Hydropower station pump machine equipment on-line monitoring analysis method and system

Abstract

The invention provides an online monitoring analysis method and system for hydropower station pump equipment, and relates to the technical field of information, and the method comprises the steps: judging that a bolt group has a fatigue failure risk when a fatigue damage degree D value reaches a preset critical threshold value, and comprehensively evaluating the health state of the bolt group in combination with bolt pre-tightening force monitoring data and a loosening failure critical threshold value of the bolt group, obtaining a residual service life prediction result of the bolt group; and by analyzing the stress nephogram and the strain nephogram of the stress concentration area around the bolt hole and the residual service life prediction result, a topological optimization and parameter optimization method is adopted to carry out optimization design on the structural arrangement and the pre-tightening force application scheme of the bolt group, and optimal bolt group design parameters for prolonging the fatigue life are obtained. Accurate prediction and optimal design of the fatigue life of the bolt group are realized, and the operation reliability and safety of the pump are improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an online monitoring and analysis method and system for pump equipment in a hydropower station. Background Art

[0002] In the bolt group connecting the pump bearing seat and the frame, the bolt preload will decay during long-term service, resulting in a decrease in the stiffness of the bolt connection and causing bolt stress redistribution. During the attenuation of the bolt preload, the axial load of the bolt gradually decreases, while the shear load and bending load increase relatively, and the stress state of the bolt changes. At the same time, due to the stress concentration phenomenon around the bolt hole, the stress amplitude in this area is large under the action of alternating loads, which is prone to fatigue damage. With the continuous accumulation of fatigue damage, the microstructure and mechanical properties of the area around the bolt hole degrade, microcracks initiate and gradually expand, and may eventually lead to fatigue fracture failure of the bolt. Therefore, it is urgent to study the stress redistribution law of the bolt group during the attenuation of the bolt preload, reveal the fatigue damage evolution mechanism and failure mode of the stress concentration area around the bolt hole, and establish a bolt fatigue life prediction model to provide guidance for the design optimization and safety assessment of the pump bearing seat bolt group. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides an online monitoring and analysis method for pump equipment in a hydropower station, which mainly includes:

[0004] According to the structural parameters and material properties of the bolt group connecting the pump bearing seat and the frame, a three-dimensional finite element model of the bolt group is established. By applying preload and external load, the stress state and stress distribution of the bolt group under actual working conditions are simulated, and the stress cloud map and strain cloud map of the stress concentration area around the bolt hole are obtained;

[0005] For the stress concentration area around the bolt hole, the stress and strain response data of the area are extracted. According to the actual working conditions of the pump and the load test data, the load spectrum of the bolt group is obtained. The load spectrum is processed by the rain flow counting method to obtain the stress cycle number and amplitude distribution, the material stress and life curve and Miner linear cumulative damage theory are obtained, and the fatigue damage degree D value is calculated;

[0006] Use strain gauges or fiber Bragg grating sensor testing methods to monitor the dynamic strain response of the bolt group in real time during operation, obtain the change data of the bolt preload at different times, combine the stress cloud map and strain cloud map of the stress concentration area around the bolt hole, and determine the loosening failure critical threshold of the bolt group by analyzing the attenuation law of the preload over time;

[0007] When the fatigue damage degree D value reaches the preset critical threshold, it is judged that the bolt group has the risk of fatigue failure. Combined with the bolt preload monitoring data and the loosening failure critical threshold of the bolt group, the health status of the bolt group is comprehensively evaluated to obtain the prediction result of the remaining service life of the bolt group;

[0008] By analyzing the stress cloud map and strain cloud map of the stress concentration area around the bolt hole and the remaining service life prediction results, the topology optimization and parameter optimization methods are used to optimize the structural layout and preload application scheme of the bolt group, and the optimal bolt group design parameters for improving fatigue life are obtained;

[0009] A bolt group failure risk assessment model is established. Combined with the change data of bolt preload, fatigue damage degree D value, stress cloud map and strain cloud map of stress concentration area around the bolt hole, the failure probability of the bolt group is predicted through the machine learning algorithm support vector machine. Combined with the optimal bolt group design parameters, a basis is provided for pump operation and maintenance decision-making.

[0010] The present invention provides an online monitoring and analysis system for pump equipment in a hydropower station, which mainly includes:

[0011] The bolt group 3D modeling module is used to establish a 3D finite element model of the bolt group according to the structural parameters and material properties of the bolt group connecting the pump bearing seat and the frame. By applying preload and external load, the stress state and stress distribution of the bolt group under actual working conditions are simulated to obtain the stress cloud map and strain cloud map of the stress concentration area around the bolt hole.

[0012] The stress-strain response and load spectrum extraction module is used to extract the stress and strain response data of the stress concentration area around the bolt hole, obtain the load spectrum of the bolt group according to the actual working conditions of the pump and the load test data, and use the rain flow counting method to process the load spectrum to obtain the stress cycle number and amplitude distribution, obtain the material stress and life curve and Miner linear cumulative damage theory, and calculate the fatigue damage degree D value;

[0013] The dynamic strain real-time monitoring module is used to use strain gauges or fiber grating sensor testing methods to monitor the dynamic strain response of the bolt group in real time during operation, obtain the change data of the bolt preload at different times, and determine the loosening failure critical threshold of the bolt group by analyzing the attenuation law of the preload over time in combination with the stress cloud map and strain cloud map of the stress concentration area around the bolt hole;

[0014] The health status assessment module is used to determine the risk of fatigue failure of the bolt group when the fatigue damage degree D value reaches the preset critical threshold. It combines the bolt preload monitoring data and the loosening failure critical threshold of the bolt group to comprehensively assess the health status of the bolt group and obtain the prediction result of the remaining service life of the bolt group.

[0015] The structural optimization design module is used to optimize the structural layout and preload application scheme of the bolt group by analyzing the stress cloud map and strain cloud map of the stress concentration area around the bolt hole and the remaining service life prediction results, and adopt topology optimization and parameter optimization methods to obtain the optimal bolt group design parameters for improving fatigue life;

[0016] The failure risk prediction module is used to establish a failure risk assessment model for bolt groups. It combines the change data of bolt preload, fatigue damage degree D value, stress cloud map and strain cloud map of stress concentration area around bolt holes, and uses machine learning algorithm support vector machine to predict the failure probability of bolt groups. Combined with the optimal bolt group design parameters, it provides a basis for pump operation and maintenance decisions.

[0017] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0018] The present invention discloses an online monitoring and analysis method and system for pump equipment in a hydropower station. The method establishes a three-dimensional finite element model of a bolt group, simulates the stress distribution under actual working conditions, and obtains a stress cloud map and a strain cloud map. Combined with the dynamic strain response monitored in real time, the preload attenuation law is analyzed to determine the critical threshold of loosening failure. The load spectrum is processed by the rain flow counting method, and the fatigue damage degree is calculated in combination with the Miner linear cumulative damage theory. When the damage degree reaches the critical value, the health state of the bolt group is comprehensively evaluated to predict the remaining service life. The optimal bolt group design parameters are obtained through topology optimization and parameter optimization. A failure risk assessment model is established using a machine learning algorithm to provide a basis for pump operation and maintenance decisions. The present invention realizes the accurate prediction and optimal design of the fatigue life of the bolt group, and improves the reliability and safety of pump operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The present invention is a flow chart of the on-line monitoring and analysis method for pump equipment in a hydropower station. DETAILED DESCRIPTION

[0020] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0021] like Figure 1 The online monitoring and analysis system for pump equipment in a hydropower station in this embodiment may specifically include:

[0022] S101. According to the structural parameters and material properties of the bolt group connecting the pump bearing seat and the frame, a three-dimensional finite element model of the bolt group is established. By applying preload and external load, the stress state and stress distribution of the bolt group under actual working conditions are simulated, and the stress cloud map and strain cloud map of the stress concentration area around the bolt hole are obtained.

[0023] Receive three-dimensional point cloud data of the connection area between the pump bearing seat and the frame, generate a solid geometric structure based on the three-dimensional point cloud data, and mesh the solid geometric structure using a hexahedral meshing method; calculate the normal stiffness of the contact surface between the frame and the bearing seat based on the mesh data of the solid geometric structure, obtain strain field distribution data on the contact surface between the frame and the bearing seat using a strain gauge array, and obtain the contact stress value from the strain field distribution data; if the contact stress value exceeds the yield limit when the preload force reaches 80% of the axial load, use the Newton-Raphson iterative method to solve the nonlinear equation group, and obtain the node displacement field data from the nonlinear equation group; extract the high stress area unit of the bolt hole wall based on the node displacement field data, use the high stress area unit to establish a stress gradient Gaussian kernel function, and calculate the principal stress direction from the stress gradient Gaussian kernel function; use a displacement sensor array to obtain displacement data in the principal stress direction, and judge the preload attenuation value of the bolt group based on the displacement data.

[0024] Specifically, a three-dimensional digital acquisition point cloud data is established for the connection area between the pump bearing seat and the frame, and the surface contour, roughness and bolt hole position of the connection area are precisely measured. The solid geometry structure is generated based on the point cloud data, and the hexahedral grid unit is divided from the solid geometry structure, and the elastic-plastic constitutive parameters and Coulomb friction parameters are added to the grid unit. According to the grid data of the solid geometry structure, the normal stiffness of the contact surface between the frame and the bearing seat is calculated, and the strain gauge array is arranged on the contact surface to measure the strain field distribution. The contact stress is inversely calculated from the strain data, and the preload loading step is set to 20% of the axial load. The iterative convergence state is judged by the displacement control method. When the preload is applied to 80% of the axial load, if the stress value exceeds the yield limit, the Newton-Raphson iteration is used to solve the nonlinear equation group. The influence of large deformation is taken into account in the solution process, and the node displacement field data is obtained from the nonlinear equation group. The node displacement field data is used to construct the stress cloud map. For the stress concentration area of ​​the bolt hole wall, the high stress area unit is extracted based on the stress cloud map, the stress gradient Gaussian kernel function is established, the stress gradient kernel function is used to calculate the principal stress direction, the displacement sensor array is arranged in the principal stress direction, and the bolt group preload attenuation value is determined from the displacement sensor data. 80 feature points are selected as reference points in the connection area between the bearing seat and the frame, and the point cloud data density is collected at 500 data points per square centimeter. The surface profile detection accuracy of the connection area is set to 0.01 mm, and the bolt hole position tolerance band width is set to 0.02 mm. The point cloud data is fitted with a cubic spline surface to reconstruct the solid geometric structure, and a quadrilateral grid unit is generated on the surface of the solid structure. The grid unit size is set to 2 mm, and the total number of units is controlled within 500,000. The elastic modulus of the material in the elastic-plastic constitutive parameters is 210 GPa, the yield strength is 235 MPa, the Coulomb friction coefficient is 0.15, and the static friction coefficient of the contact surface is 0.2. The contact area unit group is extracted from the solid structure mesh data, and an axial preload of 240 kN is applied to the contact surface boundary. The preload loading is divided into 4 steps, and the displacement convergence threshold is set to 0.001 mm for each step. The spacing of the contact surface strain gauge array is 10 mm, and the strain gauge sensitivity coefficient is 2.1. The strain signal output is measured by the Wheatstone bridge circuit, and the strain sampling frequency is set to 1000 Hz. In the iterative solution process, the stress update criterion is used to judge the convergence, and the stress convergence threshold is set to 0.1 MPa. The node displacement data is extracted from the converged solution to generate a cloud map, and the cloud map color range is set to blue to red corresponding to 0 to 400 MPa. For the stress concentration area of ​​the bolt hole wall, the unit group with a stress value greater than 300 MPa is extracted, and the Gaussian kernel function bandwidth parameter is established as 5. Five displacement sensors are arranged in the 120-degree sector in the principal stress direction. The displacement sensor sensitivity is 0.1 micron, and the sampling time interval is set to 100 milliseconds.

[0025] S102. For the stress concentration area around the bolt hole, extract the stress and strain response data of the area, obtain the load spectrum of the bolt group according to the actual working conditions of the pump and the load test data, process the load spectrum using the rain flow counting method, obtain the stress cycle number and amplitude distribution, obtain the material stress and life curve and Miner linear cumulative damage theory, and calculate the fatigue damage degree D value.

[0026] A displacement sensor is used to obtain strain data at the edge of the bolt hole, and a stress-strain coefficient matrix is ​​obtained according to the strain data and the calibration load; the radial force and axial force of the bearing seat are converted according to the stress-strain coefficient matrix, and a stress-time curve is obtained by recording the radial force and the axial force; the four-point rule is used to count the rain flow count for the stress-time curve, and the number of cycles in the stress level interval is counted from the rain flow counting result, and the number of cycles is normalized by the maximum stress value to obtain a frequency distribution diagram; the cycle ratio of each stress level is extracted from the frequency distribution diagram, and the damage amount is calculated according to the cycle ratio of each stress level and the material stress-life curve data, and the total damage degree value is obtained by summing the damage amounts of each stress level; the total damage degree value formula is:

[0027]

[0028] Where D represents the total damage value, n i represents the cycle ratio of the i-th stress level, N f(i) represents the material fatigue life corresponding to the i-th stress level. If the total damage value exceeds the critical threshold, a support vector machine regressor is used to fit the stress-life curve, and the remaining number of cycles corresponding to the maximum stress level is obtained from the stress-life curve, and the remaining operating time is obtained according to the operating speed.

[0029] Specifically, for the stress concentration area around the bolt hole of the pump, 12 displacement sensors are evenly arranged in a circle around the edge of the bolt hole, and the sampling frequency is set to 2000 Hz. The strain data is continuously collected for 24 hours. The stress-strain coefficient matrix is ​​obtained by applying a calibrated load of 0 to 500 kN to the bolt group. The radial force and axial force data of the bearing seat are converted from the stress-strain matrix, and the stress time-history curve data of the pump start-up and shutdown process are recorded. According to the stress-time history curve data, the four-point rule is used to count the rain flow, identify the closed interval of the stress cycle, divide it into 20 stress level intervals at intervals of 10 MPa, count the number of stress cycles in each stress level interval, and use the maximum stress value of 400 MPa to normalize the cycle number data, and obtain the stress cycle frequency distribution diagram from the normalized data. The cycle ratio of each stress level is extracted from the frequency distribution diagram, and the stress-life curve data of the material under 107 cycles is queried. A 1.2-fold nonlinear correction coefficient is used between the stress-cycle ratio and the life data, and the cycle ratio is substituted into the linear cumulative damage calculation formula. The damage amount is calculated for each of the 20 stress levels, and the total damage value is obtained by summing the damage amount. If the total damage value exceeds the critical threshold of 0.85, the support vector machine regressor is used to fit the stress-life curve, and the remaining life is calculated by extrapolating the stress-life curve. The remaining number of cycles corresponding to the heaviest stress level is extracted from the life calculation results, and the remaining operating time is converted according to the actual operating speed. Twelve resistance strain gauges are evenly arranged on the circumference of a 200 mm diameter circle at the edge of the pump bolt hole. The strain gauge sensitivity coefficient is 2.08, the Wheatstone bridge excitation voltage is 5 volts, and the output signal is collected through 24-bit analog-to-digital conversion with a sampling frequency of 2000 Hz. The 24-hour operation data is continuously recorded to generate a total of 172,800,000 data points. By applying 100, 200, 300, 400, and 500 kN calibration loads to the bolt group in sequence, a stress-strain linear coefficient of 0.86 is obtained. The strain data is multiplied by the linear coefficient to obtain the stress value, and the complete stress-time history curve of the pump from static start-up, stable operation to shutdown is recorded. The stress peak value in the startup process reaches 380 MPa, the stress amplitude in the stable operation section fluctuates between 120 and 260 MPa, and an instantaneous shock of 320 MPa occurs during the shutdown process.The stress-time curve is divided into 20 stress levels. The first level is 30 MPa, the difference between adjacent levels is 10 MPa, and the highest stress level is 220 MPa. The four-point rule is used to extract the stress cycle. When the stress value of the first point among the four adjacent data points is less than the fourth point, it is judged to be a closed cycle. The number of cycles for each stress level is 5620, 8940, 12460, 15780, and 18240 times, respectively. The stress-life curve with a cycle limit of 180 MPa for 107 times of the material is queried. The high stress effect is taken into account with a 1.2 times nonlinear correction coefficient, and the total damage degree is calculated to be 0.92. The kernel function parameter is set to 0.01 in the support vector machine, and the stress-life prediction model is trained to obtain the predicted remaining life of 4.2×105 cycles, which is converted to 290 hours of remaining operating time at a speed of 1450 revolutions per minute.

[0030] S103. Use strain gauges or fiber grating sensor testing methods to monitor the dynamic strain response of the bolt group during operation in real time, obtain the change data of the bolt preload at different times, combine the stress cloud map and strain cloud map of the stress concentration area around the bolt hole, and determine the critical threshold of loosening failure of the bolt group by analyzing the attenuation law of the preload over time.

[0031] The wavelength drift of the fiber grating sensor between the contact surfaces of the pump bolt group is obtained, and the fiber grating sensor is arranged between the contact surfaces at an annular interval; the compensated wavelength drift data is obtained by sliding window processing according to the wavelength drift, and the compensated wavelength drift data is converted to obtain strain data; the dynamic strain standard deviation is obtained by wavelet decomposition processing for the strain data, and the preload force attenuation curve is obtained by least squares position matching according to the dynamic strain standard deviation; the preload force attenuation speed curve is obtained by fitting the preload force attenuation curve through a support vector regressor, and the preload force attenuation rate is calculated according to the preload force attenuation speed curve; if the strain data exceeds the preload force threshold, the stress exceeding standard area unit group is extracted according to the stress cloud map, and the preload force attenuation acceleration is obtained by fitting the stress exceeding standard area unit group through Chebyshev polynomial.

[0032] Specifically, 16 fiber grating sensors were arranged circumferentially at intervals of 22.5 degrees between the contact surfaces of the pump bolt group, using a central wavelength of 1550 nanometers, a grating sensitivity coefficient of 0.8 picometers per microstrain, and a temperature compensation coefficient of 0.1 picometers per degree. The grating wavelength drift was collected by a spectrometer at a sampling frequency of 1000 Hz, and a one-minute sliding window was used to process the wavelength data in real time. The strain data was converted from the compensated wavelength drift data, and the real-time change value of the preload force was obtained based on the strain data. For the strain cloud map data, the db4 wavelet four-level decomposition was used to process the operation data, the mean square deviation of the grating sensor wavelength drift was counted every hour, the dynamic strain standard deviation was calculated through the mean square deviation data, the stress cloud map and the strain cloud map were aligned using the least squares position matching, the preload force attenuation curve was drawn from the standard deviation data, and the strain change rate was solved based on the attenuation curve. According to the strain change rate data, the preload attenuation value within the adjacent hour is calculated, and the support vector regressor is used to fit the preload attenuation speed curve. The kernel function selects the Gaussian kernel and the relaxation factor is 0.01. The preload attenuation rate is calculated from the fitting curve, and the attenuation rate is used to determine whether the preload decrease exceeds 2% per hour. If the strain data exceeds the 85% threshold of the initial preload, the stress cloud map is used to extract the regional unit group where the stress around the bolt hole exceeds 300 MPa, and the fifth-order Chebyshev polynomial is used to fit the preload attenuation curve. The accelerated attenuation rate is calculated from the second-order derivative of the attenuation curve, and the bolt loosening failure is determined based on the preload attenuation acceleration exceeding 4% per square hour. Fiber Bragg grating sensors are arranged every 22.5 degrees on the circumference of the pump bolt group with a diameter of 240 mm. The central wavelength of the sensor is 1550 nanometers, the grating length is 10 mm, and the reflectivity is 98%. The wavelength data is continuously collected at a sampling frequency of 1000 Hz. The wavelength drift caused by ambient temperature fluctuations is 0.15 picometers per degree. The temperature compensation value is recorded in real time. The wavelength drift after compensation varies in the range of 85 to 125 picometers, and the corresponding preload range is 480 to 520 kN. The collected data is processed by a 60-second sliding window. The wavelength data in the window is decomposed by the four-level db4 wavelet. The high-frequency coefficients above the third level are selected to reconstruct the strain signal. The standard deviation of the strain signal is counted once every hour, and the standard deviation value gradually increases from 0.25 microstrain to 0.85 microstrain. The stress cloud map is aligned with the strain cloud map by least squares position matching, and the registration error is less than 0.1 mm. The calculated preload attenuation rate increases from 0.8% per hour to 2.3% per hour. The Gaussian kernel function support vector regression was used to fit the preload attenuation curve, with a relaxation factor of 0.01 and a penalty factor of 100. It was predicted that the preload would drop to 82% of the initial value after 24 hours. Units with stress exceeding 300 MPa were extracted at the edge of the bolt hole, and the fifth-order Chebyshev polynomial was used to fit the unit stress change curve. The maximum value of the second-order derivative reached 4.8% per square hour, which exceeded the critical value of 4% per square hour, indicating that the bolt had loosened and failed.

[0033] S104. When the fatigue damage degree D value reaches the preset critical threshold, it is determined that the bolt group has a risk of fatigue failure. Combined with the bolt preload monitoring data and the loosening failure critical threshold of the bolt group, the health status of the bolt group is comprehensively evaluated to obtain the remaining service life prediction result of the bolt group.

[0034] The stress and strain data of the bolt group are received, and the damage value is obtained by using the modified Miner criterion according to the stress and strain data. If the damage value reaches the critical threshold, the fatigue failure probability distribution is obtained by using the random forest algorithm. The preload monitoring data of the bolt group is obtained, and the preload monitoring data is subjected to Fourier transform, and the preload change slope is obtained by using a Butterworth low-pass filter. The grid point data is constructed according to the preload change slope and the loosening threshold, and a predictor is trained by using the grid point data, and the preload failure moment is determined from the output data of the predictor. The preload failure moment and the fatigue failure risk time are determined, and if the preload failure moment is earlier than the fatigue failure risk time, the damage value and the preload attenuation value are superimposed according to the weighted average principle. The weighted superposition value is calculated by using the stress-life curve, and the remaining number of cycles is obtained from the calculation result, and the final remaining life value is determined according to the remaining number of cycles and the unit speed.

[0035] Specifically, based on the stress and strain data of the bolt group, the modified Miner criterion is used to monitor the damage value in real time. When the damage value reaches the critical threshold of 0.85, the fatigue failure probability distribution is calculated by the random forest algorithm, and the confidence level is set to 95%. The fatigue failure risk time is obtained by extrapolating the data from the SN curve, and the weight coefficient of the damage curve is 0.6. For the preload monitoring data within 24 hours, 4096 points of data are collected for Fourier transform, and the cutoff frequency is set to 20 Hz to extract the low-frequency attenuation trend. The high-frequency component is removed by the Butterworth low-pass filter, and the preload change slope is extracted from the filtered data. The preload data is given a weight coefficient of 0.4. According to the preload attenuation slope and the loosening threshold, 100×100 grid point data composed of time and preload are constructed. The predictor is trained using a three-layer convolution structure, the convolution kernel length is 32, and the activation function uses linear rectification. The output data of the trained predictor is used to determine the failure time when the preload attenuation rate exceeds 2% per hour. If the preload failure time is earlier than the fatigue failure risk time, the damage value and the preload attenuation value are superimposed according to the weighted average principle, and the stress life curve is used to calculate the remaining number of cycles. The operating hours are converted from the unit speed of 1450 rpm to obtain the final remaining life value. During the operation monitoring of the bolt group, the stress value fluctuates between 120 and 360 MPa, and the strain value varies in the range of 0.4 to 1.2 microstrain. The modified Miner criterion is used to calculate the cumulative damage value. The damage curve rises from 0.25 to 0.87 within 168 hours, exceeding the critical threshold of 0.85. The random forest algorithm uses 500 decision trees, and the feature dimensions select stress amplitude, number of cycles, and load frequency. The training obtains the failure probability distribution at 95% confidence level. The failure risk time is extrapolated from the stress life curve to be 312 hours, and the damage data is given a weight of 0.6. The preload data was analyzed by Fourier transform of 4096 points, and the preload attenuation trend was extracted by a fourth-order Butterworth filter with a cutoff frequency of 20 Hz. The preload decreased from the initial value of 520 kN to 452 kN, and the attenuation data was given a weight of 0.4. On the 100×100 grid points formed by the time preload, the convolution kernel sizes of the three-layer convolution network were 32, 16, and 8, respectively, and the number of output channels in each layer was 64, 32, and 16. A linear rectification function with a slope of 0.1 was used to predict that the preload attenuation rate reached 2.3% per hour at 240 hours, exceeding the critical value of 2% per hour. The damage degree and preload attenuation values ​​were superimposed by weighting coefficients of 0.6 and 0.4, and the comprehensive life prediction value was calculated to be 232 hours. The remaining life was converted to 201 hours according to the operating speed of 1450 rpm.

[0036] S105. By analyzing the stress cloud map and strain cloud map of the stress concentration area around the bolt hole and the remaining service life prediction results, the topology optimization and parameter optimization methods are used to optimize the structural layout and preload application scheme of the bolt group to obtain the optimal bolt group design parameters for improving fatigue life.

[0037] According to the stress cloud map data around the bolt hole, the variable density method is used to generate grid cells in the high stress area, the grid cells are deleted or retained by the density threshold, and the bolt arrangement coordinate data is obtained from the grid node distribution; the quadratic polynomial basis function is used to construct the response surface for the bolt arrangement coordinate data and the structural strain distribution, the optimal combination is obtained within the preload search range by the Bayesian optimizer, and the preload sensitivity ranking is obtained from the output data of the Bayesian optimizer; the preload application sequence is optimized by the particle swarm search algorithm according to the preload sensitivity ranking data, and the node displacement field data is obtained by the grid unit group calculation; the structural stiffness distribution matrix is ​​obtained from the node displacement field data, and the stiffness deviation value is calculated by the distribution matrix; if the stiffness deviation value exceeds the preset threshold, the sequential quadratic programming solver is used to iteratively optimize the structural parameters, and the bolt group design parameters are obtained from the converged solution.

[0038] Specifically, based on the stress cloud map data around the bolt hole, the variable density method is used to generate grid cells in the high stress area. The cell size varies in the range of 0.5 to 5 mm. The cells are deleted or retained through the 0.6 density threshold. 200 response sampling points are arranged in the stress concentration area. The bolt arrangement coordinates and aperture size are obtained from the optimized grid node distribution. The life of the arrangement scheme is verified to verify that the life improvement meets the 20% indicator. For the preload data and structural strain distribution, the quadratic polynomial basis function is used to construct the response surface, and the optimal combination of preload is searched in the range of 300 to 600 kN. The number of sampling points is set to 400 through the Bayesian optimizer, the penalty factor is set to 0.1, and the search step is set to 10 kN. The preload sensitivity ranking is extracted from the optimizer output data, and the sensitivity coefficient is normalized. According to the sensitivity normalized data, the preload application sequence is optimized, and a swarm search algorithm with a particle swarm size of 200 and an iteration number of 500 is used. The application time interval is set to 10 seconds, and the grid unit group is divided on the surface of the connection structure. The unit size is 2 mm, and the node displacement field data is calculated. The structural stiffness distribution matrix is ​​obtained from the displacement field data. If the stiffness distribution deviation exceeds 10%, the sequential quadratic programming solver is used to iteratively optimize the structural parameters, set the relative error convergence threshold of 0.1%, and the parameter correction step size of 0.1 mm. The preload value and application sequence are iteratively updated, and the bolt group design parameters are extracted from the converged solution. The parameter scheme is verified by fatigue life calculation. 60,000 tetrahedral mesh units are divided in the area around the bolt hole. The unit size is set to 5 mm in the area with stress below 100 MPa, and is refined to 0.5 mm in the area with stress exceeding 300 MPa. The middle area is linearly gradient according to the stress value. A density threshold of 0.6 is used to delete 42,000 low-stress units, and 18,000 high-stress units are retained. 200 response sampling points are evenly arranged on the remaining units, and the bolt arrangement coordinates are offset by 3.5 mm from the original position, and the hole diameter is optimized from the original 24 mm to 26 mm. The response surface is constructed using 15 quadratic polynomials, and 400 preload sampling points are set in the range of 300 to 600 kN. The Bayesian optimization penalty factor is 0.1, and the search step is 10 kN. The optimal preload combinations are 485, 472, 456, and 442 kN. The particle swarm optimization set 200 particles, the speed was limited to 5 to 15 kN per second, the application time interval was 10 seconds, the connection structure was divided into 86,000 2 mm grid units, and the maximum displacement field was reduced from 1.2 mm to 0.85 mm. The sequence quadratic programming converged after 125 iterations, the relative error was reduced to 0.08%, the parameter correction accumulated displacement of 4.2 mm, the preload adjustment was 72 kN, the application sequence was changed to 13 seconds interval, and the fatigue life was finally increased from the original 232 hours to 298 hours, reaching a 28.4% increase.

[0039] S106. Establish a bolt group failure risk assessment model, combine the bolt preload change data, fatigue damage degree D value, stress cloud map and strain cloud map of the stress concentration area around the bolt hole, and use the machine learning algorithm support vector machine to predict the failure probability of the bolt group. Combined with the optimal bolt group design parameters, provide a basis for pump operation and maintenance decisions.

[0040] The bolt preload attenuation rate, fatigue damage value and hole circumferential stress distribution data are received, and a failure risk assessment matrix is ​​constructed in a three-dimensional feature space according to the data, and a characteristic vector is obtained from the failure risk assessment matrix through principal component analysis; a risk index is constructed according to the characteristic vector, and the risk index is used as failure probability reference data to input a failure probability accumulation function, and a failure risk prediction value is obtained from the failure probability accumulation function; for the failure risk prediction value, a support vector machine failure probability predictor is trained in a three-dimensional space composed of time, preload and damage degree, and a failure probability distribution map is obtained through the failure probability predictor; if the failure probability in the failure probability distribution map exceeds a preset threshold, the preload is redistributed according to the sensitivity sorting according to the optimal design parameters to obtain a preload optimization scheme; a Markov chain state transition matrix is ​​used to predict the maintenance time window for the preload optimization scheme, and a bolt replacement time point is determined by the state transition probability, and a maintenance execution sequence is generated from the bolt replacement time point.

[0041] Specifically, according to the bolt preload attenuation rate, fatigue damage value, and hole circumferential stress distribution data, a 5×5×5 failure risk assessment matrix is ​​constructed in the three-dimensional feature space, and the principal component analysis is used to extract the feature vector. The feature contribution rate threshold is set to 90%. The risk index is constructed through the feature vector, and the risk index is used as the failure probability benchmark data. The benchmark data is input into the failure probability accumulation function. For the stress cloud map and strain cloud map data, the Hu matching invariant moment is used to extract the image features, and the 8×8 grid is divided in the stress concentration area to extract the maximum principal stress direction and strain superposition value. The gradient boosting tree is set to set 500 base learners and the learning rate is 0.1 to construct the risk prediction curve, and the failure risk level is calculated from the prediction curve. According to the failure risk level, a 100×100×100 grid unit group is set in the three-dimensional space composed of time, preload, and damage degree. The failure probability predictor is trained with a support vector machine penalty factor of 100 and a radial basis function bandwidth parameter of 0.1. The failure probability distribution is updated every 4 hours, and the area range where the risk exceeds the 0.8 threshold is calculated from the failure probability distribution map. If the failure probability exceeds the threshold of 0.8, the preload is redistributed according to the sensitivity order according to the optimal design parameters. The Markov chain state transfer matrix is ​​used to predict the 8-hour maintenance time window. The bolt replacement time point is determined by the state transfer probability, and the maintenance execution sequence is generated from the replacement time point data. In the process of bolt group failure risk assessment, a 5×5×5 risk matrix is ​​constructed for the preload attenuation rate of 2.3% per hour, the fatigue damage value of 0.87, and the maximum stress of 320 MPa. The eigenvalues ​​extracted by principal component analysis are 0.82, 0.65, and 0.43, respectively, and the cumulative contribution rate reaches 92.5%. The risk index value calculated from the eigenvector is 0.76. The stress cloud map is divided into 8×8 grids to obtain 64 feature regions. The 72-dimensional feature vector is extracted using Hu matching invariant moments. The maximum principal stress direction deviates from the axial direction by 32 degrees, and the strain superposition value reaches 1.2 microns per millimeter. The failure risk level of 0.83 is obtained by training 500 base learners of the gradient boosting tree. One million cells were generated in the three-dimensional computing grid, the number of support vector machine training samples was 8,000, the penalty factor was 100, the radial basis function bandwidth was 0.1, and the predicted failure probability distribution curve exceeded the 0.8 threshold at 168 hours of operation. The preload of the four bolts was redistributed from high to low sensitivity to 510, 485, 460, and 435 kN. The Markov chain state transition probability matrix predicted that the maintenance time window would be entered at 192 hours, and the maintenance window would last for 8 hours. The optimal replacement time point was calculated to be 196 hours through state probability calculation.

[0042] Image segmentation and feature recognition are performed on the stress cloud map and strain cloud map of the stress concentration area around the bolt hole, and the coordinates and stress-strain values ​​of the key stress concentration points are extracted. The obtained attenuation law of the preload force over time, the critical threshold of loosening failure of the bolt group and the characteristics of the stress concentration point are used as input, and a bolt group failure probability prediction model is established through the machine learning algorithm support vector machine.

[0043] The sliding window is used to process the preload time-varying data, and the variance value and the mean value are calculated from the preload time-varying data. The preload feature quantity is obtained by normalizing the variance value and the mean value by the maximum and minimum values. The fatigue damage degree value is obtained according to the preload feature quantity, and the failure risk is judged from the comparison result of the fatigue damage degree value and the preset threshold value, and the preload failure feature vector is constructed according to the failure risk. For the stress cloud map and strain cloud map data, the stress gradient threshold is used to perform regional growth segmentation, and the coordinate points of the contour line of the stress concentration area are obtained from the regional growth segmentation. The contour lines in the contour line coordinate points are identified by Hough transform, and the stress and strain values ​​are extracted for the inflection points in the contour lines whose curvature is greater than the preset threshold value, and the stress gradient data are obtained around the inflection points. A training data set is constructed according to the preload failure feature vector and the stress gradient data, and a support vector machine classifier is used to train the training data set. The accuracy of the support vector machine classifier is calculated by cross-validation. If the accuracy is lower than the preset threshold, the optimal classification parameters are obtained by grid search.

[0044] Specifically, according to the time-varying data of bolt preload, the variance and mean are calculated using a 60-second sliding window, and 8 preload feature quantities are obtained through maximum and minimum value normalization. When the preload decay rate exceeds 2% per hour, the failure point is recorded, and the failure risk is judged from the fatigue damage threshold of 0.85. The preload failure feature vector is constructed by the risk value, and the stress-strain features are superimposed on the feature vector. For the stress cloud map and strain cloud map data, the stress gradient 20% threshold is used for regional growth segmentation, the contour coordinate points are extracted in the stress concentration area, the contour lines are identified by Hough transform, the stress and strain values ​​are extracted from the inflection points of the contour lines with a curvature greater than 0.5, and the stress gradient data are extracted around the inflection points. 2000 sets of training samples are obtained from the gradient data. According to the preload feature vector and stress concentration features, a training data set is constructed and randomly divided into 80% training set and 20% validation set. The support vector machine classifier is used to train the failure data, and the penalty factor is searched in the range of 1 to 1000. The accuracy is calculated by five-fold cross validation, and the prediction error is calculated from the validation data. If the accuracy is lower than 95%, the support vector machine parameters are optimized by grid search, and a parameter combination is constructed within the range of penalty factor 1 to 1000 and kernel function parameter 0.01 to 100. The parameter value with the smallest error in the validation set is selected from the parameter combination to calculate the predicted value of the failure probability of the bolt group. In the process of preload data monitoring, the initial preload of 520 kN is processed with a 60-second sliding window, and the mean and standard deviation of the preload are calculated to be 486 kN and 24 kN, respectively. After normalization, the 8 preload eigenvalues ​​are 0.92, 0.88, 0.85, 0.82, 0.79, 0.76, 0.74, and 0.71, respectively. The preload decay rate reaches 2.3% per hour in 168 hours, exceeding the 2% per hour threshold, and the fatigue damage value of 0.87 exceeds the 0.85 threshold, generating a failure eigenvector. The stress-strain cloud map was segmented using a 20% stress gradient threshold, and 86 stress concentration area contour points were extracted. The Hough transform identified 12 contour lines, of which 4 inflection points had a curvature of more than 0.5, the maximum stress value at the inflection point was 320 MPa, and the strain value was 1.2 microns per millimeter. The stress gradient was extracted around the inflection point to form 2000 sets of training data. The support vector machine uses a radial basis kernel function, and through five-fold cross-validation, the optimal penalty factor of 520 and kernel function parameter of 0.05 are obtained. There are 1600 training data sets and 400 validation data sets, and the validation set accuracy rate reaches 93.5%. Through grid search, 50 points are set in the penalty factor interval of 1 to 1000, and 40 points are set in the kernel function parameter interval of 0.01 to 100, and the optimal parameter combination penalty factor of 685 and kernel function parameter of 0.08 are obtained. After retraining, the validation set accuracy rate is increased to 96.8%, and the predicted failure probability after 24 hours is 0.82.

[0045] The present invention provides an online monitoring and analysis system for pump equipment in a hydropower station, which mainly includes:

[0046] The bolt group 3D modeling module is used to establish a 3D finite element model of the bolt group according to the structural parameters and material properties of the bolt group connecting the pump bearing seat and the frame. By applying preload and external load, the stress state and stress distribution of the bolt group under actual working conditions are simulated to obtain the stress cloud map and strain cloud map of the stress concentration area around the bolt hole.

[0047] The stress-strain response and load spectrum extraction module is used to extract the stress and strain response data of the stress concentration area around the bolt hole, obtain the load spectrum of the bolt group according to the actual working conditions of the pump and the load test data, and use the rain flow counting method to process the load spectrum to obtain the stress cycle number and amplitude distribution, obtain the material stress and life curve and Miner linear cumulative damage theory, and calculate the fatigue damage degree D value;

[0048] The dynamic strain real-time monitoring module is used to use strain gauges or fiber grating sensor testing methods to monitor the dynamic strain response of the bolt group in real time during operation, obtain the change data of the bolt preload at different times, and determine the loosening failure critical threshold of the bolt group by analyzing the attenuation law of the preload over time in combination with the stress cloud map and strain cloud map of the stress concentration area around the bolt hole;

[0049] The health status assessment module is used to determine the risk of fatigue failure of the bolt group when the fatigue damage degree D value reaches the preset critical threshold. It combines the bolt preload monitoring data and the loosening failure critical threshold of the bolt group to comprehensively assess the health status of the bolt group and obtain the prediction result of the remaining service life of the bolt group.

[0050] The structural optimization design module is used to optimize the structural layout and preload application scheme of the bolt group by analyzing the stress cloud map and strain cloud map of the stress concentration area around the bolt hole and the remaining service life prediction results, and adopt topology optimization and parameter optimization methods to obtain the optimal bolt group design parameters for improving fatigue life;

[0051] The failure risk prediction module is used to establish a failure risk assessment model for bolt groups. It combines the change data of bolt preload, fatigue damage degree D value, stress cloud map and strain cloud map of stress concentration area around bolt holes, and uses machine learning algorithm support vector machine to predict the failure probability of bolt groups. Combined with the optimal bolt group design parameters, it provides a basis for pump operation and maintenance decisions.

[0052] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. An online monitoring and analysis method for pump equipment in a hydropower station, characterized in that: The method comprises: According to the structural parameters and material properties of the bolt group connecting the pump bearing seat and the frame, a three-dimensional finite element model of the bolt group is established. By applying preload and external load, the stress state and stress distribution of the bolt group under actual working conditions are simulated, and the stress cloud map and strain cloud map of the stress concentration area around the bolt hole are obtained; For the stress concentration area around the bolt hole, the stress and strain response data of the area are extracted. According to the actual working conditions of the pump and the load test data, the load spectrum of the bolt group is obtained. The load spectrum is processed by the rain flow counting method to obtain the stress cycle number and amplitude distribution, the material stress and life curve and Miner linear cumulative damage theory are obtained, and the fatigue damage degree D value is calculated; Use strain gauges or fiber Bragg grating sensor testing methods to monitor the dynamic strain response of the bolt group in real time during operation, obtain the change data of the bolt preload at different times, combine the stress cloud map and strain cloud map of the stress concentration area around the bolt hole, and determine the loosening failure critical threshold of the bolt group by analyzing the attenuation law of the preload over time; When the fatigue damage degree D value reaches the preset critical threshold, it is judged that the bolt group has the risk of fatigue failure. Combined with the bolt preload monitoring data and the loosening failure critical threshold of the bolt group, the health status of the bolt group is comprehensively evaluated to obtain the prediction result of the remaining service life of the bolt group; By analyzing the stress cloud map and strain cloud map of the stress concentration area around the bolt hole and the remaining service life prediction results, the topology optimization and parameter optimization methods are used to optimize the structural layout and preload application scheme of the bolt group, and the optimal bolt group design parameters for improving fatigue life are obtained; A bolt group failure risk assessment model is established. Combined with the change data of bolt preload, fatigue damage degree D value, stress cloud map and strain cloud map of stress concentration area around the bolt hole, the failure probability of the bolt group is predicted through the machine learning algorithm support vector machine. Combined with the optimal bolt group design parameters, a basis is provided for pump operation and maintenance decision-making.

2. The method according to claim 1, characterized in that According to the structural parameters and material properties of the bolt group connecting the pump bearing seat and the frame, a three-dimensional finite element model of the bolt group is established, and the stress state and stress distribution of the bolt group under actual working conditions are simulated by applying preload and external load, and the stress cloud map and strain cloud map of the stress concentration area around the bolt hole are obtained, including: Receiving three-dimensional point cloud data of the connection area between the pump bearing seat and the frame, generating a solid geometric structure according to the three-dimensional point cloud data, and meshing the solid geometric structure using a hexahedral meshing method; Calculating the normal stiffness of the contact surface between the frame and the bearing seat according to the grid data of the solid geometric structure, acquiring strain field distribution data on the contact surface between the frame and the bearing seat using a strain gauge array, and obtaining a contact stress value from the strain field distribution data; If the contact stress value exceeds the yield limit when the preload reaches 80% of the axial load, a Newton-Raphson iteration method is used to solve the nonlinear equation group, and node displacement field data is obtained from the nonlinear equation group; Extracting high stress area units of bolt hole wall according to the node displacement field data, establishing stress gradient Gaussian kernel function by using the high stress area units, and calculating principal stress direction from the stress gradient Gaussian kernel function; A displacement sensor array is used to acquire displacement data in the principal stress direction, and the preload attenuation value of the bolt group is determined according to the displacement data.

3. The method according to claim 1, characterized in that The stress concentration area around the bolt hole is extracted to extract the stress and strain response data of the area, and the load spectrum of the bolt group is obtained according to the actual working conditions of the pump and the load test data. The load spectrum is processed by the rain flow counting method to obtain the stress cycle number and amplitude distribution, obtain the material stress and life curve and Miner linear cumulative damage theory, and calculate the fatigue damage degree D value, including: A displacement sensor is used to obtain strain data at the edge of the bolt hole, and a stress-strain coefficient matrix is ​​obtained according to the strain data and the calibration load; Convert the radial force and axial force of the bearing seat according to the stress-strain coefficient matrix, and obtain a stress time history curve by recording the radial force and axial force; The four-point rule is used to count the rain flow according to the stress time history curve, the number of cycles within the stress level interval is counted from the rain flow counting result, and the number of cycles is normalized by the maximum stress value to obtain a frequency distribution diagram; Extracting the cycle ratio of each stress level from the frequency distribution diagram, calculating the damage amount according to the cycle ratio of each stress level and the material stress life curve data, and obtaining the total damage value by summing the damage amounts of each stress level; If the total damage value exceeds a critical threshold, a support vector machine regressor is used to fit the stress-life curve, the remaining number of cycles corresponding to the maximum stress level is obtained from the stress-life curve, and the remaining operating time is obtained according to the operating speed.

4. The method according to claim 1, characterized in that The testing method using strain gauges or fiber grating sensors is used to monitor the dynamic strain response of the bolt group in real time during operation, obtain the change data of the bolt preload at different times, and determine the loosening failure critical threshold of the bolt group by analyzing the attenuation law of the preload over time in combination with the stress cloud map and strain cloud map of the stress concentration area around the bolt hole, including: Obtaining the wavelength drift of the fiber grating sensor between the contact surfaces of the pump bolt group, wherein the fiber grating sensor is arranged between the contact surfaces at an annular interval; According to the wavelength drift amount, compensated wavelength drift data is obtained by sliding window processing, and strain data is obtained by converting the compensated wavelength drift data; The strain data is processed by wavelet decomposition to obtain a dynamic strain standard deviation, and a preload force attenuation curve is obtained by least squares position matching according to the dynamic strain standard deviation; The preload force attenuation curve is adopted to obtain a preload force attenuation speed curve through support vector regressor fitting, and the preload force attenuation rate is calculated according to the preload force attenuation speed curve; If the strain data exceeds the preload threshold, a stress-exceeding region unit group is extracted according to the stress cloud map, and the preload attenuation acceleration is obtained by fitting the stress-exceeding region unit group through Chebyshev polynomials.

5. The method according to claim 1, characterized in that When the fatigue damage degree D value reaches the preset critical threshold, it is judged that the bolt group has a fatigue failure risk. Combined with the bolt preload monitoring data and the loosening failure critical threshold of the bolt group, the health status of the bolt group is comprehensively evaluated to obtain the remaining service life prediction result of the bolt group, including: receiving stress and strain data of the bolt group, obtaining a damage value according to the stress and strain data by using a modified Miner criterion, and obtaining a fatigue failure probability distribution by using a random forest algorithm if the damage value reaches a critical threshold; Obtaining bolt group preload monitoring data, performing Fourier transform on the preload monitoring data, and obtaining a preload change slope through a Butterworth low-pass filter; Constructing grid point data according to the preload change slope and the looseness threshold, training a predictor through the grid point data, and determining the preload failure moment from the output data of the predictor; Determine the preload failure moment and the fatigue failure risk time, and if the preload failure moment is earlier than the fatigue failure risk time, superimpose the damage value and the preload attenuation value according to the weighted average principle; The weighted superposition value is calculated using the stress-life curve, the remaining number of cycles is obtained from the calculation result, and the final remaining life value is determined based on the remaining number of cycles and the unit speed.

6. The method according to claim 1, characterized in that By analyzing the stress cloud map and strain cloud map of the stress concentration area around the bolt hole and the remaining service life prediction results, the topology optimization and parameter optimization methods are used to optimize the structural layout and preload force application scheme of the bolt group, and the optimal bolt group design parameters for improving fatigue life are obtained, including: According to the stress cloud map data around the bolt hole, the variable density method is used to generate grid cells in the high stress area, the grid cells are deleted or retained according to the density threshold, and the bolt arrangement coordinate data is obtained from the grid node distribution; A response surface is constructed using a quadratic polynomial basis function for the bolt arrangement coordinate data and the structural strain distribution, an optimal combination is obtained within the preload search range through a Bayesian optimizer, and a preload sensitivity ranking is obtained from the output data of the Bayesian optimizer; The preload force application sequence is optimized using a particle swarm search algorithm according to the preload force sensitivity ranking data, and node displacement field data is obtained by calculating the grid unit group; Acquire a structural stiffness distribution matrix from the node displacement field data, and calculate a stiffness deviation value through the distribution matrix; If the stiffness deviation value exceeds a preset threshold, a sequential quadratic programming solver is used to iteratively optimize the structural parameters, and the bolt group design parameters are obtained from the converged solution.

7. The method according to claim 1, characterized in that The bolt group failure risk assessment model is established, combined with the change data of bolt preload, fatigue damage degree D value, stress cloud map and strain cloud map of stress concentration area around bolt hole, and the failure probability of bolt group is predicted by support vector machine of machine learning algorithm, combined with the optimal bolt group design parameters, to provide a basis for pump operation and maintenance decision, including: Receiving bolt preload attenuation rate and fatigue damage value and hole circumference stress distribution data, constructing a failure risk assessment matrix in a three-dimensional feature space according to the data, and obtaining a characteristic vector from the failure risk assessment matrix through principal component analysis; constructing a risk index according to the characteristic vector, using the risk index as failure probability reference data to input into a failure probability accumulation function, and obtaining a failure risk prediction value from the failure probability accumulation function; According to the failure risk prediction value, a support vector machine failure probability predictor is trained in a three-dimensional space consisting of time, preload force and damage degree, and a failure probability distribution map is obtained through the failure probability predictor; If the failure probability in the failure probability distribution diagram exceeds a preset threshold, the preload force is redistributed according to the sensitivity ranking based on the optimal design parameters to obtain a preload force optimization solution; A Markov chain state transfer matrix is ​​used to predict the maintenance time window for the preload optimization scheme, a bolt replacement time point is determined by the state transfer probability, and a maintenance execution sequence is generated from the bolt replacement time point.

8. The method according to claim 7, characterized in that The bolt group failure risk assessment model is established, combined with the change data of bolt preload, fatigue damage degree D value, stress cloud map and strain cloud map of stress concentration area around bolt hole, and the failure probability of bolt group is predicted by support vector machine of machine learning algorithm, combined with the optimal bolt group design parameters, to provide a basis for pump operation and maintenance decision, and also includes: Image segmentation and feature recognition are performed on the stress cloud map and strain cloud map of the stress concentration area around the bolt hole, and the coordinates and stress-strain values ​​of the key stress concentration points are extracted. The obtained attenuation law of the preload force over time, the critical threshold of loosening failure of the bolt group and the characteristics of the stress concentration point are used as input, and a bolt group failure probability prediction model is established through the machine learning algorithm support vector machine.

9. The method according to claim 8, characterized in that The obtained stress cloud map and strain cloud map of the stress concentration area around the bolt hole are subjected to image segmentation and feature recognition, the coordinates and stress-strain values ​​of the key stress concentration points are extracted, the obtained attenuation law of the preload force over time, the loosening failure critical threshold of the bolt group and the stress concentration point characteristics are used as input, and a bolt group failure probability prediction model is established through a machine learning algorithm support vector machine, including: The preload time-varying data is processed by using a sliding window, a variance value and a mean value are calculated from the preload time-varying data, and a preload characteristic value is obtained by normalizing the variance value and the mean value by maximum and minimum values; Obtaining a fatigue damage value according to the preload characteristic quantity, judging a failure risk from a comparison result of the fatigue damage value and a preset threshold, and constructing a preload failure characteristic vector according to the failure risk; For the stress cloud map and strain cloud map data, the stress gradient threshold is used to perform regional growth segmentation, and the coordinate points of the contour line of the stress concentration area are obtained from the regional growth segmentation; Identify the contour lines in the coordinate points of the contour line by Hough transformation, extract stress and strain values ​​for inflection points in the contour lines whose curvature is greater than a preset threshold, and obtain stress gradient data around the inflection points; A training data set is constructed according to the preload failure feature vector and the stress gradient data, and a support vector machine classifier is used to train the training data set. The accuracy of the support vector machine classifier is calculated through cross-validation. If the accuracy is lower than a preset threshold, the optimal classification parameters are obtained through grid search. 10.Online monitoring and analysis system for pump equipment in hydropower station, characterized by: The system comprises: The bolt group 3D modeling module is used to establish a 3D finite element model of the bolt group according to the structural parameters and material properties of the bolt group connecting the pump bearing seat and the frame. By applying preload and external load, the stress state and stress distribution of the bolt group under actual working conditions are simulated to obtain the stress cloud map and strain cloud map of the stress concentration area around the bolt hole. The stress-strain response and load spectrum extraction module is used to extract the stress and strain response data of the stress concentration area around the bolt hole, obtain the load spectrum of the bolt group according to the actual working conditions of the pump and the load test data, and process the load spectrum using the rain flow counting method to obtain the stress cycle number and amplitude distribution, obtain the material stress and life curve and Miner linear cumulative damage theory, and calculate the fatigue damage degree D value; The dynamic strain real-time monitoring module is used to use strain gauges or fiber grating sensor testing methods to monitor the dynamic strain response of the bolt group in real time during operation, obtain the change data of the bolt preload at different times, and determine the loosening failure critical threshold of the bolt group by analyzing the attenuation law of the preload over time in combination with the stress cloud map and strain cloud map of the stress concentration area around the bolt hole; The health status assessment module is used to determine the risk of fatigue failure of the bolt group when the fatigue damage degree D value reaches the preset critical threshold. It combines the bolt preload monitoring data and the loosening failure critical threshold of the bolt group to comprehensively assess the health status of the bolt group and obtain the prediction result of the remaining service life of the bolt group. The structural optimization design module is used to optimize the structural layout and preload application scheme of the bolt group by analyzing the stress cloud map and strain cloud map of the stress concentration area around the bolt hole and the remaining service life prediction results, and adopt topology optimization and parameter optimization methods to obtain the optimal bolt group design parameters for improving fatigue life; The failure risk prediction module is used to establish a failure risk assessment model for bolt groups. It combines the change data of bolt preload, fatigue damage degree D value, stress cloud map and strain cloud map of stress concentration area around bolt holes, and uses machine learning algorithm support vector machine to predict the failure probability of bolt groups. Combined with the optimal bolt group design parameters, it provides a basis for pump operation and maintenance decisions.

Images