Hydropower unit vibration fault diagnosis and health management method and system

Abstract

The application provides a method and system for fault diagnosis and health management of hydroelectric generating set vibration, belonging to the technical field of hydroelectric generating set fault diagnosis. The method comprises the following steps: determining vibration monitoring targets based on the characteristics and operation requirements of the hydroelectric generating set, then determining and executing vibration data collection and diagnosis strategies to obtain vibration monitoring data of the hydroelectric generating set; obtaining the current operation condition of the generating set and the vibration characteristic data in the vibration monitoring data and calculating the condition deviation; if the condition deviation is greater than the deviation threshold, obtaining the health state evaluation value of the generating set components, extracting and evaluating the part of monitoring data with the health state evaluation value less than the first health threshold as the second vibration monitoring data to obtain the fault feature quality evaluation result, selecting the corresponding target correction strategy from the fault diagnosis correction strategy library based on the result to obtain the fault diagnosis correction strategy and correcting the fault diagnosis result of the second vibration monitoring data based on the fault diagnosis correction strategy to obtain the target fault diagnosis result.

Description

Technical Field

[0001] This application relates to the technical field of fault diagnosis of hydropower station equipment, and in particular to a method and system for fault diagnosis and health management of vibration of hydropower station units. Background Technology

[0002] Hydropower station generating units, as equipment for clean energy production, operate under complex conditions for extended periods. Their vibration status directly impacts the safety and stability of the unit's operation. During operation, abnormal vibrations can occur due to various factors such as hydraulic impact, mechanical wear, and electromagnetic interference. Failure to diagnose and address these issues promptly and accurately can lead to component damage, unit shutdown, and even safety accidents, resulting in significant economic losses. Existing vibration fault diagnosis methods often suffer from indiscriminate selection of monitoring targets, lack of specificity in data collection and diagnosis strategies, and insufficient consideration of the impact of operating condition deviations on diagnostic results. Furthermore, they suffer from insufficient accuracy in fault feature extraction and difficulty in effectively correcting diagnostic biases. These methods fail to achieve accurate identification of unit vibration faults and comprehensive health management, thus failing to meet the actual needs of efficient, safe, and stable operation of hydropower station units.

[0003] Therefore, there is an urgent need for a fault diagnosis and health management method and system for the vibration of hydropower station units. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method and system for fault diagnosis and health management of vibration in hydropower station units.

[0005] A first aspect of this application provides a method for fault diagnosis and health management of vibration in hydropower station units, comprising:

[0006] Based on the unit characteristics and operational requirements of the target hydropower station, the corresponding vibration monitoring targets are determined;

[0007] Based on the vibration monitoring target, determine the corresponding vibration data acquisition and diagnosis strategy;

[0008] The vibration data acquisition and diagnosis strategy is executed, and vibration monitoring data of the target hydropower station unit is acquired based on vibration sensors;

[0009] The system acquires the current operating conditions of the unit during monitoring and the vibration characteristic data from the vibration monitoring data, and calculates the deviation between the unit's operating conditions and the preset standard operating conditions.

[0010] If the deviation of the operating condition is greater than the preset deviation threshold, the health status assessment value of the unit component in the vibration monitoring data is obtained, and the monitoring data in the vibration monitoring data in which the health status assessment value of the unit component is less than the first health threshold is used as the second vibration monitoring data.

[0011] Fault feature extraction and quality assessment are performed on the second vibration monitoring data to obtain fault feature quality assessment results;

[0012] Based on the fault characteristic quality assessment results, a corresponding target correction strategy is selected from the preset fault diagnosis correction strategy library to obtain the preset fault diagnosis correction strategy. The fault diagnosis results of the second vibration monitoring data are then corrected based on the preset fault diagnosis correction strategy to obtain the target fault diagnosis result.

[0013] A second aspect of this application provides a fault diagnosis and health management system for vibration of hydropower station units, including:

[0014] The target determination module is used to determine the corresponding vibration monitoring targets based on the unit characteristics and operational requirements of the target hydropower station.

[0015] The diagnostic strategy module is used to determine the corresponding vibration data acquisition and diagnostic strategy based on the vibration monitoring target.

[0016] The execution strategy module is used to execute the vibration data acquisition and diagnosis strategy and acquire vibration monitoring data of the target hydropower station unit based on the vibration sensor;

[0017] The operating condition calculation module is used to obtain the current operating condition of the unit during monitoring and the vibration characteristic data in the vibration monitoring data, and to calculate the deviation of the operating condition between the unit operating condition and the preset standard operating condition.

[0018] The data filtering module is used to obtain the health status assessment value of the unit components in the vibration monitoring data if the deviation of the operating condition is greater than a preset deviation threshold, and to take the part of the monitoring data in the vibration monitoring data whose health status assessment value of the unit components is less than a first health threshold as the second vibration monitoring data.

[0019] The quality assessment module is used to extract fault features and assess the quality of the second vibration monitoring data to obtain fault feature quality assessment results.

[0020] The strategy selection module is used to select a corresponding target correction strategy from a preset fault diagnosis correction strategy library based on the fault feature quality assessment results, obtain a preset fault diagnosis correction strategy, and correct the fault diagnosis results of the second vibration monitoring data based on the preset fault diagnosis correction strategy to obtain the target fault diagnosis result.

[0021] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for fault diagnosis and health management of vibration of hydropower station units.

[0022] In a fourth aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for fault diagnosis and health management of vibration of hydropower station units.

[0023] The beneficial effects of the fault diagnosis and health management method and system for hydropower station unit vibration provided in this application are as follows: This application first determines the monitoring target and the acquisition and diagnosis strategy, and then filters abnormal vibration data according to the deviation of operating conditions and health status, which can effectively eliminate invalid data under normal operating conditions and reduce diagnostic interference; by adaptively selecting correction strategies through fault feature quality assessment, the diagnostic results are optimized, thereby improving the accuracy and reliability of hydropower station unit vibration fault diagnosis, and realizing refined and intelligent fault diagnosis and health management. Attached Figure Description

[0024] Figure 1 A flowchart illustrating a method for fault diagnosis and health management of hydropower station unit vibration provided in an embodiment of this application;

[0025] Figure 2 A structural block diagram of a fault diagnosis and health management system for hydropower station unit vibration provided in an embodiment of this application;

[0026] Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0027] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0028] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0029] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for fault diagnosis and health management of hydropower station unit vibration according to an embodiment of this application. The method includes:

[0030] S101: Based on the unit characteristics and operational requirements of the target hydropower station, determine the corresponding vibration monitoring target.

[0031] In this embodiment, the target hydropower station is the hydropower station for which unit vibration fault diagnosis and health management are to be performed. It can be distinguished based on installed capacity, unit type (e.g., mixed-flow, axial-flow), and years of operation, and is the object of the entire method. Unit characteristics are the inherent attributes of the target hydropower station's units, including unit model, structural parameters, operating parameter ranges, historical fault records, material fatigue characteristics, and operating stress characteristics, which serve as the basis for determining the monitoring target. Among these, structural parameters include rotor diameter and stator winding structure; operating parameter ranges include rated speed, rated load, and rated head.

[0032] In this embodiment, operational requirements refer to the specific requirements for vibration monitoring during the actual operation and maintenance of the target hydropower station. These requirements mainly include monitoring accuracy requirements, real-time monitoring requirements, and monitoring cost constraints, which directly determine the selection criteria for monitoring targets. Specifically, monitoring accuracy requirements include the range of vibration amplitude monitoring errors; real-time monitoring requirements include data acquisition intervals and diagnostic response times; and monitoring cost constraints include the number of sensors deployed and the upper limit of operation and maintenance costs. The vibration monitoring targets are the key components of the unit that ultimately require vibration data acquisition and fault monitoring, such as rotors, stators, bearings, and main shafts. These components are the objects of vibration data acquisition and diagnostic strategy formulation and must meet operational requirements and be compatible with the unit's characteristics.

[0033] S102: Based on the vibration monitoring target, determine the corresponding vibration data acquisition and diagnosis strategy.

[0034] In this embodiment, vibration data acquisition is a comprehensive acquisition scheme developed to obtain effective vibration data from the vibration monitoring target. This scheme includes: sensor deployment locations, sensor types, data acquisition frequency, acquisition duration, and data filtering standards (rules for removing noisy data), ensuring that the acquired vibration data is accurate, complete, and meets diagnostic requirements. The diagnostic strategy is a complete solution for diagnosing faults in the vibration monitoring target based on the acquired vibration data. This includes: fault feature extraction methods, fault identification standards, diagnostic procedures, and fault grading standards, used to accurately determine whether the monitored target has a fault, the type of fault, and its severity.

[0035] S103: Execute vibration data acquisition and diagnosis strategies, and acquire vibration monitoring data of the target hydropower station unit based on vibration sensors.

[0036] In this embodiment, vibration sensors are devices used to collect vibration signals from the generating unit, such as accelerometers, velocity sensors, and displacement sensors. These sensors convert the physical quantities of vibration (displacement, velocity, acceleration) of the unit components into readable and analyzable electrical signals, making them a key tool for acquiring vibration monitoring data. The target hydropower station unit is the generating unit corresponding to the target hydropower station, such as a 300MW mixed-flow turbine unit. It is the object on which the vibration sensors are installed and also the source carrier of the vibration monitoring data. The vibration monitoring data consists of raw data and pre-processed data representing the vibration state of the vibration monitoring target, collected by the vibration sensors. This includes vibration amplitude, vibration frequency, phase, and spectral characteristics, and forms the data foundation for operating condition analysis and fault diagnosis.

[0037] S104: Obtain the current operating conditions of the unit and the vibration characteristic data in the vibration monitoring data during monitoring, and calculate the deviation between the unit's operating conditions and the preset standard operating conditions.

[0038] In this embodiment, the current operating condition of the unit refers to the actual operating status parameters of the unit at the same moment the vibration data is collected. These parameters mainly include: active load, reactive load, head, speed, guide vane opening, excitation current, cooling water temperature, and lubricating oil temperature. They are used to characterize whether the unit is currently under full load, no load, phase adjustment, start-up / shutdown, or in a transitional process. Vibration characteristic data are feature quantities extracted from the original vibration monitoring data that can represent the health status of components, such as: vibration amplitude, spectral characteristic frequency, harmonic amplitude, kurtosis, margin index, phase difference, and vibration trend change rate. The preset standard operating condition is a benchmark operating condition determined by the hydropower station design documents, manufacturer's instructions, or long-term stable operating data. It is the stable operating condition of the unit under rated load, rated head, and rated speed, used as a reference standard to judge whether the operating condition is abnormal. The operating condition deviation is a comprehensive index used to represent the degree of difference between the current operating condition and the standard operating condition. It is obtained by normalizing and weighting parameters such as load, head, and speed. The larger the value, the more unstable the operating condition and the more likely it is to cause abnormal vibrations.

[0039] S105: If the deviation of the operating condition is greater than the preset deviation threshold, the health status assessment value of the unit components in the vibration monitoring data is obtained, and the monitoring data in the vibration monitoring data in which the health status assessment value of the unit components is less than the first health threshold is used as the second vibration monitoring data.

[0040] In this embodiment, the preset deviation threshold is a critical value set to determine whether the operating condition is abnormal. It is determined by the unit design specifications and historical operating experience. If the deviation from the operating condition is greater than the preset deviation threshold, it indicates that the unit's operating condition is unstable, and abnormal vibration data needs to be investigated. The preset deviation threshold is determined by cross-validation. For example, some labeled cases of stable operation and abnormal operating conditions of the unit are randomly selected from the historical database. The anomaly identification accuracy and fault missed detection rate under different deviation thresholds are simulated. The minimum deviation value that makes the abnormal operating condition identification accuracy meet the fault diagnosis requirements and ensures that no effective fault signals are missed is selected as the preset deviation threshold.

[0041] In this embodiment, the health status assessment value of the unit component is a numerical value representing the health level of the vibration monitoring target, calculated based on vibration characteristic data using a preset assessment model. The smaller the value, the worse the component's health status and the greater the risk of failure. A first health threshold is a set critical value used to determine whether the health status of a unit component is abnormal. If the health status assessment value is less than the first health threshold, it indicates that the corresponding component has a potential fault or has already experienced a minor fault. The first health threshold is determined through a component deterioration trend curve. For example, historical health assessment values ​​are trend-fitted, and the assessment value corresponding to the inflection point where the deterioration rate abruptly changes is selected as the first health threshold.

[0042] Specifically, the preset evaluation model adopts a health status assessment network structure based on feature matching and multi-level feature fusion, which is divided into three layers: an input feature preprocessing layer, a multi-dimensional feature matching layer, and a health status output layer. The input feature preprocessing layer is used to normalize and align the vibration time-domain features, frequency-domain features, and operating condition features. The multi-dimensional feature matching layer has multiple sub-matching units, corresponding to power frequency features, harmonic features, harmonic features, and trend features, respectively. It realizes the matching and comparison of real-time features with benchmark features through similarity calculation. The health status output layer adopts a fully connected mapping structure, which comprehensively maps the matching similarity to a health status evaluation value in the range of 0-1. The smaller the value, the worse the health of the component and the higher the risk of failure.

[0043] The training process of the pre-defined evaluation model consists of three steps: sample construction, feature matching benchmark learning, and mapping relationship optimization. First, a labeled sample set is constructed, including normal, sub-healthy, and faulty states. The input data includes vibration amplitude, spectral features, vibration trends, and synchronous operating parameters. Second, using the normal state samples as a benchmark, the feature centers and feature boundaries corresponding to each health level are learned. Finally, the mapping relationship is optimized through supervised learning of labeled samples to ensure that the output health assessment value is consistent with the actual health level. The input to the pre-defined evaluation model consists of time-domain features, frequency-domain features, and operating condition feature vectors extracted from vibration monitoring data. The output of the pre-defined evaluation model is a continuous health state assessment value within the range of 0-1. This assessment value is directly compared with the first health threshold to select the second vibration monitoring data.

[0044] In this embodiment, the second vibration monitoring data is a subset of abnormal data selected from the original vibration monitoring data that meets the conditions where the deviation from the operating condition is greater than a preset deviation threshold and the corresponding component health status assessment value is less than the first health threshold. This subset is the object of subsequent fault feature extraction and diagnosis correction (excluding invalid data that is faultless or abnormal).

[0045] S106: Extract fault features and assess the quality of the second vibration monitoring data to obtain the fault feature quality assessment results.

[0046] In this embodiment, fault feature extraction involves extracting features that characterize the type and severity of the fault from the second vibration monitoring data using signal processing algorithms (spectral analysis, wavelet transform, envelope demodulation, time-domain index calculation, etc.). These features include amplitude, power frequency / octave / division frequency, harmonic ratio, kurtosis, margin, and shaft center trajectory. Quality assessment evaluates the effectiveness, reliability, signal-to-noise ratio (SNR), and completeness of the extracted fault features. It determines whether the features are clear, whether they are affected by noise, and whether they are sufficient to support accurate diagnosis, avoiding misdiagnosis due to poor feature quality. The fault feature quality assessment result is a quantitative or graded conclusion, representing an evaluation of the effectiveness of the fault features. This includes: feature SNR, number of effective features, degree of distortion, confidence level, and defect type (e.g., low SNR, blurred features, missing data, severe interference), which is used to select diagnostic correction strategies later.

[0047] S107: Based on the fault characteristic quality assessment results, select the corresponding target correction strategy from the preset fault diagnosis correction strategy library to obtain the preset fault diagnosis correction strategy, and correct the fault diagnosis results of the second vibration monitoring data based on the preset fault diagnosis correction strategy to obtain the target fault diagnosis result.

[0048] In this embodiment, the preset fault diagnosis correction strategy library is a pre-established set of strategies for correcting diagnostic biases, including: low signal-to-noise ratio correction strategy, data missing completion strategy, transfer learning correction strategy, expert rule constraint strategy, and operating condition adaptation correction strategy. The target correction strategy is one or more correction strategies best suited to the current data situation, matched from the strategy library based on the fault feature quality assessment results. The fault diagnosis result of the second vibration monitoring data is the result obtained by directly performing a preliminary diagnosis on the abnormal data before correction, which may have problems such as bias, misjudgment, and low confidence. The target fault diagnosis result is the final accurate, reliable, and credible fault diagnosis conclusion obtained after optimization by the target correction strategy, including the fault location, fault type, fault severity, and handling suggestions.

[0049] As can be seen from the above, this application, by first determining the monitoring target and the data collection and diagnosis strategy, and then screening abnormal vibration data based on the deviation of operating conditions and health status, can effectively eliminate invalid data under normal operating conditions and reduce diagnostic interference; by using fault feature quality assessment to adaptively select correction strategies to optimize diagnostic results, it improves the accuracy and reliability of vibration fault diagnosis of hydropower station units, and realizes refined and intelligent fault diagnosis and health management.

[0050] In one embodiment of this application, based on the unit characteristics and operational requirements of the target hydropower station, the corresponding vibration monitoring target is determined, including:

[0051] Obtain the vibration monitoring requirements corresponding to the target hydropower station. The vibration monitoring requirements include monitoring accuracy requirements, monitoring real-time requirements, and monitoring cost constraints.

[0052] Based on the vibration monitoring requirements, basic information of candidate monitoring components is obtained from the unit archive database. The basic information includes historical fault records, component structural parameters, material fatigue characteristics, and operating stress data.

[0053] Based on the monitoring accuracy requirements, components with a fault frequency greater than a preset frequency threshold in historical fault records are selected from the candidate monitoring components as the first candidate components.

[0054] Based on the real-time monitoring requirements, the feasibility of monitoring the first candidate component is evaluated, and the component whose signal acquisition capability is greater than the preset acquisition threshold is selected as the second candidate component.

[0055] Based on the monitoring cost constraint, the sensor deployment cost of the second candidate component is evaluated, and the component whose estimated monitoring cost is less than the preset cost threshold is selected to obtain the selected component.

[0056] The selected components are identified as the final vibration monitoring targets, and corresponding component identification information is generated for each vibration monitoring target.

[0057] In this embodiment, vibration monitoring requirements refer to the comprehensive requirements put forward by users or hydropower stations for unit vibration monitoring, including monitoring accuracy requirements, real-time monitoring requirements, and monitoring cost constraints. Monitoring accuracy requirements include the amplitude and frequency measurement error range; real-time monitoring requirements include the acquisition frequency and diagnostic delay; monitoring cost constraints include the number of sensors, hardware investment, and upper limits on operation and maintenance costs. The unit archive database is a database storing information on the entire life cycle of the target hydropower station unit, including data on the structure, materials, historical faults, maintenance records, operating stress, fatigue characteristics, etc., of each component. Candidate monitoring components are components within the unit that are prone to vibration faults and are initially included in the monitoring scope, such as guide bearings, thrust bearings, main shafts, rotors, stators, frames, and runners. Basic information is data used to evaluate the monitoring value of components, including: historical fault records, component structural dimensions, material fatigue limits, operating stress distribution, and maintenance data over the years.

[0058] In this embodiment, the fault frequency is the number or frequency of failures of a component during its historical operation, used to determine whether the component is a high-risk critical component. The preset frequency threshold is a set critical value for the frequency of fault occurrence; values ​​exceeding this threshold indicate a high incidence of component failures and should be prioritized for monitoring. The preset frequency threshold is determined by combining historical fault statistics with operational and maintenance requirements. For example, the average annual number of faults for each component is statistically analyzed from the unit's historical fault database, and a fault frequency distribution curve is plotted. Based on the hydropower station's sensitivity requirements for critical component fault early warning, the minimum fault frequency value that can cover more than 90% of actual fault events is selected as the preset frequency threshold.

[0059] In this embodiment, the first candidate component is a high-failure-frequency component selected from the candidate components according to the monitoring accuracy requirements. Monitoring feasibility assessment is an evaluation process that determines whether the component's location is convenient for sensor installation, whether the signal is easily interfered with, and whether the vibration signal can be effectively acquired. Signal acquireability is an indicator that measures whether the sensor can acquire a stable, clear, and low-noise vibration signal at the component's location. The preset acquireability threshold is the minimum standard for determining whether the signal can be effectively acquired; only signals exceeding this preset acquireability threshold can proceed to the next round of selection. The preset acquireability threshold is determined through on-site measurement point environmental statistics and signal transmission verification. For example, actual measurements are performed on the measurement points of each candidate component of the unit, considering spatial installation conditions, electromagnetic interference intensity, and vibration transmission attenuation characteristics. The success rate of acquiring effective signals and the diagnostic availability rate under different acquireability thresholds are simulated, and the lowest acquireability score that ensures stable vibration signal acquisition without feature distortion is selected as the preset acquireability threshold.

[0060] In this embodiment, the second candidate component is a component that meets the signal acquisition requirement based on the first candidate component. Sensor deployment cost assessment estimates the costs required for sensor model, quantity, wiring, installation, commissioning, and subsequent maintenance. A preset cost threshold is the upper limit of the allowable monitoring cost; if this limit is exceeded, monitoring of that component is abandoned. The preset cost threshold is determined using the annual operation and maintenance budget allocation method for hydropower stations. For example, based on the total annual budget for equipment monitoring and fault diagnosis of the hydropower station, the cost limit for vibration monitoring of a single unit is allocated proportionally according to the importance of the unit, the number of monitoring points, the sensor type, and subsequent operation and maintenance costs. This cost limit is used as the preset cost threshold. The selected component is the final component that simultaneously meets the requirements of high fault frequency, signal acquisition capability, and controllable cost. Component identification information is a unique number / name assigned to each monitoring target, such as upper guide bearing-01, main shaft-02, facilitating subsequent data management and diagnostic positioning.

[0061] As can be seen from the above, this embodiment can prioritize key components that are prone to failure, easy to collect data, and low in cost by screening monitoring components step by step from multiple dimensions such as monitoring accuracy, real-time performance, and cost constraints. This can optimize sensor deployment, reduce system redundancy and investment, and improve the rationality of monitoring resource allocation while meeting monitoring needs.

[0062] In one embodiment of this application, a corresponding vibration data acquisition and diagnosis strategy is determined based on the vibration monitoring target, including:

[0063] Acquire multi-source state data of the vibration monitoring target, including historical vibration data, historical operating condition data, maintenance record data, and data on similar unit failure cases;

[0064] Based on historical vibration data, identify the vibration baseline characteristics, typical fault spectrum characteristics, and vibration trend change patterns of the vibration monitoring target.

[0065] Based on historical operating data, the vibration response characteristics of the vibration monitoring target under different loads, heads and speeds are extracted to identify sensitive areas under operating conditions.

[0066] Based on maintenance record data, the correlation between the failure modes and maintenance cycles of vibration monitoring targets is analyzed to determine the types of common failures.

[0067] Based on data from similar unit failure cases, the failure distribution patterns and characteristic frequencies of similar structural components of the vibration monitoring target were statistically analyzed.

[0068] Based on the characteristics of the vibration baseline, the sensitive areas of the operating conditions, and the types of faults that are prone to occur, a preliminary set of diagnosable faults is determined.

[0069] Based on the vibration trend change law, fault distribution law and characteristic frequency, the fault diagnosability assessment of the vibration monitoring target is carried out to obtain the fault diagnosability assessment results.

[0070] Based on the set of diagnosable faults and the results of fault diagnosability assessment, the corresponding vibration data acquisition and diagnosis strategies are determined.

[0071] In this embodiment, multi-source state data is a collection of various relevant data gathered around the vibration monitoring target (e.g., a specific upper guide bearing or main shaft). This data supports the formulation of acquisition and diagnostic strategies from different dimensions, avoiding the limitations of single data sources. Vibration history data consists of vibration-related data collected during the past operation of the vibration monitoring target, including vibration amplitude, frequency, phase, and spectral curves, which forms the basis for identifying vibration patterns. Operating condition history data records parameter data of the vibration monitoring target under different operating conditions, including load, head, speed, and guide vane opening, used to analyze the impact of operating conditions on vibration. Maintenance record data includes past fault repairs and periodic maintenance records of the vibration monitoring target, including fault occurrence time, fault type, maintenance plan, and maintenance cycle. Similar unit fault case data is a collection of fault cases of corresponding monitoring components from other hydropower stations with similar unit models and structures to the target hydropower station (e.g., a 300MW mixed-flow turbine unit), including fault characteristics, causes, and handling methods.

[0072] In this embodiment, the vibration baseline characteristics are the standard vibration characteristics of the vibration monitoring target under normal and stable operating conditions, such as the vibration amplitude range and spectral reference curve under normal operating conditions, which serve as the benchmark for judging whether the vibration is abnormal. Typical fault spectral characteristics are the unique features of the vibration signal spectrum when a specific fault occurs in the vibration monitoring target; for example, a loose bearing corresponds to a specific octave, and a bent spindle corresponds to an abnormal power frequency amplitude, used for fault type identification. The vibration trend change pattern is the pattern of change of the vibration characteristics (amplitude, frequency) of the vibration monitoring target with operating time and conditions (e.g., the vibration amplitude gradually increases with the number of years of operation). The vibration response characteristics are the pattern of change of the vibration signal of the vibration monitoring target under different loads, heads, speeds, and other operating conditions (e.g., when the load increases, the vibration amplitude in a certain frequency band increases significantly).

[0073] In this embodiment, the sensitive operating condition area is the operating condition range where the vibration characteristics of the vibration monitoring target change significantly and abnormal vibrations are prone to occur (e.g., when the load is 200-250MW, the vibration sensitivity of the main shaft increases significantly). Failure mode refers to the specific manifestation and generation mechanism of the failure of the vibration monitoring target (e.g., bearing failure modes include abnormal clearance, wear, and loosening). Maintenance cycle correlation is the correspondence between the failure time of the vibration monitoring target and the maintenance cycle (e.g., the bearing failure rate decreases within 6 months after maintenance). Common failure types are the most frequent types of failures occurring in the vibration monitoring target based on maintenance records (e.g., upper guide bearings are prone to abnormal clearance, and the main shaft is prone to bending deformation). Similar structural components are components with the same or similar structure, material, stress mode, and operating environment as the target vibration monitoring target, such as the upper guide bearing of the same model unit. Failure distribution pattern refers to the distribution characteristics of the failure frequency, failure time period, and failure severity of similar structural components in similar units.

[0074] In this embodiment, the diagnosable fault set is a preliminary set of all fault types (e.g., spindle bending, abnormal bearing clearance, etc.) that can be identified later through vibration data diagnosis, based on the vibration baseline, sensitive operating area, and common fault types. Fault diagnosability assessment is the process of determining whether various faults in the diagnosable fault set can be effectively detected, identified, and warned against; it assesses the identifiability of fault characteristics and the lead time for warnings. The results of the fault diagnosability assessment are quantitative or graded conclusions, including the detectability probability of various faults, the lead time for warnings, and the comprehensive diagnosability index, used to optimize the acquisition and diagnosis strategy. The vibration data acquisition and diagnosis strategy is the final, complete scheme adapted to the vibration monitoring target, including acquisition strategies (sensor deployment, acquisition frequency, etc.) and diagnosis strategies (feature extraction methods, fault identification standards, etc.).

[0075] As can be seen from the above, this embodiment fully explores the vibration baseline, operating condition sensitivity and fault distribution patterns by integrating multi-source historical data and similar unit cases, making the vibration acquisition and diagnosis strategy more in line with the actual characteristics of the unit, improving the coverage and pertinence of fault identification, and avoiding blind acquisition and ineffective diagnosis.

[0076] In one embodiment of this application, based on the vibration trend change law, fault distribution law, and characteristic frequency, a fault diagnosability assessment is performed on the vibration monitoring target to obtain the fault diagnosability assessment result, including:

[0077] Based on the vibration trend change pattern, trend feature indicators are extracted, including vibration amplitude change rate, characteristic frequency offset and harmonic component evolution rate.

[0078] Based on the fault distribution pattern and characteristic frequency, and according to the dynamic characteristics of the vibration monitoring target, a fault feature identifiability matrix is ​​established. The fault feature identifiability matrix is ​​used to characterize the degree of separation between the characteristic frequencies of various faults and the background noise spectrum.

[0079] By performing correlation analysis between trend characteristic indicators and fault characteristic identifiability matrix, the detectability probability of each fault type under preset operating conditions is calculated, and the first diagnostic index is obtained.

[0080] Based on the failure distribution pattern, the probability of occurrence of each failure type in different operating stages in historical failure records is statistically analyzed. Based on the predicted value of the remaining service life of the components, the predictable advance of each failure type is calculated to obtain the second diagnostic index.

[0081] Based on the coupling relationship between the characteristic frequency and the unit's natural frequency, the risk coefficient of resonance caused by each fault type is calculated, and the third diagnostic index is obtained.

[0082] A pre-defined comprehensive diagnosticability evaluation method is used to integrate the first, second, and third diagnosticability indicators to obtain a comprehensive diagnosticability index for each fault type.

[0083] Based on the comprehensive diagnosability index, the diagnosability classification of each fault type of the vibration monitoring target is carried out to obtain the diagnosability classification results.

[0084] The diagnosticability grading results and the comprehensive diagnosticability index for each fault type are used as the results of the fault diagnosticability assessment.

[0085] In this embodiment, the vibration trend change pattern refers to the pattern of vibration characteristics (amplitude, frequency) of the vibration monitoring target changing with operating time and conditions (e.g., the annual increase in vibration amplitude of the upper guide bearing is 0.03 mm / s). Trend characteristic indicators are parameters extracted from the vibration trend change pattern to quantify the vibration deterioration trend, used to determine the fault development speed, and include three categories: vibration amplitude change rate: the amount of change in vibration amplitude per unit time, e.g., mm / s / year; characteristic frequency offset: the deviation between the fault characteristic frequency and the standard characteristic frequency, e.g., Hz; harmonic component evolution rate: the rate of change of harmonic amplitude over time, e.g., mm / s / month. The fault distribution pattern refers to the distribution characteristics of the frequency, time period, and severity of faults in similar structural components of similar units. Characteristic frequencies are the unique spectral frequencies corresponding to various types of faults.

[0086] In this embodiment, dynamic characteristics refer to the mechanical properties of the vibration monitoring target itself, including stiffness, damping, natural frequency, and vibration transmission path, which determine the transmission and attenuation patterns of fault vibration signals. The fault feature identifiability matrix is ​​a two-dimensional matrix (rows: fault type, columns: measurement point location), used to quantify the degree of separation between the characteristic frequencies of various faults and the background noise spectrum. The larger the value, the easier the fault features are to identify, effectively determining the detectability of weak faults. The background noise spectrum is the vibration spectrum generated by non-fault factors (hydraulic noise, electromagnetic noise, environmental interference) during unit operation, which can interfere with the identification of fault features and serves as the reference benchmark for the identifiability matrix. Correlation analysis combines trend characteristic indicators (fault development trend) with the identifiability matrix (fault feature clarity) to analyze their correlation and determine the degree of identifiability of features during fault development.

[0087] In this embodiment, the preset operating conditions are common operating conditions of the unit (e.g., rated load 300MW, rated head 120m, rated speed 300r / min, and sensitive operating areas) used to calculate the detectability probability of faults under different operating conditions. The detectability probability is the probability (value 0-1) that a certain type of fault can be effectively detected by vibration data under the preset operating conditions. The higher the probability, the easier it is to detect the fault in a timely manner, which is the core of the first diagnosticability index. The first diagnosticability index quantifies the detectability of faults; it is the detectability probability of each fault type, indicating whether the fault can be effectively identified. The operating phase refers to different periods of unit operation, such as the startup phase, stable operation phase, initial post-maintenance phase, and long-term operation phase. The probability of fault occurrence varies in different phases. The predicted remaining service life of a component is based on the component's material fatigue characteristics, operating stress, and vibration trends, predicting how long the component can still operate normally; for example, the remaining service life of the upper guide bearing is 2 years. The early warning lead time is the time interval from detecting fault characteristics to the fault developing to the point of affecting the normal operation of the unit; it is the second diagnosticability index, representing the early warning capability of the fault.

[0088] In this embodiment, the second diagnosticability indicator is a quantification of fault predictability, representing the lead time for each fault type, indicating whether the fault can be detected in advance and allow for processing time. The unit's natural frequency is its inherent vibration frequency (determined by structure and materials), an inherent property of the unit. The coupling relationship is the interaction between the fault's characteristic frequency and the unit's natural frequency; for example, when the characteristic frequency is close to the natural frequency, resonance is easily triggered. The resonance risk coefficient is the probability (value 0-1) that the characteristic frequency of a certain type of fault will cause unit resonance; the higher the coefficient, the greater the resonance risk, and this is the core of the third diagnosticability indicator. The third diagnosticability indicator is a quantification of fault risk level, representing the resonance risk coefficient for each fault type, indicating the severity and urgency of the fault.

[0089] In this embodiment, the preset comprehensive diagnosticability evaluation method is a method for integrating three diagnosticability indicators. This embodiment adopts a weighted summation method, assigning different weights to the three indicators according to the operation and maintenance needs of the hydropower station, and calculating a comprehensive index. The comprehensive diagnosticability index is a comprehensive numerical value (ranging from 0 to 1) quantifying the diagnosticability of a certain type of fault after integrating the first, second, and third diagnosticability indicators. The higher the value, the stronger the detectability, predictability, and risk controllability of the fault. Diagnosticability grading is based on the comprehensive diagnosticability index to classify the diagnosticability of each fault type into levels (e.g., excellent, good, average, poor), facilitating subsequent targeted optimization of diagnostic strategies. The diagnosticability grading result is the conclusion of the diagnosticability level classification for each fault type (e.g., abnormal bearing clearance is good). The fault diagnosticability assessment result is the final output assessment conclusion, including the comprehensive diagnosticability index and diagnostic grading result for each fault type, used to determine the vibration data acquisition and diagnostic strategies.

[0090] As can be seen from the above, this embodiment constructs a comprehensive evaluation system through multi-dimensional diagnosticability indicators, and conducts quantitative and graded assessments of each fault type. This clarifies the detectability, predictability, and risk level of each fault, providing a scientific basis for the formulation of diagnostic strategies and enhancing the ability to predict and identify faults early.

[0091] In one embodiment of this application, based on the fault distribution pattern and characteristic frequency, and according to the dynamic characteristics of the vibration monitoring target, a fault feature identifiability matrix is ​​established, including:

[0092] Obtain the background noise spectrum of the vibration monitoring target. The background noise spectrum includes the spectral distribution of hydraulic noise, mechanical noise and electromagnetic noise.

[0093] Based on the fault distribution pattern, the characteristic frequency set corresponding to each type of fault is determined. The characteristic frequency set includes power frequency, harmonics, sub-harmonics and high-frequency natural frequency components.

[0094] Based on the dynamic characteristics of the vibration monitoring target, a vibration transmission path model is established, and the amplitude gain of the transfer function of each measuring point with respect to different characteristic frequency components is calculated.

[0095] Based on the transfer function amplitude gain, the theoretical amplitudes of each frequency component in the characteristic frequency set are corrected to obtain the expected observable amplitudes of each fault type at each measurement point.

[0096] Calculate the ratio of the expected observable amplitude to the noise amplitude at the corresponding frequency position in the background noise spectrum to obtain the characteristic signal-to-noise ratio of each fault type at each measurement point;

[0097] Arrange the characteristic signal-to-noise ratios of each fault type at each measuring point according to the fault type and the location of the measuring point to generate a two-dimensional matrix structure, and obtain the fault feature identifiability matrix.

[0098] In this embodiment, the background noise spectrum is the inherent non-fault noise spectrum during normal operation of the unit, including: hydraulic noise: noise caused by water flow, eddies, and pressure pulsations; mechanical noise: noise caused by structural friction, loose connections, and normal mechanical vibration; electromagnetic noise: noise generated by the electromagnetic excitation of the generator, used to compare with fault characteristics to determine whether the signal is submerged in noise. The characteristic frequency set is a set of typical frequencies corresponding to a certain fault, including: power frequency: the unit's rotational frequency (1X); harmonics: 2X, 3X, 4X, etc.; sub-frequency divisions: 0.4X, 0.5X, etc.; high-frequency inherent frequencies: the resonant frequencies of the component's own structure.

[0099] In this embodiment, the vibration transmission path model describes the transmission process of vibration from the fault source (e.g., inside the bearing) through components, frame, and base to the sensor, and is used to calculate signal attenuation and amplification. The transfer function amplitude gain is the factor by which a certain frequency component is amplified or attenuated when the vibration travels from the fault source to the measurement point, indicating the sensitivity of the measurement point to that frequency. The theoretical amplitude is the ideal amplitude of the fault characteristic frequency without considering transmission attenuation. The expected observable amplitude is the fault characteristic amplitude that can actually be measured at the sensor location after transmission path attenuation / amplification correction. The characteristic signal-to-noise ratio (SNR) is the expected observable amplitude divided by the noise amplitude at the corresponding frequency; the larger the ratio, the clearer and easier to identify the fault characteristics. The two-dimensional matrix structure includes: rows: different fault types; columns: different measurement point locations; and matrix values: corresponding characteristic SNRs, forming an identifiability matrix.

[0100] The vibration transmission path model is a multi-degree-of-freedom dynamic model adapted to the structural characteristics of vibration monitoring targets (bearings, main shafts, etc.) in hydropower stations. It is divided into three levels: the fault source excitation layer, the transmission path layer, and the measurement point response layer. The fault source excitation layer simulates vibration excitation signals generated by various faults (abnormal bearing clearance, main shaft bending, etc.), corresponding to power frequency and harmonic frequencies in the characteristic frequency set. The transmission path layer, based on the actual structure of the monitoring target, is decomposed into transmission units such as bearing housings, frames, and bases, simulating the transmission process of vibration from the fault source to the measurement point. The measurement point response layer corresponds to the measurement point locations of each sensor, outputting the vibration response signal after transmission attenuation / amplification, providing a basis for subsequent calculation of the transfer function amplitude gain. The various levels of the vibration transmission path model are correlated through dynamic parameters, ensuring that the vibration transmission process is consistent with the actual operating state of the unit.

[0101] As can be seen from the above, this embodiment, by establishing a fault feature identifiability matrix, quantifies the degree of separation between fault feature frequency and noise and the transmission characteristics of measurement points, can accurately determine the observability of fault signals, improve the ability to extract weak fault features, and reduce the probability of missed diagnosis and misdiagnosis.

[0102] In one embodiment of this application, based on the fault feature quality assessment results, a corresponding target correction strategy is selected from a preset fault diagnosis correction strategy library to obtain a preset fault diagnosis correction strategy, including:

[0103] Establish a mapping relationship library between diagnostic deviation types and correction strategies, which serves as a preset fault diagnosis correction strategy library;

[0104] Based on the defect type and defect severity identified in the fault characteristic quality assessment results, match the corresponding diagnostic deviation type;

[0105] Based on the type of diagnostic deviation, select the corresponding target correction strategy from the preset fault diagnosis correction strategy library;

[0106] Based on the signal-to-noise ratio characteristics, operating condition deviation, and component health status values ​​of the second vibration monitoring data, the target correction strategy is evaluated to obtain the final preset fault diagnosis correction strategy.

[0107] In this embodiment, the diagnostic deviation type refers to the type of diagnostic error caused by data or feature issues, such as: low signal-to-noise ratio deviation, feature missing deviation, operating condition deviation, noise interference deviation, and minor fault omission deviation. The mapping relationship library is a rule base that binds diagnostic deviation types to corresponding correction strategies one-to-one, enabling automatic matching. The defect type refers to the specific problem with the fault feature, such as insufficient signal-to-noise ratio, missing high-frequency features, data mutation, spectral ambiguity, and severe interference. The defect severity is the level of severity of the defect, such as minor, moderate, or severe, used to determine the appropriate correction strategy. The target correction strategy is one or more of the most suitable correction strategies matched from the strategy library based on the defect type and severity.

[0108] In this embodiment, the signal-to-noise ratio (SNR) is the ratio of the fault characteristic amplitude to the noise amplitude, representing the clarity of the signal. The operating condition deviation is the degree of difference between the current operating condition and the standard operating condition, used to determine the impact of the operating condition on the diagnosis. The component health status value is a quantitative score of the component's current health level, used to determine the stage of fault development. The final preset fault diagnosis correction strategy is a final strategy determined after comprehensive evaluation and that can be directly used to correct the diagnostic results.

[0109] As can be seen from the above, this embodiment achieves adaptive selection of correction methods by using fault feature quality matching diagnosis deviation and correction strategy, making diagnosis correction more targeted and improving the robustness of diagnosis results under complex working conditions.

[0110] In one embodiment of this application, the fault diagnosis result of the second vibration monitoring data is corrected based on a preset fault diagnosis correction strategy to obtain a target fault diagnosis result, including:

[0111] The type of the final preset fault diagnosis and correction strategy is determined. The preset fault diagnosis and correction strategy includes transfer learning correction strategy and expert rule fusion correction strategy.

[0112] If the preset fault diagnosis correction strategy is a transfer learning correction strategy, then the preset source domain model transfer method is used to correct the fault diagnosis results of the second vibration monitoring data to obtain the target fault diagnosis results.

[0113] If the preset fault diagnosis correction strategy is an expert rule fusion correction strategy, then the preset expert rule reasoning method is used to correct the fault diagnosis results of the second vibration monitoring data to obtain the target fault diagnosis results.

[0114] In this embodiment, the correction strategy falls into two main categories of preset fault diagnosis correction strategies, which are flexibly selected based on data conditions and diagnostic needs. These two categories include: transfer learning correction strategy and expert rule fusion correction strategy. The transfer learning correction strategy utilizes mature diagnostic models from similar generating units (source domain) to adapt to the vibration data of the target hydropower station (target domain), addressing the issues of insufficient data and poor diagnostic generalization ability for the target generating unit. The expert rule fusion correction strategy uses diagnostic rules developed based on the experience of experts in the hydropower field to constrain and verify the preliminary diagnostic results, improving the engineering practicality and reliability of the diagnostic results.

[0115] In this embodiment, the source domain model transfer method is an implementation of the transfer learning correction strategy. It reuses pre-trained diagnostic models from similar generating units (source domain), adjusts model parameters to adapt to the vibration data and unit characteristics of the target hydropower station, and achieves accurate diagnosis and correction. The expert rule reasoning method is an implementation of the expert rule fusion correction strategy. It calls a preset set of expert rules, such as rules corresponding to fault characteristics and fault types, and rules relating operating conditions and faults, to perform logical reasoning, verification, and correction on the preliminary diagnostic results.

[0116] As can be seen from the above, this embodiment, by setting two correction methods—transfer learning and expert rule fusion—can flexibly switch according to data conditions, taking into account both the accuracy of data-driven diagnosis and the reliability of expert knowledge, thus having a wider range of applicable scenarios and stronger fault tolerance.

[0117] In one embodiment of this application, if the preset fault diagnosis correction strategy is a transfer learning correction strategy, then the fault diagnosis result of the second vibration monitoring data is corrected using a preset source domain model transfer method to obtain the target fault diagnosis result, including:

[0118] Acquire historical fault diagnosis data and pre-trained diagnostic models from source hydropower stations with similar turbine models and operating conditions to the target hydropower station;

[0119] Extract network layer parameters from the pre-trained diagnostic model used to extract general features of vibration signals to obtain a transferable feature extractor;

[0120] Using the portion of the second vibration monitoring data with a confidence level greater than a preset confidence threshold, the classification layer after the transferable feature extractor is adjusted and trained to obtain a corrected diagnostic model adapted to the target hydropower station.

[0121] The second vibration monitoring data is input into the modified diagnostic model to obtain the modified fault type probability distribution.

[0122] If the maximum probability value in the corrected fault type probability distribution is greater than the preset reliability threshold, then the corresponding fault type will be used as the corrected fault diagnosis result.

[0123] If a pre-trained diagnostic model cannot be obtained, or if the maximum probability value in the corrected fault type probability distribution is less than or equal to the preset reliability threshold, then the vibration amplitude and spectrum characteristics of the second vibration monitoring data are matched and verified with the preset fault mode library to obtain the fused fault diagnosis result, which is used as the target fault diagnosis result.

[0124] In this embodiment, the target hydropower station is the specific hydropower station currently undergoing vibration fault diagnosis, serving as the adaptation object and final application object for the modified diagnostic model. The source hydropower station refers to other hydropower stations with the same / similar turbine models and similar operating conditions (load, head, speed range) as the target hydropower station. Their unit fault data and diagnostic models can be used as reference templates for migration and reuse. Historical fault diagnosis data, collected from the source hydropower station in the past, includes historical data on unit fault types, vibration characteristics, operating parameters, and diagnostic results, forming the training basis for the pre-trained diagnostic model. The pre-trained diagnostic model is a model pre-trained based on the historical fault diagnosis data of the source hydropower station, capable of being directly used for fault identification. This embodiment employs a deep learning model, possessing mature vibration feature extraction and fault classification capabilities.

[0125] The pre-trained diagnostic model employs a deep learning network structure adapted to the vibration fault diagnosis scenario of hydropower stations. It is divided into four layers: an input layer, a general feature extraction layer, a feature fusion layer, and a fault classification layer. These layers are interconnected through parameters to form a complete diagnostic link. The input layer receives multi-dimensional feature data from vibration signals, standardizing the data format. The general feature extraction layer, consisting of three convolutional sub-layers and two pooling sub-layers, extracts common vibration features (power frequency, harmonics, and time-domain statistical features) shared by different types of hydropower units, serving as the core module for model transferability. The feature fusion layer uses a fully connected structure to fuse the extracted multi-dimensional general features, eliminating feature redundancy. The fault classification layer uses a Softmax classification structure, outputting the predicted probability of various faults to complete fault type identification. The close connection between these layers ensures the accuracy of feature extraction and fault classification.

[0126] In this embodiment, the common features are those shared by similar generating units in different hydropower stations and can be used for fault identification, such as the harmonic characteristics of bearing faults and the power frequency characteristics of main shaft faults, and are not limited to a specific hydropower station. The network layer parameters are model parameters used in the pre-trained diagnostic model to extract common features of vibration signals, such as the convolutional and pooling layer parameters of a deep learning model; these are the core transferable parts of the model. The transferable feature extractor is a module extracted from the pre-trained diagnostic model that is solely responsible for extracting common features of vibration signals. It can be directly reused without retraining, reducing training costs.

[0127] In this embodiment, the second vibration monitoring data is a subset of selected abnormal vibration data, such as abnormal vibration data from the upper guide bearing and the main shaft. This data serves as fine-tuning data and the diagnostic target for the revised diagnostic model. Confidence level is the reliability of the preliminary diagnostic result corresponding to each data point in the second vibration monitoring data, ranging from 0 to 1. Higher confidence levels indicate more reliable data, which can be used for model adjustment. The preset confidence threshold is a critical value set for selecting reliable adjustment data. Data exceeding this threshold can be used to adjust the training classification layer, ensuring the adjustment effect. The preset confidence threshold is determined based on historical diagnostic accuracy statistics. For example, the proportion of correct diagnoses corresponding to different confidence levels in historical diagnostic data is statistically analyzed, and a confidence value covering more than 95% of correct diagnostic results is selected as the preset confidence threshold.

[0128] In this embodiment, the classification layer is the module in the diagnostic model responsible for mapping the extracted vibration features to specific fault types (e.g., abnormal bearing clearance, main shaft bending). It needs to be adjusted based on the target hydropower station data to adapt to the characteristics of the target unit. Adjustment training only optimizes the parameters of the classification layer after the transferable feature extractor, without changing the general feature extraction module, achieving reuse and adaptation, and reducing training workload. The corrected diagnostic model, after adjustment based on the target hydropower station data, is a diagnostic model adapted to the vibration characteristics of the target unit and capable of accurately identifying faults in the target unit; it is a tool for transfer learning correction.

[0129] In this embodiment, the fault type probability distribution is the output of the modified diagnostic model. The probability of occurrence (summing up to 1) for each fault type (e.g., abnormal bearing clearance, loose bearing) is used to determine the most likely fault type. The maximum probability value is the highest probability value in the fault type probability distribution (i.e., the probability corresponding to the most likely fault), used to determine the reliability of the diagnostic result. The preset reliability threshold is a set critical value used to determine whether the diagnostic result is reliable. If the maximum probability value is greater than the preset reliability threshold, the diagnostic result is considered reliable. The preset reliability threshold is determined using a cross-validation method based on historical diagnostic accuracy. For example, multiple sets of typical fault samples are selected from a historical fault annotation library to simulate the fault detection rate and false alarm rate under different reliability thresholds. The probability value that satisfies the hydropower station's operation and maintenance early warning requirements and has the lowest false alarm rate is selected as the preset reliability threshold. The corrected fault diagnosis result is the fault type (the most likely fault) corresponding to the maximum probability value being greater than the preset reliability threshold; this is the preliminary corrected diagnostic conclusion.

[0130] In this embodiment, the fault mode library is a pre-established database storing typical vibration characteristics (amplitude, spectrum, and phase) corresponding to various faults, including the correspondence between fault types and vibration characteristics, used for fault matching verification. Matching verification compares the vibration amplitude and spectrum characteristics of the second vibration monitoring data with the typical characteristics of various faults in the fault mode library, calculates the matching degree, and verifies the accuracy of the preliminary diagnostic results. The fused fault diagnosis result is the final diagnostic conclusion obtained by combining the model output with the fault mode library matching results when a pre-trained model is unavailable or the corrected diagnostic results are unreliable, ensuring diagnostic reliability.

[0131] As can be seen from the above, this embodiment can achieve high-precision diagnosis even under small sample conditions by using transfer learning to reuse similar unit models and adjusting them to adapt to the target unit; the verification by matching the fault mode library further improves the reliability of the results and solves the diagnostic difficulties caused by insufficient data for new units.

[0132] In one embodiment of this application, if the preset fault diagnosis correction strategy is an expert rule fusion correction strategy, then the fault diagnosis result of the second vibration monitoring data is corrected using a preset expert rule reasoning method to obtain the target fault diagnosis result, including:

[0133] Based on the current operating condition parameters and vibration characteristics corresponding to the second vibration monitoring data, the expert rule set matching the operating condition is retrieved from the preset fault mode library.

[0134] Analyze the confidence weights and applicable conditions of each rule in the expert rule set, and establish a rule reasoning network;

[0135] Using a rule-based reasoning network, rule matching reasoning is performed on the preliminary diagnostic results of the second vibration monitoring data to obtain the rule support for each candidate fault type.

[0136] The output probability of the preliminary diagnosis result is obtained, and the output probability of the preliminary diagnosis result is fused with the rule support using the fuzzy logic method to obtain the fused fault type confidence.

[0137] The fault type with the highest confidence level is determined as the final corrected target fault diagnosis result, and the corresponding fault cause analysis and handling suggestions are output.

[0138] In this embodiment, the current operating condition parameters are the unit operating parameters synchronized with the second vibration monitoring data acquisition time, including active load, head, speed, guide vane opening, lubricating oil temperature, etc., used to match expert rules adapted to the current operating condition. Vibration characteristics are features extracted from the second vibration monitoring data, including vibration amplitude, spectral characteristics (power frequency, octaves), phase, kurtosis, etc., which are the basis for rule matching. The preset fault mode library is a pre-established database that stores typical vibration characteristics, operating conditions, and corresponding expert diagnostic rules for various unit faults, and is the basis for retrieving expert rules. The expert rule set is a set of expert diagnostic rules retrieved from the fault mode library that match the current operating condition parameters. Each rule corresponds to the operating condition + vibration characteristics to obtain the logical relationship of fault type. For example, a load of 200-250MW + an increase in the amplitude of the upper guide bearing at 2X octaves indicates abnormal bearing clearance.

[0139] In this embodiment, the confidence weight is a weight value assigned to each rule in the expert rule set, ranging from 0 to 1. The weight is determined by the rule's reliability, expert acceptance, and historical application accuracy; the higher the weight, the greater the rule's reference value. The applicability condition refers to the applicable operating condition range for each expert rule. A rule can only be used for reasoning when the current operating condition and vibration characteristics meet the applicability condition. For example, a rule may only apply to operating conditions with a load of 200-250MW and a speed of 300-305r / min. The rule reasoning network is an interconnected and collaboratively reasoning network structure built from all rules in the expert rule set according to the logical relationship between operating condition, characteristic, and fault. This enables multi-rule joint reasoning, avoiding the limitations of a single rule. The preliminary diagnostic result is the fault conclusion obtained by directly diagnosing the second vibration monitoring data before expert rule correction.

[0140] In this embodiment, rule-matching reasoning utilizes a rule-based reasoning network to match the current operating parameters and vibration characteristics with each rule in the expert rule set, determining whether the rule's applicability conditions are met, and thus obtaining the degree of support each rule provides for each candidate fault type. Candidate fault types are all possible fault types involved in the preliminary diagnostic results (e.g., abnormal bearing clearance, loose bearing, slight spindle bending, etc.), and are the objects of rule-based reasoning. Rule support is the degree of support each expert rule provides for a particular candidate fault type, ranging from 0 to 1. A higher support indicates that the current operating conditions and vibration characteristics more closely match the fault type corresponding to that rule.

[0141] In this embodiment, the output probability is the probability of occurrence of each candidate fault type in the preliminary diagnosis results (summing up to 1), representing the preliminary diagnosis's tendency to judge each fault type. Fuzzy logic is a fusion method used to handle uncertainty and fuzziness. It can reasonably fuse the preliminary diagnosis output probability and rule support, two different dimensions of indicators, to obtain a more reliable fault confidence score. The fused fault type confidence score is the final confidence score (value 0-1) of each candidate fault type after fuzzy logic fusion, taking into account both the model output of the preliminary diagnosis and the empirical judgment of expert rules, making it more reliable than a single indicator.

[0142] In this embodiment, the final corrected target fault diagnosis result is the fault type with the highest confidence score among the fused fault types. This type is the fault conclusion that best reflects the actual situation of the unit after expert rule correction. Fault cause analysis analyzes the specific causes of the fault based on the current operating conditions, vibration characteristics, and historical unit data, targeting the finally diagnosed fault type. For example, abnormal bearing clearance might be caused by improper clearance adjustment after maintenance or long-term wear. Treatment recommendations are based on the fault type and cause analysis, providing specific treatment measures, priorities, and timelines tailored to the actual operation and maintenance of the hydropower station. For example, bearing clearance testing and tightening checks should be conducted during the next shutdown, with maintenance personnel given priority for investigation.

[0143] As can be seen from the above, this embodiment uses expert rule reasoning and fuzzy logic to initially diagnose probabilities, and makes full use of industry experience to constrain the diagnostic output, so that the fault conclusions are more in line with engineering practice, more interpretable, and easier for operation and maintenance personnel to understand and implement.

[0144] Corresponding to the fault diagnosis and health management method for hydropower station unit vibration in the above embodiment, Figure 2 This is a structural block diagram of a fault diagnosis and health management system for hydropower station unit vibration according to an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The fault diagnosis and health management system 20 for the vibration of the hydropower station units includes: target determination module 21, diagnosis strategy module 22, execution strategy module 23, operating condition calculation module 24, data filtering module 25, quality assessment module 26, and strategy selection module 27.

[0145] Among them, the target determination module 21 is used to determine the corresponding vibration monitoring target based on the unit characteristics and operating requirements of the target hydropower station;

[0146] The diagnostic strategy module 22 is used to determine the corresponding vibration data acquisition and diagnostic strategy based on the vibration monitoring target;

[0147] The execution strategy module 23 is used to execute the vibration data acquisition and diagnosis strategy and acquire vibration monitoring data of the target hydropower station unit based on the vibration sensor.

[0148] The operating condition calculation module 24 is used to acquire the current operating condition of the unit and the vibration characteristic data in the vibration monitoring data during monitoring, and to calculate the deviation of the operating condition between the unit operating condition and the preset standard operating condition.

[0149] The data filtering module 25 is used to obtain the health status assessment value of the unit components in the vibration monitoring data if the deviation of the operating condition is greater than the preset deviation threshold, and to take the part of the monitoring data in the vibration monitoring data whose health status assessment value of the unit components is less than the first health threshold as the second vibration monitoring data.

[0150] The quality assessment module 26 is used to extract fault features and assess the quality of the second vibration monitoring data to obtain the fault feature quality assessment results.

[0151] The strategy selection module 27 is used to select the corresponding target correction strategy from the preset fault diagnosis correction strategy library based on the fault feature quality assessment results, obtain the preset fault diagnosis correction strategy, and correct the fault diagnosis results of the second vibration monitoring data based on the preset fault diagnosis correction strategy to obtain the target fault diagnosis result.

[0152] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the target determination module 21, diagnostic strategy module 22, execution strategy module 23, working condition calculation module 24, data filtering module 25, quality assessment module 26, and strategy selection module 27 are shown.

[0153] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0154] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0155] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0156] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the fault diagnosis and health management method for hydropower station unit vibration provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0157] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0158] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0159] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0160] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0161] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0162] The units described as separate components may or may not be physically separate. The components shown as units 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 units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0163] Furthermore, the functional units in the various embodiments of this application 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 as a software functional unit.

[0164] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for fault diagnosis and health management of hydroelectric generating unit vibration, characterized in that, include: Based on the unit characteristics and operational requirements of the target hydropower station, the corresponding vibration monitoring targets are determined; Based on the vibration monitoring target, determine the corresponding vibration data acquisition and diagnosis strategy; The vibration data acquisition and diagnosis strategy is executed, and vibration monitoring data of the target hydropower station unit is acquired based on vibration sensors; The system acquires the current operating conditions of the unit during monitoring and the vibration characteristic data from the vibration monitoring data, and calculates the deviation between the unit's operating conditions and the preset standard operating conditions. If the deviation of the operating condition is greater than the preset deviation threshold, the health status assessment value of the unit component in the vibration monitoring data is obtained, and the monitoring data in the vibration monitoring data in which the health status assessment value of the unit component is less than the first health threshold is used as the second vibration monitoring data. Fault features are extracted and quality is assessed from the second vibration monitoring data to obtain fault feature quality assessment results. A preliminary diagnosis is then performed on the second vibration monitoring data to obtain fault diagnosis results. Based on the fault characteristic quality assessment results, a corresponding target correction strategy is selected from the preset fault diagnosis correction strategy library to obtain the preset fault diagnosis correction strategy. The fault diagnosis results of the second vibration monitoring data are then corrected based on the preset fault diagnosis correction strategy to obtain the target fault diagnosis result.

2. The method according to claim 1, characterized in that, The determination of corresponding vibration monitoring targets based on the unit characteristics and operational requirements of the target hydropower station includes: Obtain the vibration monitoring requirements corresponding to the target hydropower station, including monitoring accuracy requirements, monitoring real-time requirements, and monitoring cost constraints; Based on the aforementioned vibration monitoring requirements, basic information of candidate monitoring components is obtained from the unit archive database. This basic information includes historical fault records, component structural parameters, material fatigue characteristics, and operating stress data. Based on the monitoring accuracy requirements, components with a fault frequency greater than a preset frequency threshold in the historical fault records are selected from the candidate monitoring components as the first candidate components. Based on the real-time monitoring requirements, the feasibility of monitoring the first candidate component is evaluated, and components whose signal collectability is greater than the preset collectability threshold are selected as the second candidate component. Based on the monitoring cost constraint, the sensor deployment cost of the second candidate component is evaluated, and the component whose estimated monitoring cost is less than the preset cost threshold is selected to obtain the selected component. The selected components are identified as the final vibration monitoring targets, and corresponding component identification information is generated for the vibration monitoring targets.

3. The method according to claim 1, wherein, The step of determining the corresponding vibration data acquisition and diagnosis strategy based on the vibration monitoring target includes: Acquire multi-source state data of the vibration monitoring target, including historical vibration data, historical operating condition data, maintenance record data, and data on similar unit failure cases; Based on the historical vibration data, the vibration baseline characteristics, typical fault spectrum characteristics, and vibration trend change patterns of the vibration monitoring target are identified. Based on the historical operating data, the vibration response characteristics of the vibration monitoring target under different loads, heads and speeds are extracted to identify sensitive areas of the operating conditions. Based on the maintenance record data, the correlation between the failure modes and maintenance cycles of the vibration monitoring target is analyzed to determine the types of common failures. Based on the fault case data of the same type of unit, the fault distribution pattern and characteristic frequency of similar structural components to the vibration monitoring target were statistically analyzed. Based on the vibration baseline characteristics, the sensitive operating conditions, and the types of common faults, a preliminary set of diagnosable faults is determined. Based on the vibration trend change pattern, the fault distribution pattern, and the characteristic frequency, the fault diagnosability assessment of the vibration monitoring target is performed to obtain the fault diagnosability assessment result. Based on the set of diagnosable faults and the results of the fault diagnosability assessment, a corresponding vibration data acquisition and diagnosis strategy is determined.

4. The method according to claim 3, wherein, Based on the vibration trend change pattern, the fault distribution pattern, and the characteristic frequency, the fault diagnosability assessment of the vibration monitoring target is performed to obtain the fault diagnosability assessment results, including: Based on the vibration trend change pattern, trend feature indicators are extracted, including vibration amplitude change rate, characteristic frequency offset, and harmonic component evolution rate. Based on the fault distribution pattern and characteristic frequency, and according to the dynamic characteristics of the vibration monitoring target, a fault feature identifiability matrix is ​​established. The fault feature identifiability matrix is ​​used to characterize the degree of separation between the characteristic frequencies of various faults and the background noise spectrum. The trend feature index is correlated with the fault feature identifiability matrix to calculate the detectability probability of each fault type under preset working conditions, and the first diagnostic index is obtained. Based on the aforementioned fault distribution pattern, the probability of occurrence of each fault type in different operating stages in historical fault records is statistically analyzed. Based on the predicted value of the remaining service life of the components, the predictable advance of each fault type is calculated to obtain the second diagnostic index. Based on the coupling relationship between the characteristic frequency and the unit's natural frequency, the risk coefficient of resonance caused by each fault type is calculated to obtain the third diagnostic index. A pre-defined comprehensive diagnosticability evaluation method is used to fuse the first diagnosticability index, the second diagnosticability index, and the third diagnosticability index to obtain a comprehensive diagnosticability index for each fault type. Based on the comprehensive diagnosability index, the diagnosability classification of each fault type of the vibration monitoring target is performed to obtain the diagnosability classification results. The diagnosticability grading results and the comprehensive diagnosticability index of each fault type are used as the fault diagnosticability assessment results.

5. The method according to claim 4, wherein, Based on the fault distribution pattern and characteristic frequency, and according to the dynamic characteristics of the vibration monitoring target, a fault feature identifiability matrix is ​​established, including: Obtain the background noise spectrum of the vibration monitoring target, wherein the background noise spectrum includes the spectral distribution of hydraulic noise, mechanical noise and electromagnetic noise; Based on the fault distribution pattern, the characteristic frequency set corresponding to each type of fault is determined. The characteristic frequency set includes power frequency, harmonics, sub-harmonics, and high-frequency inherent frequency components. Based on the dynamic characteristics of the vibration monitoring target, a vibration transmission path model is established, and the amplitude gain of the transfer function of each measuring point with respect to different characteristic frequency components is calculated. Based on the transfer function amplitude gain, the theoretical amplitudes of each frequency component in the characteristic frequency set are corrected to obtain the expected observable amplitudes of each fault type at each measurement point. Calculate the ratio of the expected observable amplitude to the noise amplitude at the corresponding frequency position in the background noise spectrum to obtain the characteristic signal-to-noise ratio of each fault type at each measurement point; The signal-to-noise ratios of each fault type at each measuring point are arranged according to the fault type and the location of the measuring point to generate a two-dimensional matrix structure, thus obtaining the fault feature identifiability matrix.

6. The method according to claim 1, wherein, Based on the fault characteristic quality assessment results, a corresponding target correction strategy is selected from a preset fault diagnosis correction strategy library to obtain a preset fault diagnosis correction strategy, including: Establish a mapping relationship library between diagnostic deviation types and correction strategies, which serves as a preset fault diagnosis correction strategy library; Based on the defect type and defect severity identified in the fault characteristic quality assessment results, the corresponding diagnostic deviation type is matched; Based on the diagnostic deviation type, select the corresponding target correction strategy from the preset fault diagnosis correction strategy library; Based on the signal-to-noise ratio characteristics, operating condition deviation, and component health status values ​​of the second vibration monitoring data, the target correction strategy is evaluated to obtain the final preset fault diagnosis and correction strategy.

7. The method according to claim 6, wherein the method further comprises: The fault diagnosis result of the second vibration monitoring data is corrected based on the preset fault diagnosis correction strategy to obtain the target fault diagnosis result, including: The fault diagnosis results of the second vibration monitoring data are corrected by using a preset expert rule reasoning method to obtain the target fault diagnosis results.

8. The method for fault diagnosis and health management of hydropower station unit vibration according to claim 7, characterized in that, The step of correcting the fault diagnosis result of the second vibration monitoring data using a preset expert rule reasoning method to obtain the target fault diagnosis result includes: Based on the current operating condition parameters and vibration characteristics corresponding to the second vibration monitoring data, an expert rule set matching the operating condition is retrieved from the preset fault mode library; Analyze the confidence weights and applicable conditions of each rule in the expert rule set, and establish a rule reasoning network; Using the rule-based reasoning network, rule matching reasoning is performed on the fault diagnosis results of the second vibration monitoring data to obtain the rule support degree of each candidate fault type; The output probability of the fault diagnosis result is obtained, and the output probability of the fault diagnosis result is fused with the rule support using a fuzzy logic method to obtain the fused fault type confidence. The fault type with the highest confidence level is determined as the final corrected target fault diagnosis result, and the corresponding fault cause analysis and handling suggestions are output.

9. A fault diagnosis and health management system for vibration of hydropower station units, characterized in that, include: The target determination module is used to determine the corresponding vibration monitoring targets based on the unit characteristics and operational requirements of the target hydropower station. The diagnostic strategy module is used to determine the corresponding vibration data acquisition and diagnostic strategy based on the vibration monitoring target. The execution strategy module is used to execute the vibration data acquisition and diagnosis strategy and acquire vibration monitoring data of the target hydropower station unit based on the vibration sensor; The operating condition calculation module is used to obtain the current operating condition of the unit during monitoring and the vibration characteristic data in the vibration monitoring data, and to calculate the deviation of the operating condition between the unit operating condition and the preset standard operating condition. The data filtering module is used to obtain the health status assessment value of the unit components in the vibration monitoring data if the deviation of the operating condition is greater than a preset deviation threshold, and to take the part of the monitoring data in the vibration monitoring data whose health status assessment value of the unit components is less than a first health threshold as the second vibration monitoring data. The quality assessment module is used to extract fault features and assess the quality of the second vibration monitoring data to obtain fault feature quality assessment results, and to perform preliminary diagnosis on the second vibration monitoring data to obtain fault diagnosis results. The strategy selection module is used to select a corresponding target correction strategy from a preset fault diagnosis correction strategy library based on the fault feature quality assessment results, obtain a preset fault diagnosis correction strategy, and correct the fault diagnosis results of the second vibration monitoring data based on the preset fault diagnosis correction strategy to obtain the target fault diagnosis result.

Images