supply pump failure; 2. Technical water supply filter failure; 3. Oil circulation system failure; 4. Early signs of oil film rupture; 5. Shaft eccentricity; 6. Bearing rubbing.

[0083] 2) Fault Sample Extraction Process

[0084] The construction of each type of fault sample follows the following process: 1. Determine the time of fault occurrence by using operation logs, expert experience, and historical diagnostic records to determine the specific time period in which the fault occurred;

[0085] 2. Generate sliding window interval features. ;

[0086] 3. Obtain multiple sample records, divide each fault interval into several windows, and generate a feature vector for each window;

[0087] 4. Sample denoising and normalization: outlier values ​​are removed, and data is normalized according to feature dimensions.

[0088] 3) Construction of statistical feature model for fault categories

[0089] For each type of fault c, its sample set is denoted as: ;

[0090] Built based on the sample set:

[0091] 1. Class mean vector ;

[0092] 2. Similar to covariance matrix ;

[0093] In scenarios with a large number of features but a scarcity of faulty samples (e.g., 12-dimensional features with only 10–20 samples per class), the aforementioned covariance matrix is ​​usually a singular matrix and cannot be directly inverted. Therefore, this invention further introduces a low-rank approximation.

[0094] (5) Low-rank Mahalanobis distance fault location

[0095] The dimensionality of the feature matrix and related features is high, but the number of samples is limited, resulting in a covariance matrix... Normally, it is not invertible. This invention uses SVD decomposition to obtain a low-rank approximate inverse matrix.

[0096] 1) Low-rank Mahalanobis distance

[0097] 1. Singularity handling process for covariance matrix

[0098] right Perform SVD decomposition: ;

[0099] in Given an eigenvalue matrix, select the first k eigenvalues ​​to form: ;

[0100] Corresponding eigenvector matrix: ;

[0101] 2. Low-rank inverse approximation ;

[0102] 2) Calculation of low-rank Mahalanobis distance

[0103] After integrating the interval feature vectors X, calculate the following for each class c: ;

[0104] Obtain the statistical distance from the current sample to the center of each fault class.

[0105] 3) Determining the optimal fault category

[0106] Final Failure Category: ;

[0107] Fault confidence level: ; 4) Contribution of key features

[0108] The most sensitive features can be sorted and output.

[0109] In summary, this application has the following effects:

[0110] (1) Improve the accuracy of anomaly identification and reduce the risk of false alarms and missed alarms: This invention expands the single threshold judgment into a multi-dimensional joint discrimination method by constructing a multi-criteria fusion identification mechanism based on temperature threshold criteria, temperature rise trend criteria, and tile consistency criteria. This allows the anomaly identification process to simultaneously consider absolute temperature level, change trend, and structural relationship between tiles. Compared with traditional methods based solely on fixed alarm thresholds, this invention can effectively reduce false alarms caused by changes in normal operating conditions such as load fluctuations, ambient temperature changes, and cooling system switching, while improving the ability to identify slow-heating anomalies and local hotspot anomalies.

[0111] (2) Achieving multi-dimensional feature modeling based on system mechanism correlation: This invention constructs interval statistical feature vectors within the anomaly window, and integrates cooling system parameters, lubrication system parameters, and shaft vibration parameters into the feature space for comprehensive analysis. This allows bearing temperature anomalies to no longer rely solely on a single temperature variable for judgment, but rather on pattern recognition through multi-system correlation features. Through the above technical solution, a statistical correlation can be established between temperature anomalies and changes in cooling capacity, oil film state, and shaft dynamic characteristics, thereby improving the physical consistency and engineering rationality of fault identification.

[0112] (3) Solving the problem of singularity of covariance matrix under small sample conditions and improving algorithm stability: In actual engineering scenarios, the number of fault samples is usually limited and the feature dimension is high, which can easily lead to non-invertibility of the covariance matrix, thus affecting the stability of traditional Mahalanobis distance calculation. This invention uses the singular value decomposition method of covariance matrix to construct a low-rank approximate inverse matrix, which suppresses the influence of noise direction while retaining the main feature direction information, so that Mahalanobis distance calculation still maintains numerical stability and statistical validity under small sample conditions, thereby improving the reliability of fault location results.

[0113] (4) Achieving quantitative localization of multiple types of faults: This invention constructs a fault standard library and establishes corresponding mean vectors and covariance matrix models for each type of fault, enabling the current abnormal feature vector to calculate its statistical distance to the center of each type of fault. By comparing the magnitude of various distances, the most likely fault category can be determined, and the corresponding confidence information can be output, thus upgrading from "abnormal prompts" to "fault category localization".

[0114] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for identifying and locating abnormal bearing temperatures in a hydro-generator set, characterized in that, include: S1: Obtain temperature data of each bearing and unit operating parameters from the database, and perform multi-level preprocessing on the temperature data; S2: Identify the current operating condition based on the unit speed and active power, and use a multi-criteria fusion method to identify abnormalities in the bearing temperature. If an abnormality exists, extract preset interval statistical features from three dimensions—cooling system, lubrication system, and shaft vibration—for the abnormal time window to form a multi-dimensional feature vector. S3: For each type of predefined typical bearing system fault, based on the multidimensional feature vector extracted within the historical fault time period, calculate the mean vector and covariance matrix corresponding to the fault type to form a fault standard library; wherein, the mean vector is obtained by taking the arithmetic mean of all valid historical samples of the fault type in each feature dimension, and is used to characterize the central prototype of the fault type in the multidimensional feature space. S4: For the multidimensional feature vector of the current abnormal window, use the mean vector of each type of fault in step S3 as the reference center and the covariance matrix as the distribution scale, and use the low-rank Mahalanobis distance method to calculate the statistical distance to each fault category respectively. Then, determine the category with the smallest distance as the most likely fault type and output the fault location result.

2. The method for identifying and locating abnormal bearing temperatures in a hydro-generator unit according to claim 1, characterized in that, The multi-level preprocessing includes outlier removal, validity assessment, missing value completion, and resampling to a 1-minute sampling period.

3. The method for identifying and locating abnormal bearing temperatures in a hydro-generator unit according to claim 1, characterized in that, The multiple criteria include temperature threshold criteria, temperature rise trend criteria, and tile consistency criteria. If any criterion is triggered, it is determined that there is a temperature anomaly.

4. The method for identifying and locating abnormal bearing temperatures in a hydro-generator set according to claim 3, characterized in that, The method of identifying the current operating condition based on the unit speed and active power includes: The operating status of the unit is determined based on the unit speed n and active power P. When 99.5% of n... r ≤n≤100.5%n r When n is in the operating state, the unit is determined to be in operation. r This is the rated speed of the unit, expressed in r / min. Determine the active power P of the unit under operating conditions. When P ≥ P th Then it is considered to be in a power generation state, P≤-P th This is the pumping state, where P th This is an empirical threshold.

5. The method for identifying and locating abnormal bearing temperatures in a hydro-generator unit according to claim 1, characterized in that, The temperature threshold criterion includes: The upper limit T of temperature at each measuring point is configured using a relational database. max As a temperature threshold, when T > T max If an error occurs, it will be recorded and displayed.

6. The method for identifying and locating abnormal bearing temperatures in a hydro-generator unit according to claim 1, characterized in that, The criteria for judging the temperature rise trend are as follows: The temperature slope is calculated using a sliding time window method, with a time window W = 5 min and a step size S = 1 min. The temperature data T within this time window is evaluated; if T < (T0 + T) max Skip the time window when T0 is 2 / 2, where T0 is the ambient temperature; the formula for calculating the temperature rise slope is: ； In the formula, k is the slope, t represents the current time, and T t Let T be the temperature at time t. t -W represents the temperature at time tW. When the slope k exceeds the threshold, an anomaly is recorded and displayed.

7. The method for identifying and locating abnormal bearing temperatures in a hydro-generator set according to claim 1, characterized in that, The tile consistency criterion includes: calculating the average temperature T of each tile over the 10 minutes. avg If the average temperature of a tile deviates from that of the lowest tile by more than 3°C, it is considered that there is a local abnormal hot spot and the abnormality is recorded.

8. The method for identifying and locating abnormal bearing temperatures in a hydro-generator set according to claim 1, characterized in that, The multidimensional feature vector includes: Average technical water supply flow rate Q avg The total technical water supply flow rate within the window is taken as the average value over time. Q i Let N be the technical water supply flow rate at the ii-th sampling point; N is the total number of sampling points within the abnormal time window, and i is the serial number of the sampling point, i=1,2,…N; Calculate the difference between the inlet and outlet water temperatures ΔT, and the average difference between the inlet and outlet water temperatures: ;T in,avg T represents the average inlet water temperature of the technical water supply within the time window. out,avg This represents the average temperature of the technical water supply outlet within the time window. Average outlet pressure p of the technical water supply pump avg , ;p i The outlet pressure of the technical water supply pump at the i-th sampling point; The pressure difference Δp across the water supply filter screen. ;p in,avg p represents the average pressure before the filter within the time window. out,avg This represents the average pressure behind the filter within the time window. Average oil level H in the oil tank avg , H i Let i be the oil level in the oil tank at the i-th sampling point; Average oil temperature T in the oil tank t-avg , ;T t,i Let i be the oil temperature in the oil tank at the i-th sampling point; X-axis swing peak value X p2p , Xmax and Xmin are the maximum and minimum values ​​of the X-axis swing signal within the time window, respectively. X-axis swing low-frequency component amplitude X l , ; This function represents the operation function for taking the amplitude of a signal. X-axis swing high-frequency component amplitude X h , ; Y-axis swing peak value Y p2p， Corresponding to the X-axis, it reflects the range of swing in the Y-axis. ; The maximum value of the Y-axis swing signal within the abnormal time window. The minimum value of the Y-axis swing signal within the abnormal time window; Y-axis swing low-frequency component amplitude Y l , ; Y-axis swing high-frequency component amplitude Y h , ; This forms an interval-type feature vector.

9. The method for identifying and locating abnormal bearing temperatures in a hydro-generator set according to claim 1, characterized in that, The formation of the fault standard library includes: Fault Category Statistical Feature Model Construction: For each fault category c, its sample set is denoted as , Xc(i) represents the multidimensional feature vector of the i-th historical sample of the c-th type of fault; Built based on the sample set: Class mean vector, Nc represents the total number of valid historical samples for the c-th type of fault. Class covariance matrix: μ c Let Σc be the mean vector of the c-th type of fault, Σc be the covariance matrix of the c-th type of fault, and T denote the matrix transpose operation.

10. The method for identifying and locating abnormal bearing temperatures in a hydro-generator unit according to claim 9, characterized in that, The low-rank Mahalanobis distance fault location includes: The process for handling singularities in the covariance matrix. Perform SVD decomposition: Where U is the eigenvector matrix, Given an eigenvalue matrix, select the first k eigenvalues ​​to form: , The k-th eigenvalue represents the variance of the data along the direction of the k-th principal component. Corresponding eigenvector matrix: For the low-rank inverse approximation, we get: ; For corresponding eigenvectors, U k Let denot be the truncated eigenvector matrix composed of the first k eigenvectors, let Λk be the truncated diagonal matrix composed of the first k largest eigenvalues, and let Σc,k be the low-rank approximation matrix of the covariance matrix of the c-th type of fault. After integrating the interval feature vectors X, calculate the following for each class c: Dc is the low-rank Mahalanobis distance from the current sample XX to the center of the cc-th type of fault; X is the multidimensional feature vector of the current sample to be tested; μc represents the mean vector of the c-th type of fault. Obtain the statistical distance from the current sample to the center of each fault class; Final Fault Category: ； Fault confidence level: ； Key feature contribution Sort the output to the most sensitive features. The index for the most likely fault category is determined. Fault confidence, W i Contribution to key features.

Images