Intelligent prediction method and system for nonlinear accumulated damage of movable guide vane of water turbine

Abstract

The invention belongs to the technical field of structural health monitoring and life prediction of water turbine generator sets, and particularly relates to an intelligent prediction method and system for nonlinear accumulated damage of a movable guide vane of a water turbine. The method comprises the following steps: acquiring operation state data, structure response data and environment data of a movable guide vane through a sensor network and a unit monitoring system to obtain an original monitoring data set; and preprocessing the original monitoring data set to obtain preprocessed multi-source fusion data. According to the method, a traditional Miner linear accumulation criterion is replaced by a nonlinear damage accumulation model considering a load sequence effect, and a nonlinear exponential function related to the current stress amplitude and front and back stress processes is obtained through deep learning network training; the method can accurately reflect the acceleration or delay effect of the alternating effect of high and low loads on damage accumulation under the variable-amplitude load condition, overcomes the defect that the linear accumulation theory assumes that all levels of load damage are mutually independent, and remarkably improves the damage prediction precision under the variable-working-condition operation condition.

Description

Technical Field

[0001] This invention belongs to the field of structural health monitoring and life prediction technology of hydro-generator units, specifically involving an intelligent prediction method and system for nonlinear cumulative damage of moving guide vanes of hydro-turbines. Background Technology

[0002] The turbine is the core equipment of a hydroelectric power generation system, and the guide vanes, as key moving components for regulating flow and controlling output, directly affect the operational safety and power generation efficiency of the entire power station. In recent years, with the increasing flexibility of power grid dispatching, hydroelectric units frequently participate in deep peak shaving, rapid start-up and shutdown, and large-scale load adjustments, causing the guide vanes to endure high-amplitude, variable-frequency alternating hydrodynamic loads for extended periods. Under this complex stress environment, the guide vane structure is highly susceptible to fatigue damage, manifested as the initiation and propagation of microcracks until eventual fracture. If the damage evolution pattern cannot be grasped in a timely manner and the remaining lifespan cannot be accurately predicted, it will be difficult to effectively prevent unit accidents caused by guide vane fracture.

[0003] Currently, fatigue assessment of turbine guide vanes mainly relies on numerical verification based on the linear cumulative damage theory (Miner's criterion), empirical judgment based on periodic disassembly inspections, and condition monitoring based on single-parameter thresholds such as vibration or temperature. These methods have significant shortcomings in engineering applications. The linear cumulative damage theory assumes that damage caused by each load level is independent and linearly superimposed, failing to consider load sequence effects and nonlinear interactions between loads, leading to large damage prediction errors under variable amplitude load conditions. Secondly, existing methods often combine offline calculations with periodic inspections, failing to achieve real-time monitoring and dynamic prediction of damage states, making it difficult to meet the early warning requirements of rapid changes in unit operating conditions. Furthermore, traditional models typically only consider uniaxial stress states, neglecting the multiaxial complex stresses actually borne by the guide vanes and their impact on fatigue life. The coupling effects of material performance degradation and water corrosion during long-term service on fatigue damage are not fully considered, causing deviations between predicted results and actual damage states. This results in a lack of deep integration of existing technologies with the unit's operating data system, making it difficult for predicted results to directly guide maintenance decisions and operational optimization.

[0004] The aforementioned technical limitations mean that current hydropower stations still mainly rely on planned maintenance and reactive repairs. This poses the risk of missing early damage due to insufficient prediction accuracy, and may also lead to unnecessary shutdowns for maintenance due to overly conservative assessments, thus hindering further improvements in the reliability and economic benefits of the generating units. Summary of the Invention

[0005] (1) Technical problems to be solved The purpose of this invention is to provide an intelligent prediction method and system for nonlinear cumulative damage of turbine guide vanes, in order to solve the problem that the risk of missing early damage due to insufficient prediction accuracy leads to overly conservative assessment of turbine guide vanes, resulting in unnecessary downtime for maintenance.

[0006] (2) Technical solution To achieve the above objectives, on the one hand, the present invention provides an intelligent prediction method for nonlinear cumulative damage of moving guide vanes of a water turbine, comprising: S1. Collect the operating status data, structural response data and environmental data of the movable guide vane through the sensor network and the unit monitoring system to obtain the raw monitoring dataset; preprocess the raw monitoring dataset to obtain the preprocessed multi-source fusion data.

[0007] S2. Based on the design drawings of the movable guide vane, a three-dimensional geometric model is established. The three-dimensional geometric model is then meshed and boundary conditions are set to obtain a finite element analysis model. Based on the strain time history data from the multi-source fusion data, and combined with the finite element analysis model, the stress tensor time history of the key parts of the guide vane is obtained by reconstructing the multiaxial stress state using the Kalman filter algorithm. In the Kalman filter algorithm, the state vector is defined as the six stress components of the key parts. The observation vector is defined as the strain value at each strain sensor measurement point, and the observation matrix is ​​constructed based on the stress-strain conversion relationship at the sensor measurement point location in the finite element analysis model.

[0008] S3. Calculate the stress intensity time history of each key component based on the stress tensor time history; perform cycle identification and statistics on the stress intensity time history using the rainflow counting method to obtain stress cycle data, which includes stress amplitude, average stress, and cycle number; based on the operating status data in the multi-source fusion data, statistically analyze the number of start-ups and shutdowns and the number of depth peak shavings under different head conditions to obtain the annual cycle frequency for each operating condition; based on the annual cycle frequency and the corresponding alternating stress amplitude in the stress cycle data, classify according to the variable load mode to obtain the load sequence characteristics.

[0009] S4. Construct a nonlinear damage accumulation model considering the load sequence effect based on stress cycle data; find the allowable number of cycles for each stress cycle based on the SN curve of the material in water; obtain a nonlinear exponential function by training historical damage data through a deep learning network; substitute the nonlinear exponential function into the nonlinear damage accumulation model to calculate the current cumulative damage degree.

[0010] S5. Based on the initial mechanical property parameters and service time of the material, establish an empirical model of material performance degradation to obtain the yield strength of the material at the current service time, and calculate the ratio of the current yield strength to the initial yield strength as the material degradation index; establish a corrosion pit morphology evolution model based on environmental data in multi-source fusion data to obtain the stress concentration factor caused by corrosion; combine the stress concentration factor with the SN curve of the material in water to obtain a modified SN curve that considers the corrosion effect.

[0011] S6. Input the stress cycle data, load sequence characteristics, environmental data and material degradation index as input features into the neural network prediction model to obtain the probability distribution of remaining service life; use the Bayesian algorithm to integrate the new monitoring data into the prior distribution to obtain the dynamically updated posterior distribution of remaining service life.

[0012] S7. Based on the current cumulative damage level, the dynamically updated posterior distribution of remaining lifespan, and the corrected SN curve, calculate the expected value of future damage rate through Monte Carlo simulation to obtain the predicted value of remaining lifespan; compare the current cumulative damage level with the preset damage level threshold to obtain the health status classification result and corresponding operation and maintenance suggestions.

[0013] Furthermore, step S1 includes: Operating status data is acquired in real time through the unit's SCADA system interface, including upstream water level, downstream water level, gross head, guide vane opening, calculated power, and cumulative operating time; gross head data is obtained by calculating the difference between upstream and downstream water levels; and the current operating condition type is obtained by querying the operating condition parameter table based on the gross head and guide vane opening.

[0014] Structural response data is obtained by collecting strain time history data at each measuring point through waterproof strain sensors arranged at the root of the guide vane, the middle of the guide vane arm, and the bottom of the pivot. The root of the guide vane is the connection between the guide vane and the guide vane arm, and the bottom of the pivot includes the location of the bolt holes at the bottom of the pivot.

[0015] Environmental data, including water temperature, sediment content, and pH value, are collected using water quality analyzers and temperature sensors.

[0016] During preprocessing, outliers are removed from the original monitoring data using the 3σ criterion to obtain outlier-free data; noise is filtered from the outlier-free data using a low-pass filter to obtain filtered data; timestamps are aligned to the filtered data based on a unified time reference to obtain aligned data; missing values ​​in the aligned data are filled using linear interpolation to obtain multi-source fused data.

[0017] Furthermore, step S2 includes: The geometric dimensions of the guide vane body, guide vane shaft, and shaft arm are extracted from the design drawings of the movable guide vane, and a three-dimensional geometric model is obtained by establishing the geometric model of the movable guide vane using three-dimensional modeling software.

[0018] The initial mesh model is obtained by meshing the three-dimensional geometric model using a hybrid meshing strategy with hexahedral mesh as the main component and triangular prism mesh as the auxiliary component. The guide vane trailing edge and top bolt hole areas in the initial mesh model are locally meshed to obtain a refined mesh model. The refined mesh model is then subjected to mesh quality checks to ensure that the mesh quality meets the preset conditions, resulting in a finite element mesh that meets the accuracy requirements.

[0019] A cylindrical coordinate system is established at the center of the guide vane shaft tip. X, Y, and Z are defined as radial, circumferential, and axial, respectively, to obtain a local coordinate system. In the local coordinate system, axial displacement constraints are applied to the top bolt hole and the bottom center to obtain axial constraint conditions. Radial displacement constraints are applied to the contact surface between the guide vane shaft and the bearing bush to obtain radial constraint conditions. Radial and circumferential displacement constraints are applied to the contact surface between the guide vane shaft and the rotating arm to obtain circumferential constraint conditions. The flow field wall pressure is mapped to the flow surface of the moving guide vane using the fluid-structure interaction mapping method to obtain pressure load boundary conditions. The axial constraint conditions, radial constraint conditions, circumferential constraint conditions, and pressure load boundary conditions are applied to the finite element mesh to obtain a complete finite element analysis model.

[0020] Furthermore, step S3 includes: Calculate the three principal stresses at each time step based on the six stress components in the stress tensor time history. , , Obtain the principal stress time history; according to the third strength theory, calculate the first principal stress at each time step. With the third principal stress Subtracting the two yields the stress intensity time history; the expression for calculating stress intensity is: ; in, for Stress intensity at time t, for The first principal stress at any given moment, for The third principal stress at time [time].

[0021] For the guide vane shaft arm, the alternating stress amplitude of the guide vane shaft arm is obtained by adding the stress intensity before the load change and the stress intensity after the load change; for the guide vane root, the alternating stress amplitude of the guide vane root is obtained by subtracting the stress intensity before the load change and the stress intensity after the load change and taking the absolute value; the alternating stress amplitude is then multiplied by the load amplification factor to obtain the corrected alternating stress amplitude that takes into account the calculation error and the influence of pressure pulsation.

[0022] The stress intensity time history was identified by the four-point rainflow counting method to extract the peak and valley values ​​of each complete stress cycle to obtain the stress cycle sequence. The stress amplitude and average stress of each cycle were calculated based on the stress cycle sequence, and the number of cycles with the same amplitude level was counted to obtain stress cycle data containing stress amplitude, average stress and number of cycles.

[0023] Based on the operational status data from the multi-source fusion data, the number of start-ups and shutdowns and the number of depth peak shavings under different head conditions are statistically analyzed to obtain the annual cycle frequency for each operating condition. Based on the annual cycle frequency and the corresponding alternating stress amplitude in the stress cycle data, the load sequence characteristics are obtained by classifying the load change modes from rated load to shutdown and from rated load to partial load. The load sequence characteristics include overload ratio and load block size.

[0024] Furthermore, in step S4, the expression for the nonlinear damage accumulation model is: ; in, To accumulate damage, This represents the total number of stress cycles. For the first The allowable number of stress cycles is determined by looking up the SN curve of the material in water. It is a non-linear exponential function. This represents the current stress amplitude. This is the previous stress amplitude. This represents the average stress of the current cycle.

[0025] Based on the SN curves in water, a correspondence between stress amplitude and allowable cycle number is established to obtain an SN curve database; the modified alternating stress amplitude is input into the SN curve database for querying to obtain the corresponding allowable cycle number.

[0026] The historical stress amplitude sequence, average stress sequence, stress ratio, and material property degradation index are used as input vectors for the deep learning network, and the measured damage degree is used as a label. The network parameters are optimized through backpropagation algorithm to obtain the trained nonlinear exponential function. The current stress cycle parameters are substituted into the trained nonlinear exponential function to obtain the nonlinear exponential value corresponding to the current cycle. The damage increment of each cycle is accumulated according to the nonlinear exponent to obtain the current cumulative damage degree.

[0027] Furthermore, step S5 includes: The expression for the empirical model of material property degradation is: ; in, Service life Yield strength at that time The initial yield strength of the material. The degradation coefficient; degradation coefficient The measured yield strength values ​​at different service time points were obtained by conducting mechanical property tests on periodically sampled specimens, and regression fitting was performed on the measured values ​​to obtain the yield strength. Based on the material yield strength at the current service time With the initial yield strength of the material The ratio of the yield strength ratio is used to calculate the current yield strength ratio and obtain the material degradation index. The corrosion depth is obtained by calculating the corrosion rate based on environmental data such as water temperature, sediment content, and pH value, combined with corrosion kinetic equations. The evolution over time; based on the geometric characteristics of the corrosion pits, the radius of the corrosion pits is established. and the radius of curvature of the pit bottom With corrosion depth The empirical relationship; the formula for calculating the stress concentration factor is: ; corrosion depth Radius of corrosion pit and the radius of curvature of the pit bottom Substituting into the above formula, we obtain the stress concentration factor caused by corrosion. ; stress concentration factor By combining the SN curve with the material in water, the stress amplitude is amplified and corrected to obtain a corrected SN curve that takes into account the corrosion effect.

[0028] Furthermore, step S6 includes: The input feature vector is obtained by vectorizing the amplitude, mean, and cycle sequence from the stress cycle data, the overload ratio and load block size from the load sequence features, the water temperature and corrosion rate from the environmental data, and the current yield strength ratio from the material degradation index. The input feature vector is then fed into an LSTM layer for temporal feature extraction to obtain the hidden state sequence. The hidden state sequence is then weighted and fused using an attention mechanism to obtain the fused feature representation. The fused feature representation is then mapped through a fully connected layer to obtain the mean and variance parameters of the remaining service life. A Gaussian distribution is then constructed based on the mean and variance parameters to obtain the probability distribution of the remaining service life.

[0029] The probability distribution of remaining useful life is used as the prior distribution. When new monitoring data is received, the likelihood function is calculated based on the new data to obtain the data likelihood value. The prior distribution is multiplied by the data likelihood value and normalized to obtain the dynamically updated posterior distribution of remaining useful life.

[0030] Furthermore, step S7 includes: Based on the alternating stress amplitude and annual cycle frequency of each working condition, and combined with the modified SN curve, the allowable number of cycles for each working condition is obtained, resulting in a table of allowable cycle counts for each working condition. The damage coefficient sequence is obtained by calculating the single-working-condition damage coefficient for each working condition based on the ratio of the annual cycle frequency to the allowable number of cycles. The annual cumulative damage coefficient is obtained by summing the damage coefficients of each working condition in the damage coefficient sequence. The design fatigue life is calculated based on the ratio of the critical damage value to the annual cumulative damage coefficient, resulting in the predicted design fatigue life. The expression for calculating the design fatigue life is as follows: ; in, To design fatigue life prediction values, This is the critical damage value. For the first Annual cycle count for various operating conditions; predicted fatigue life value Used to calculate the annual cumulative damage coefficient.

[0031] Based on the current cumulative damage level and annual cumulative damage coefficient, a sample set of future damage rates is obtained by generating future load scenario samples through Monte Carlo simulation; the expected value of the future damage rate is obtained by calculating the statistical mean of the sample set of future damage rates. ; Through the formula: Calculations were performed to obtain the predicted remaining useful life. ; the remaining useful life prediction value This will be output as the final prediction result.

[0032] The current cumulative damage level is compared with the preset damage level threshold. A value <0.3 indicates a healthy state, and normal monitoring and maintenance suggestions are output; when 0.3 ≤ When the value is less than 0.7, it is considered a state of concern, and maintenance suggestions to increase the monitoring frequency are output; when 0.7 ≤ When the value is less than 0.9, it is considered a warning state, and maintenance suggestions for planned repairs are output; when... If the value is ≥0.9, it is considered a dangerous state, and maintenance suggestions for immediate shutdown and repair are output.

[0033] Furthermore, the load amplification factor ranges from 1.0 to 1.1, and is used to compensate for the combined effects of alternating stress calculation errors, water flow pressure pulsations, average stress effects, and local stress concentrations. For the load variation stress difference condition obtained directly through finite element calculation, the alternating stress amplitude is multiplied by the load amplification factor to obtain the corrected alternating stress amplitude at the guide vane root; for the guide vane shaft arm, the alternating stress amplitude of all fatigue calculation conditions is multiplied by the load amplification factor to obtain the corrected alternating stress amplitude at the guide vane shaft arm; the corrected alternating stress amplitude is input into the SN curve database to obtain the corresponding allowable number of cycles.

[0034] Based on the same inventive concept, this invention also provides an intelligent prediction system for nonlinear cumulative damage of moving guide vanes of water turbines, including a data acquisition module, a guide vane modeling module, a data calculation module, a damage calculation module, a data correction module, a data prediction module, and a result output module.

[0035] The data acquisition module is used to collect operating status data, structural response data, and environmental data of the movable guide vanes through the sensor network and the unit monitoring system to obtain the raw monitoring dataset; the raw monitoring dataset is preprocessed to obtain preprocessed multi-source fusion data.

[0036] The guide vane modeling module is used to create a three-dimensional geometric model based on the design drawings of the active guide vane. The three-dimensional geometric model is meshed and boundary conditions are set to obtain a finite element analysis model. Based on the strain time history data in the multi-source fusion data, the stress tensor time history of the key parts of the guide vane is obtained by reconstructing the multiaxial stress state through the Kalman filter algorithm in combination with the finite element analysis model.

[0037] The data calculation module is used to calculate the stress intensity time history of each key part based on the stress tensor time history; the stress intensity time history is used to identify and statistically analyze the cycle of stress intensity time history to obtain stress cycle data, which includes stress amplitude, average stress and cycle number; the number of start-ups and shutdowns and the number of depth peak shavings under different head conditions are statistically analyzed based on the operating status data in the multi-source fusion data to obtain the annual cycle frequency of each working condition; based on the annual cycle frequency and the corresponding alternating stress amplitude in the stress cycle data, the load sequence characteristics are obtained by classifying according to the variable load mode.

[0038] The damage calculation module is used to construct a nonlinear damage accumulation model that considers the load sequence effect based on stress cycle data; to find the allowable number of cycles for each stress cycle based on the SN curve of the material in water; to obtain a nonlinear exponential function by training historical damage data through a deep learning network; and to calculate the current cumulative damage by substituting the nonlinear exponential function into the nonlinear damage accumulation model.

[0039] The data correction module is used to establish an empirical model of material performance degradation based on the initial mechanical property parameters and service time of the material to obtain the yield strength of the material at the current service time, and to calculate the ratio of the current yield strength to the initial yield strength as a material degradation index; to establish a corrosion pit morphology evolution model based on environmental data in multi-source fusion data to obtain the stress concentration factor caused by corrosion; and to combine the stress concentration factor with the SN curve of the material in water to obtain a corrected SN curve that takes into account the corrosion effect.

[0040] The data prediction module is used to input stress cycle data, load sequence characteristics, environmental data and material degradation indicators as input features into the neural network prediction model to obtain the probability distribution of remaining service life; the new monitoring data is incorporated into the prior distribution through the Bayesian algorithm to obtain the dynamically updated posterior distribution of remaining service life.

[0041] The results output module is used to calculate the expected value of future damage rate and obtain the predicted value of remaining service life by Monte Carlo simulation based on the current cumulative damage level, the dynamically updated posterior distribution of remaining service life, and the corrected SN curve; and to obtain the health status classification result and corresponding operation and maintenance suggestions by comparing the current cumulative damage level with the preset damage level threshold.

[0042] (3) Beneficial effects Compared with existing technologies, the advantages of this invention are as follows: This invention replaces the traditional Miner linear accumulation criterion with a nonlinear damage accumulation model that considers the load sequence effect. Through deep learning training, it obtains a nonlinear exponential function related to the current stress amplitude and the stress history before and after, accurately reflecting the acceleration or delaying effect of alternating high and low loads on damage accumulation under variable amplitude load conditions. This overcomes the defect of the linear accumulation theory's assumption that damage at each load level is independent, significantly improving the damage prediction accuracy under variable operating conditions. Simultaneously, this invention establishes an empirical model of material performance degradation and a corrosion pit morphology evolution model, quantitatively incorporating the decrease in material yield strength due to long-term service and the stress concentration effect caused by water corrosion into fatigue damage analysis, obtaining a modified SN curve considering corrosion effects. This allows the prediction results to truly reflect the performance degradation law of materials in actual service environments. Furthermore, this invention sets multi-level thresholds based on the cumulative damage degree for health status classification, outputting corresponding operation and maintenance suggestions for different states. The prediction results are directly transformed into actionable maintenance decisions, facilitating timely and targeted measures by maintenance personnel and effectively avoiding unnecessary downtime for maintenance due to overly conservative assessments or the risk of missing early damage due to insufficient prediction accuracy. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 Design drawings and three-dimensional geometric model of the movable guide vane; Figure 2 Mesh generation and boundary condition diagram for the active guide vane; Figure 3Cloud maps showing the stress intensity distribution of the guide vane under different operating conditions; Figure 4 The SN curve of ZG04Cr13Ni5Mo material in water; Figure 5 This is a graph showing the variation of guide vane stress intensity under different operating conditions. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Throughout this specification, references to "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "an embodiment," "an example," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0047] Example 1: This example uses the movable guide vane of a right bank unit in a giant hydropower station as an example to describe in detail the intelligent prediction method for nonlinear cumulative damage of the movable guide vane of the present invention. The hydropower station unit has a rated power of 870MW, a rated head of 137m, a maximum head of 172.27m, a minimum head of 106m, and the movable guide vane material is ZG04Cr13Ni5Mo.

[0048] Step S1: Data Acquisition and Preprocessing Operational status data is acquired in real time through the unit's SCADA system interface. In this embodiment, the collected operational status data includes upstream water level, downstream water level, gross head, guide vane opening, calculated power, and cumulative operating time. The gross head data is obtained based on the difference between the upstream and downstream water levels. For example, when the upstream water level is 986.17m and the downstream water level is 813.9m, the calculated gross head is 172.27m, corresponding to the maximum head condition; when the upstream water level is 986.17m and the downstream water level is 849.17m, the calculated gross head is 137m, corresponding to the rated head condition; and when the upstream water level is 959.02m and the downstream water level is 853.02m, the calculated gross head is 106m, corresponding to the minimum head condition.

[0049] Structural response data were collected using waterproof strain sensors located at the root of the guide vane, the middle of the guide vane arm, and the bottom of the pivot. The root of the guide vane is the connection point between the guide vane body and the guide vane arm, which is a stress concentration area. The bottom of the pivot, including the location of the bolt holes at the bottom of the pivot, also exhibits stress concentration.

[0050] Environmental data, including water temperature, sediment content, and pH value, was collected using a water quality analyzer and temperature sensors. In this embodiment, the movable guide vane operates in an underwater environment for extended periods, where water quality parameters significantly influence the corrosion rate of the material.

[0051] During preprocessing, outliers are removed from the original monitoring data using the 3σ criterion to obtain outlier-free data; noise is filtered from the outlier-free data using a low-pass filter to obtain filtered data; timestamps are aligned to the filtered data based on a unified time reference to obtain aligned data; missing values ​​in the aligned data are filled using linear interpolation to obtain multi-source fused data.

[0052] Step S2: Finite element modeling The geometric dimensions of the guide vane body, guide vane shaft, and shaft arm were extracted from the design drawings of the movable guide vane. A 3D geometric model of the movable guide vane was then created using 3D modeling software. To avoid calculation errors caused by poor mesh quality, the transition section at the connection between the guide vane and the guide vane shaft was ignored in the 3D modeling. The material of the movable guide vane is ZG04Cr13Ni5Mo, and its main material properties are shown in Table 1.

[0053] Table 1 Material Properties and Evaluation Criteria

[0054] An initial mesh model was obtained by meshing the 3D geometric model using a hybrid meshing strategy, primarily using hexahedral meshes and secondarily using triangular prism meshes. The mesh quality was consistently maintained above 0.6. Local mesh refinement was then applied to the guide vane trailing edge and the top bolt hole area of ​​the initial mesh model to obtain a refined mesh model. In this embodiment, the finite element model has a total of 1,252,645 elements and 1,285,875 nodes.

[0055] A cylindrical coordinate system is established at the center of the guide vane shaft tip, with X, Y, and Z defined as radial, circumferential, and axial directions, respectively, to obtain a local coordinate system. Within this local coordinate system, axial displacement constraints are applied to the top bolt hole and the bottom center to obtain axial constraint conditions; radial displacement constraints are applied to the contact surface between the guide vane shaft and the bearing bush to obtain radial constraint conditions; and radial and circumferential displacement constraints are applied to the contact surface between the guide vane shaft and the rotating arm to obtain circumferential constraint conditions. Using the fluid-structure interaction mapping method and leveraging topology and coordinate consistency, the flow field wall pressure is mapped to the flow surface of the movable guide vane to obtain pressure load boundary conditions. The axial, radial, circumferential, and pressure load boundary conditions are then applied to the finite element mesh to obtain a complete finite element analysis model.

[0056] Based on the finite element analysis model, the Kalman filter algorithm is used to reconstruct the multiaxial stress state. The state-space model of the Kalman filter is constructed as follows: State vector Defined as the stress tensor components of key parts of the guide vane (guide vane root, guide vane arm middle, and pivot bottom), i.e. The dimension is 6. Observation vector Defined as the measured strain value at each strain sensor measurement point. In this embodiment, several strain sensors are arranged, and the dimension of the observation vector is equal to the number of sensors. State transition matrix. Based on the changing patterns of operating conditions at adjacent time points, the quasi-static operating condition can be approximated by the identity matrix. Observation matrix The construction method is as follows: In the finite element analysis model, a unit stress component is applied to each sensor measurement point, and the corresponding strain response is calculated. A linear mapping relationship between the stress component and the sensor strain is then established, and this mapping matrix is ​​the observation matrix. Process noise covariance matrix and observation noise covariance matrix The parameters are set according to the uncertainty of the working conditions and the measurement accuracy of the sensors. Through the prediction-update iterative process of Kalman filtering, the prior information of the finite element model and the measured strain data are integrated to obtain the optimal stress tensor estimates of key parts at each time point, thus constructing the stress tensor time history. Step S3: Stress Calculation and Cyclic Statistics Calculate the three principal stresses at each time step based on the six stress components in the stress tensor time history. , , The principal stress time histories are obtained. According to the third strength theory, the stress intensity time histories are obtained by subtracting the first principal stress σ1 from the third principal stress σ3 at each time point. The expression for calculating stress intensity is as follows: ;in, for Stress intensity at time t, for The first principal stress at any given moment, for The third principal stress is used at a specific moment. The third strength theory is adopted for stress intensity calculation because the maximum shear stress occurs at a 45-degree angle to the axis when low-carbon steel is stretched. The material slips along this plane and cracks are generated, which is consistent with the fatigue failure mechanism. At the same time, the calculation results show that the stress intensity under the same working conditions is greater than the equivalent stress, indicating that using stress intensity to assess fatigue is safer.

[0057] The strength calculation in this embodiment considers different load stability conditions and their corresponding shutdown conditions under three water heads, totaling 20 calculation conditions. The specific operating parameters are shown in Table 2.

[0058] Table 2 Calculation Conditions for the Strength of the Moving Guide Vane

[0059] Stress intensity distribution analysis for each calculation condition shows that the maximum stress intensity of the movable guide vane is mainly distributed at the root of the guide vane (where the guide vane connects to the guide vane arm) or at the bolt hole at the bottom of the pivot. The maximum stress intensity for each calculation condition is shown in Table 3.

[0060] Table 3 Maximum stress intensity under various calculation conditions

[0061] At the same head and upstream and downstream water levels, the maximum stress intensity of the movable guide vane gradually decreases as the opening degree gradually increases. Under the same load, the higher the unit operating head, the greater the maximum stress intensity of the movable guide vane, while the water level has a relatively small impact on the maximum stress intensity. Compared with the normal operating condition strength assessment standard of 174 MPa and the special operating condition strength assessment standard of 348 MPa, the stress and deformation of the movable guide vane under various operating conditions meet the requirements for safe and stable operation of the unit.

[0062] Regarding the calculation of alternating stress amplitude at the guide vane shaft arm and guide vane root, this invention, through analysis of the stress intensity distribution before and after load changes, yielded the following findings: (1) The stress difference of the guide vane shaft arm under load change is almost equal to the sum of the stress before load change and the stress after load change. It can be inferred that the stress signs of the guide vane shaft arm before and after load change are opposite.

[0063] (2) The stress difference at the root of the guide vane under load change is almost equal to the absolute value of the difference between the stress before and after the load change. It can be inferred that the stress sign of the guide vane is the same before and after the load change.

[0064] (3) For most loaded conditions, the root of the guide vane shaft arm has the maximum stress intensity before and after the load change, and it is located on the head side of the guide vane; while for the shutdown condition, it is located on the tail side. The location of the maximum value of the load change stress difference at the root of the guide vane shaft arm is basically the same as that of the shutdown condition, indicating that it is likely to depend on the maximum value of the stress distribution before and after the load change.

[0065] Therefore, for the guide vane shaft arm, the alternating stress amplitude of the guide vane shaft arm is obtained by adding the stress intensity before the load change and the stress intensity after the load change; for the guide vane root, the alternating stress amplitude of the guide vane root is obtained by subtracting the stress intensity before the load change and the stress intensity after the load change from the absolute value.

[0066] Considering the impact of fatigue on factors such as calculation errors of alternating stress at the guide vane root, water flow pressure pulsation, average stress, and local stress concentration, the alternating stress amplitude is corrected by multiplying it by a load amplification factor of 1.05. For the guide vane root, fatigue conditions where the stress difference due to load changes is not directly calculated are addressed. The alternating stress is multiplied by the load amplification factor; for the guide vane shaft arm, the alternating stress for all fatigue calculation conditions is multiplied by the load amplification factor. In this embodiment, the fatigue calculation conditions are divided into 15 groups according to the variable load mode, and the alternating stress amplitude under each variable load condition is shown in Table 4.

[0067] Table 4. Alternating stress amplitude under various variable load conditions

[0068] Step S4: Nonlinear damage accumulation calculation A nonlinear damage accumulation model considering the load sequence effect is constructed based on stress cycle data. The allowable number of cycles for each stress cycle is determined from the SN curve of the material in water. A multi-layer fully connected neural network is constructed, taking the current stress amplitude, the previous stress amplitude, the current cycle average stress, the stress ratio, and the material degradation index as inputs, and the nonlinear exponent value as output. This network is trained using historical damage data to obtain a nonlinear exponent prediction network. The current stress cycle parameters are input into the trained network to obtain the nonlinear exponent value, which is then substituted into the nonlinear damage accumulation model to calculate the current cumulative damage. The expression for the nonlinear damage accumulation model is as follows: ;in, To accumulate damage, This represents the total number of stress cycles. For the first The allowable number of stress cycles is determined by looking up the SN curve of the material in water. It is a non-linear exponential function. This represents the current stress amplitude. This is the previous stress amplitude. The average stress of the current cycle; fatigue is the process by which certain regions of a material gradually undergo permanent structural changes under cyclic loading, leading to crack formation or fracture after a certain number of cycles. When the stress amplitude caused by load cycles varies, the traditional method generally uses the linear accumulation method for fatigue analysis.

[0069] The deep learning network used in this invention is a multi-layer fully connected neural network, and its specific structure is as follows: the number of neurons in the input layer is 5, corresponding to the current stress amplitude. Previous stress amplitude The average stress of the current cycle Stress ratio and material degradation index The first hidden layer contains 32 neurons, using ReLU activation; the second hidden layer contains 16 neurons, also using ReLU activation; the output layer contains 1 neuron, outputting the nonlinear exponential value of the current loop. The output layer does not use an activation function to ensure the continuity of the output values. Nonlinear exponent. The physical meaning is: when When the value is greater than 1, it indicates that the damage accumulation rate in the current cycle is lower than the linear accumulation assumption, corresponding to the delayed effect of low load after high load; when When <1, it indicates that the damage accumulation rate of the current cycle is higher than the linear accumulation assumption, corresponding to the acceleration effect of high load after low load; when When = 1, it degenerates into the traditional Miner linear accumulation criterion. Based on the physical laws of fatigue damage, The reasonable range for its value is 0.5 to 2.0.

[0070] The training data was constructed as follows: crack length or stiffness attenuation data at each inspection moment were extracted from the periodic inspection records of historical service units, and the damage increment at adjacent inspection moments was used as the basis for the data. and the number of stress cycles within the corresponding time period By reverse calculation, the average damage increment for each cycle is determined; combined with the stress amplitude and allowable number of cycles for each cycle, the damage increment is calculated. According to the formula The nonlinear exponential label values ​​for each cycle are calculated.

[0071] The linear cumulative damage theory states that under cyclic loading, fatigue damage can accumulate linearly, with each stress being independent and uncorrelated. When the accumulated damage reaches a certain value, fatigue failure occurs in the specimen or component. A typical example of linear cumulative damage theory is Miner's theory. In Miner's theory, the damage caused by one cycle is D = 1 / N, where N is the fatigue life corresponding to the current load level S. Under constant amplitude loading, the damage caused by n cycles is Dn = n / N. Under variable amplitude loading, the damage caused by n cycles is the sum of the damage from each load level. The critical fatigue damage (DCR) is approximately 1 under random loading.

[0072] Miner's theory is a linear fatigue cumulative damage theory that does not consider the influence of load sequence. The nonlinear damage accumulation model proposed in this invention introduces a nonlinear exponential function α related to the current stress amplitude and the stress histories before and after it. This model can accurately reflect the acceleration or delay effect of alternating high and low loads on damage accumulation under variable amplitude loading conditions, overcoming the deficiency of the linear cumulative theory that assumes that damage at each load level is independent.

[0073] High-cycle fatigue refers to a condition where the alternating stress on a material is far below its yield strength, and the number of stress cycles before fracture exceeds 10^5. The fatigue characteristics of this material are typically described using an SN curve. The SN curve, also known as the stress-life curve, refers to the alternating stress amplitude (S) and the number of cycles (N) required for fatigue failure at the corresponding stress amplitude. Based on the SN curve of ZG04Cr13Ni5Mo material in water, a database of SN curves was established to correspond the stress amplitude to the allowable number of cycles.

[0074] Historical stress amplitude sequences, average stress sequences, stress ratios, and material property degradation indices are used as input vectors for a deep learning network, with measured damage levels as labels. Backpropagation is used to optimize the network parameters, resulting in a trained nonlinear exponential function. The current stress cycle parameters are then substituted into this trained function to obtain the nonlinear exponential value for the current cycle. Finally, the damage increments from each cycle are summed according to the nonlinear exponent to obtain the current cumulative damage level.

[0075] Step S5: Material Degradation and Corrosion Correction Empirical models of material property degradation are used to describe the decay of a material's yield strength over service time. The expression for an empirical model of material property degradation is as follows: ;in, Service life Yield strength at that time The initial yield strength of the material. The degradation coefficient; degradation coefficient The measured yield strength values ​​at different service time points were obtained by conducting mechanical property tests on periodically sampled specimens. The degradation coefficient was then obtained through regression fitting of the measured values. The measured yield strength values ​​at different service time points were obtained by conducting mechanical property tests on periodically sampled specimens, and regression fitting was performed on the measured values ​​to obtain the yield strength of the material at the current service time. With the initial yield strength of the material The ratio of yield strength to current yield strength is used to calculate the material degradation index.

[0076] A corrosion pit morphology evolution model is used to calculate the stress concentration factor caused by corrosion. Based on environmental data such as water temperature, sediment content, and pH, the corrosion rate is calculated using corrosion kinetic equations to determine the corrosion depth. The evolution over time. Based on the geometric characteristic measurement data of the corrosion pits, the radius of the corrosion pits is established. and the radius of curvature of the pit bottom With corrosion depth The empirical relationship. The expression for calculating the stress concentration factor is: ; to increase corrosion depth Radius of corrosion pit and the radius of curvature of the pit bottom Substituting into the above formula, we obtain the stress concentration factor caused by corrosion. The stress concentration factor By combining the SN curve with the material in water, the stress amplitude is amplified and corrected to obtain a corrected SN curve that takes into account the corrosion effect.

[0077] Step S6: Lifetime Prediction and Probability Distribution The input feature vector is obtained by vectorizing the amplitude, mean, and cycle sequences from the stress cycle data, the overload ratio and load block size from the load sequence features, the water temperature and corrosion rate from the environmental data, and the current yield strength ratio from the material degradation index. This input feature vector is then fed into an LSTM layer for temporal feature extraction to obtain the hidden state sequence. The hidden state sequence is then weighted and fused using an attention mechanism to obtain the fused feature representation. This fused feature representation is then mapped through a fully connected layer to obtain the mean and variance parameters of the remaining service life. A Gaussian distribution is constructed based on the mean and variance parameters to obtain the probability distribution of the remaining service life.

[0078] Using the probability distribution of remaining useful life as the prior distribution, when new monitoring data is received, a likelihood function is calculated based on the new data to obtain the data likelihood value. The prior distribution is then multiplied by the data likelihood value and normalized to obtain the dynamically updated posterior distribution of remaining useful life.

[0079] Step S7: Fatigue life calculation and health status classification In this embodiment, the annual cycle number of the corresponding alternating stress can be determined by combining the unit's operating conditions. Based on the alternating stress amplitude and annual cycle frequency for each operating condition, and combined with the modified SN curve, the allowable cycle number for each operating condition is obtained, resulting in a table of allowable cycle numbers for each operating condition. The damage coefficient sequence for each operating condition is obtained by calculating the single-condition damage coefficient based on the ratio of the annual cycle frequency to the allowable cycle number. The annual cumulative damage coefficient is obtained by summing the damage coefficients for each operating condition in the damage coefficient sequence. The design fatigue life is calculated based on the ratio of the critical damage value to the annual cumulative damage coefficient, resulting in the predicted design fatigue life value. The expression for calculating the design fatigue life is as follows: ;in, To design fatigue life prediction values, This is the critical damage value. For the first Annual cycle count for various operating conditions; predicted fatigue life value Used to calculate the annual cumulative damage coefficient.

[0080] The fatigue calculation results of the movable guide vane in this embodiment are shown in Table 5.

[0081] Table 5 Fatigue Calculation of Moving Guide Vanes

[0082] Based on the above calculations, for a single unit, with 1000 start-ups / shutdowns / year, 730 deep peak shavings (0.2Pr-Pr) / year, and 2190 deep peak shavings (0.5Pr-Pr) / year, the lifespan of the movable guide vane root is 18.31 years. If the unit's design life is 40 years, the movable guide vane root is at risk of fatigue failure under frequent start-ups / shutdowns and deep peak shavings, while the guide vane shaft arm, due to its relatively small alternating stress, will not experience fatigue failure within its design life. Since the movable guide vane skirt was not considered in the geometric modeling, the calculated stress may be greater than the actual stress under stress concentration. Therefore, without the influence of factors such as severe flow-induced vibration, the actual lifespan of the movable guide vane is likely to be greater than 18.31 years.

[0083] Based on the current cumulative damage level and annual cumulative damage coefficient, a sample set of future damage rates is generated using Monte Carlo simulation to produce future load scenario samples. The expected value E of the future damage rate is obtained by calculating the statistical mean of the sample set. The predicted remaining service life Te is then calculated using the formula Te=(1-D) / E. The predicted remaining service life Te is output as the final prediction result.

[0084] The current cumulative damage level is compared with the preset damage level threshold to classify the health status: when D < 0.3, it is judged as a healthy state, and the operation and maintenance suggestion for normal monitoring is output; when 0.3 ≤ D < 0.7, it is judged as a state of concern, and the operation and maintenance suggestion for increasing the monitoring frequency is output; when 0.7 ≤ D < 0.9, it is judged as a state of warning, and the operation and maintenance suggestion for planned maintenance is output; when D ≥ 0.9, it is judged as a state of danger, and the operation and maintenance suggestion for immediate shutdown and maintenance is output.

[0085] This embodiment also provides an intelligent prediction system for nonlinear cumulative damage of moving guide vanes of a water turbine, which is used to implement the method described in Embodiment 1. Figure 2 As shown, the system includes a data acquisition module, a guide vane modeling module, a data calculation module, a damage calculation module, a data correction module, a data prediction module, and a result output module.

[0086] The data acquisition module is used to collect operating status data, structural response data, and environmental data of the movable guide vanes through the sensor network and the unit monitoring system to obtain the raw monitoring dataset; the raw monitoring dataset is preprocessed to obtain preprocessed multi-source fusion data.

[0087] The guide vane modeling module is used to create a three-dimensional geometric model based on the design drawings of the active guide vane. The three-dimensional geometric model is meshed and boundary conditions are set to obtain a finite element analysis model. Based on the strain time history data in the multi-source fusion data, the stress tensor time history of the key parts of the guide vane is obtained by reconstructing the multiaxial stress state through the Kalman filter algorithm in combination with the finite element analysis model.

[0088] The data calculation module is used to calculate the stress intensity time history of each key part based on the stress tensor time history; the stress intensity time history is used to identify and statistically analyze the cycle of stress intensity time history to obtain stress cycle data, which includes stress amplitude, average stress and cycle number; the number of start-ups and shutdowns and the number of depth peak shavings under different head conditions are statistically analyzed based on the operating status data in the multi-source fusion data to obtain the annual cycle frequency of each working condition; based on the annual cycle frequency and the corresponding alternating stress amplitude in the stress cycle data, the load sequence characteristics are obtained by classifying according to the variable load mode.

[0089] The damage calculation module is used to construct a nonlinear damage accumulation model that considers the load sequence effect based on stress cycle data; to find the allowable number of cycles for each stress cycle based on the SN curve of the material in water; to obtain a nonlinear exponential function by training historical damage data through a deep learning network; and to calculate the current cumulative damage by substituting the nonlinear exponential function into the nonlinear damage accumulation model.

[0090] The data correction module is used to establish an empirical model of material performance degradation based on the initial mechanical property parameters and service time of the material to obtain the yield strength of the material at the current service time, and to calculate the ratio of the current yield strength to the initial yield strength as a material degradation index; to establish a corrosion pit morphology evolution model based on environmental data in multi-source fusion data to obtain the stress concentration factor caused by corrosion; and to combine the stress concentration factor with the SN curve of the material in water to obtain a corrected SN curve that takes into account the corrosion effect.

[0091] The data prediction module is used to input stress cycle data, load sequence characteristics, environmental data and material degradation indicators as input features into the neural network prediction model to obtain the probability distribution of remaining service life; the new monitoring data is incorporated into the prior distribution through the Bayesian algorithm to obtain the dynamically updated posterior distribution of remaining service life.

[0092] The results output module is used to calculate the expected value of future damage rate and obtain the predicted value of remaining service life by Monte Carlo simulation based on the current cumulative damage level, the dynamically updated posterior distribution of remaining service life, and the corrected SN curve; and to obtain the health status classification result and corresponding operation and maintenance suggestions by comparing the current cumulative damage level with the preset damage level threshold.

[0093] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0094] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent prediction of non-linear cumulative damage of hydraulic turbine wicket gate, characterized in that, The method comprises the following steps: S1, collecting the operation state data, structural response data and environmental data of the movable guide vane through a sensor network and a unit monitoring system to obtain an original monitoring data set; and preprocessing the original monitoring data set to obtain multi-source fusion data after preprocessing; S2, establishing a three-dimensional geometric model according to the design drawings of the movable guide vane, performing mesh division on the three-dimensional geometric model and setting boundary conditions to obtain a finite element analysis model; and according to the strain time history data in the multi-source fusion data, combining the finite element analysis model and performing multi-axis stress state reconstruction through a Kalman filtering algorithm to obtain a stress tensor time history of key positions of the guide vane; S3, calculating a stress intensity time history of each key position according to the stress tensor time history; performing cycle identification and statistics on the stress intensity time history through a rainflow counting method to obtain stress cycle data, the stress cycle data including a stress amplitude, an average stress and a cycle number; and according to the operation state data in the multi-source fusion data, counting the start-stop machine number and the deep peak regulation number under different water head conditions to obtain the annual cycle frequency of each working condition; and according to the annual cycle frequency and the corresponding alternating stress amplitude in the stress cycle data, classifying according to a variable load mode to obtain load sequence characteristics; S4, constructing a nonlinear damage accumulation model considering load sequence effects according to the stress cycle data; finding the allowable cycle number of each stress cycle according to the material S-N curve in water; training historical damage data through a deep learning network to obtain a nonlinear exponential function; and substituting the nonlinear exponential function into the nonlinear damage accumulation model to perform damage calculation to obtain a current cumulative damage degree; S5, establishing a material performance degradation empirical model according to material initial mechanical performance parameters and service time to obtain a material yield strength at a current service time, and calculating a ratio of the current yield strength to the initial yield strength as a material degradation index; establishing a corrosion pit morphology evolution model according to the environmental data in the multi-source fusion data to obtain a stress concentration coefficient caused by corrosion; and combining the stress concentration coefficient with the material S-N curve in water to obtain a modified S-N curve considering the corrosion effect; S6, inputting the stress cycle data, the load sequence characteristics, the environmental data and the material degradation index as input features into a neural network prediction model to obtain a probability distribution of the remaining useful life; and fusing new monitoring data into a prior distribution through a Bayesian algorithm to obtain a dynamically updated remaining life posterior distribution; S7, calculating a future damage rate expectation value through Monte Carlo simulation according to the current cumulative damage degree, the dynamically updated remaining life posterior distribution and the modified S-N curve to obtain a remaining useful life prediction value; and comparing the current cumulative damage degree with a preset damage degree threshold to obtain a health state classification result and a corresponding operation and maintenance suggestion.

2. The intelligent prediction method for non-linear cumulative damage of guide vanes of a hydraulic turbine according to claim 1, characterized in that, The step S1 comprises: The operation state data is obtained in real time through a unit SCADA system interface, including an upstream water level, a downstream water level, a gross water head, a guide vane opening, a calculated power and a cumulative running time; the gross water head data is obtained by calculating the difference between the upstream water level and the downstream water level; and the current running working condition type is obtained by querying a working condition parameter table according to the gross water head and the guide vane opening. The structural response data are collected by waterproof strain sensors arranged at the guide vane root, the guide vane arm middle part and the pivot bottom, wherein the guide vane root is the connection between the guide vane and the guide vane arm, and the pivot bottom includes the bolt hole position of the pivot bottom; The environmental data are collected by a water quality analyzer and a temperature sensor, including water temperature, sediment content and pH value.

3. The intelligent prediction method for non-linear cumulative damage of guide vanes of a hydraulic turbine according to claim 2, characterized in that, The step S2 comprises: According to the design drawings of the movable guide vane, the geometric dimensions of the guide vane body, the guide vane shaft and the shaft arm are extracted, and a three-dimensional geometric model of the movable guide vane is established by using a three-dimensional modeling software to obtain a three-dimensional geometric model; The three-dimensional geometric model is meshed by using a mixed meshing strategy of hexahedral meshing as the main method and triangular prism meshing as the auxiliary method to obtain an initial mesh model; the guide vane trailing edge and the top bolt hole position in the initial mesh model are locally meshed to obtain an encrypted mesh model; and the mesh quality of the encrypted mesh model is checked to ensure that the mesh quality meets the preset conditions to obtain a finite element mesh meeting the accuracy requirements; A cylindrical coordinate system is established at the center of the top end of the guide vane shaft, and X, Y and Z are defined as the radial direction, the circumferential direction and the axial direction to obtain a local coordinate system; in the local coordinate system, an axial displacement constraint is applied to the top end bolt hole and the bottom center to obtain an axial constraint condition; a radial displacement constraint is applied to the contact surface between the guide vane shaft and the bearing bush to obtain a radial constraint condition; a radial and circumferential displacement constraint is applied to the contact surface between the guide vane shaft and the rotating arm to obtain a circumferential constraint condition; the wall surface pressure of the flow field is mapped to the flow surface of the movable guide vane by using a fluid-structure interaction mapping method to obtain a pressure load boundary condition; and the axial constraint condition, the radial constraint condition, the circumferential constraint condition and the pressure load boundary condition are applied to the finite element mesh to obtain a complete finite element analysis model.

4. The intelligent prediction method for non-linear cumulative damage of guide vanes of a hydraulic turbine according to claim 3, characterized in that, The step S3 comprises: According to six stress components in stress tensor time history, three principal stresses at each time are calculated , , Principal stress time history is obtained; according to the third strength theory, the first principal stress at each time is subtracted from the third principal stress to obtain stress intensity time history; the calculation expression of stress intensity is: ； wherein is the stress intensity at the moment, is the first principal stress at the moment, is the third principal stress at the moment; For the guide vane shaft arm position, the stress intensity before the load change and the stress intensity after the load change are added to obtain the alternating stress amplitude of the guide vane shaft arm; for the guide vane root, the stress intensity before the load change and the stress intensity after the load change are subtracted to obtain the absolute value of the alternating stress amplitude of the guide vane root; and the alternating stress amplitude is multiplied by a load amplification factor to obtain a modified alternating stress amplitude considering the calculation error and the pressure pulsation influence; The stress intensity time history is cyclically identified by using a four-point rainflow counting method, the peak value and the valley value of each complete stress cycle are extracted to obtain a stress cycle sequence; the stress amplitude and the average stress of each cycle are calculated according to the stress cycle sequence, and the cycle number of the same amplitude level is counted to obtain stress cycle data containing the stress amplitude, the average stress and the cycle number; According to the operation state data in the multi-source fusion data, the start-stop machine number and the deep peak regulation number under different water head conditions are counted to obtain the annual cycle frequency of each working condition; according to the annual cycle frequency and the corresponding alternating stress amplitude in the stress cycle data, the load sequence characteristics are classified according to the variable load mode from the rated load to shutdown and from the rated load to partial load; the load sequence characteristics include the overload ratio and the load block size.

5. The intelligent prediction method for non-linear cumulative damage of guide vanes of a hydraulic turbine according to claim 4, characterized in that, In the step S4, the expression of the nonlinear damage accumulation model is: ； wherein, is the accumulated damage degree, is the total number of stress cycles, is the allowable number of cycles for the th stress cycle obtained from the S-N curve of the material in water, is a non-linear exponential function, is the current stress amplitude, is the previous stress amplitude, is the average stress of the current cycle; According to the material S-N curve in water, a corresponding relationship between the stress amplitude and the allowable cycle number is established to obtain an S-N curve database; and the corrected alternating stress amplitude is input into the S-N curve database to obtain the corresponding allowable cycle number; The historical stress amplitude sequence, the average stress sequence, the stress ratio and the material performance degradation index are taken as the input vectors of the deep learning network, and the measured damage degree is taken as the label, and the network parameter optimization is performed through the back propagation algorithm to obtain the trained nonlinear exponential function; the current stress cycle parameters are substituted into the trained nonlinear exponential function to obtain the nonlinear exponential value corresponding to the current cycle; and the damage increments of each cycle are accumulated according to the nonlinear exponential to obtain the current cumulative damage degree.

6. The intelligent prediction method for non-linear cumulative damage of guide vanes of a hydraulic turbine according to claim 5, characterized in that, The step S5 comprises: The expression of the material performance degradation empirical model is: ； wherein, is the yield strength at the service time is the yield strength at the service time is the initial yield strength of the material, is the degradation coefficient; the degradation coefficient The measured value of the yield strength at different service time points is obtained by testing the mechanical properties of the periodically sampled test pieces, and the measured value is regressed and fitted to obtain a ratio of the current yield strength to the initial yield strength of the material a ratio of the current yield strength to the initial yield strength of the material a ratio of the current yield strength to the initial yield strength of the material The corrosion depth is obtained by calculating the corrosion rate according to the water temperature, the silt content and the pH value in the environmental data and the corrosion kinetics equation The evolution relationship with time; the empirical relationship between the corrosion pit radius and the pit bottom curvature radius and the corrosion depth The calculation expression of the stress concentration coefficient is: ； corrosion depth , corrosion pit radius and pit bottom curvature radius Substitute the above formula to get the stress concentration factor caused by corrosion ; the stress concentration factor combined with the material S-N curve in water, the stress amplitude is amplified to get the modified S-N curve considering the effect of corrosion.

7. The intelligent prediction method for non-linear cumulative damage of guide vanes of a hydraulic turbine according to claim 6, characterized in that, The step S6 comprises: The amplitude, the mean value and the number sequence in the stress cycle data, the overload ratio and the load block size in the load sequence feature, the water temperature and the corrosion rate in the environmental data, and the current yield strength ratio in the material degradation index are vectorized to obtain the input feature vector; the input feature vector is input into the LSTM layer to extract the time sequence feature to obtain the hidden state sequence; the hidden state sequence is weighted and fused through the attention mechanism to obtain the fused feature representation; the fused feature representation is mapped through the full connection layer to obtain the mean value and the variance parameter of the remaining useful life; and the Gaussian distribution is constructed according to the mean value and the variance parameter to obtain the probability distribution of the remaining useful life; The probability distribution of the remaining useful life is taken as the prior distribution, when the new monitoring data is received, the likelihood function is calculated according to the new data to obtain the data likelihood value; the prior distribution is multiplied by the data likelihood value and is normalized to obtain the dynamically updated remaining life posterior distribution.

8. The intelligent prediction method for non-linear cumulative damage of guide vanes of a hydraulic turbine according to claim 7, characterized in that, The step S7 comprises: According to the alternating stress amplitude and the annual cycle frequency of each working condition, the allowable cycle number of each working condition is obtained by combining the corrected S-N curve to obtain the allowable cycle number table of each working condition; the single-working-condition damage coefficient of each working condition is obtained by calculating the ratio of the annual cycle frequency to the allowable cycle number to obtain the damage coefficient sequence; the damage coefficients of each working condition in the damage coefficient sequence are accumulated to obtain the annual cumulative damage coefficient; the design fatigue life is calculated according to the ratio of the critical damage value to the annual cumulative damage coefficient to obtain the design fatigue life prediction value; and the calculation expression of the design fatigue life is: ； wherein, is a design fatigue life prediction value, is a critical damage value, is a number of annual cycles for a design fatigue life prediction value for calculating an annual cumulative damage coefficient; According to the current accumulated damage degree and the annual accumulated damage coefficient, a future damage rate sample set is generated by Monte Carlo simulation of future load scenarios; a statistical mean is calculated for the future damage rate sample set to obtain a future damage rate expectation value ; a remaining service life prediction value is calculated by the formula: ; and the remaining service life prediction value is output as the final prediction result.​ The current accumulated damage degree is compared with a preset damage degree threshold, and when <0.3 is determined as a healthy state, and an operation and maintenance suggestion of normal monitoring is output; when 0.3≤ <0.7 is determined as a state of attention, and an operation and maintenance suggestion of strengthening monitoring frequency is output; when 0.7≤ <0.9 is determined as a pre-warning state, and an operation and maintenance suggestion of planned maintenance is output; when ≥0.9 is determined as a dangerous state, and an operation and maintenance suggestion of immediate shutdown for repair is output.

9. The intelligent prediction method for non-linear cumulative damage of guide vanes of a hydraulic turbine according to claim 8, characterized in that, The load amplification factor is in the range of 1.0 to 1.1, which is used to compensate the comprehensive effects of the alternating stress calculation error, the water flow pressure fluctuation, the average stress effect and the local stress concentration; For the load change stress difference working condition obtained by direct finite element calculation, the alternating stress amplitude is multiplied by the load amplification factor to obtain the corrected alternating stress amplitude of the guide vane root; for the guide vane shaft arm part, the alternating stress amplitudes of all fatigue calculation working conditions are multiplied by the load amplification factor to obtain the corrected alternating stress amplitude of the guide vane shaft arm; and the corrected alternating stress amplitude is input into the S-N curve database to obtain the corresponding allowable cycle number.

10. A system for intelligent prediction of non-linear cumulative damage of hydraulic turbine wicket gate, for implementing the method of any one of claims 1-9, characterized in that, The system comprises a data acquisition module, a guide vane modeling module, a data calculation module, a damage calculation module, a data correction module, a data prediction module and a result output module; The data acquisition module is configured to acquire operation state data, structural response data and environmental data of the movable guide vane through a sensor network and a unit monitoring system to obtain an original monitoring data set; and the original monitoring data set is preprocessed to obtain multi-source fusion data after preprocessing. The guide vane modeling module is configured to establish a three-dimensional geometric model according to design drawings of the movable guide vane, perform meshing on the three-dimensional geometric model and set boundary conditions to obtain a finite element analysis model; and according to strain time history data in the multi-source fusion data, the stress tensor time history of key positions of the guide vane is reconstructed by a Kalman filtering algorithm in combination with the finite element analysis model. The data calculation module is configured to calculate stress intensity time histories of the key positions according to the stress tensor time history. The stress intensity time histories are subjected to cycle identification and statistics by a rainflow counting method to obtain stress cycle data, which includes a stress amplitude, an average stress and a cycle number. The annual cycle frequency of each working condition is obtained according to the operation state data in the multi-source fusion data, the start-stop number and the deep peak regulation number under different water head conditions; and the load sequence characteristics are obtained according to the variable load mode in combination with the annual cycle frequency and the corresponding alternating stress amplitude in the stress cycle data. The damage calculation module is configured to construct a nonlinear damage accumulation model considering the load sequence effect according to the stress cycle data; to obtain the allowable cycle number of each stress cycle according to a material S-N curve in water; to obtain a nonlinear exponential function by training historical damage data through a deep learning network; and to obtain the current cumulative damage degree by substituting the nonlinear exponential function into the nonlinear damage accumulation model for damage calculation. The data correction module is configured to establish a material performance degradation empirical model according to initial mechanical performance parameters of the material and a service time to obtain a material yield strength at the current service time, and to calculate a ratio of the current yield strength to the initial yield strength as a material degradation index; to establish a corrosion pit morphology evolution model according to the environmental data in the multi-source fusion data to obtain a stress concentration coefficient caused by corrosion; and to combine the stress concentration coefficient with the material S-N curve in water to obtain a modified S-N curve considering the corrosion effect. The data prediction module is configured to input the stress cycle data, the load sequence characteristics, the environmental data and the material degradation index as input features into a neural network prediction model to obtain a probability distribution of the remaining useful life; and to obtain a dynamically updated posterior distribution of the remaining life by fusing new monitoring data into a prior distribution through a Bayesian algorithm. The result output module is configured to calculate a future damage rate expectation value to obtain a remaining useful life prediction value by Monte Carlo simulation according to the current cumulative damage degree, the dynamically updated posterior distribution of the remaining life and the modified S-N curve; and to obtain a health state classification result and a corresponding operation and maintenance suggestion by comparing the current cumulative damage degree with a preset damage degree threshold.

Images