Unit clearance vibration test method based on transient numerical simulation

Abstract

The invention relates to the technical field of unit gap vibration testing, in particular to a unit gap vibration testing method based on transient numerical simulation, and the method comprises the steps: building a multi-degree-of-freedom dynamic model through obtaining unit structure parameters and operation parameters, carrying out the global optimization of stiffness matrix and damping matrix parameters through employing a genetic algorithm, and generating a high-precision dynamic model; a thermal fluid coupling module is integrated in the model, a coupling weight factor is dynamically corrected in combination with a particle swarm optimization algorithm, a multi-physical-field unsteady-state coupling vibration response equation is generated, and a Pareto optimal solution set is solved in combination with a multi-objective genetic algorithm to optimize gap parameters. According to the invention, real-time evaluation of the vibration state and closed-loop output of fault early warning signals are realized. According to the invention, the problem of insufficient characterization of unsteady state multi-field coupling characteristics in the prior art is solved, and high-precision technical support is provided for online monitoring of rotating machinery.

Description

Technical Field

[0001] The present invention relates to the technical field of unit clearance vibration testing, and in particular to a unit clearance vibration testing method based on transient numerical simulation. Background Art

[0002] During power plant operation, gap vibration within rotating equipment such as turbines and generators directly impacts equipment safety and reliability. A vibration testing method based on transient numerical simulation, by constructing a dynamic model of the unit and simulating transient excitations under actual operating conditions (such as start-up and shutdown, sudden load changes, etc.), can accurately analyze the transfer characteristics of vibration energy within the gap and the dynamic response patterns of key components. This method can identify abnormal vibration modes caused by temperature gradients, fluid impact, or mechanical imbalance, and assist in optimizing gap design parameters and operational control strategies, thereby reducing the risk of unplanned downtime and extending equipment lifecycles. Its application provides theoretical support and technical means for online monitoring, fault warning, and preventive maintenance of power plant units, significantly improving the operational reliability of large rotating machinery.

[0003] In the unit gap vibration testing technology based on transient numerical simulation, the existing methods have limitations in characterizing the non-steady-state coupling characteristics between the dynamic excitation source and the actual operating conditions of the unit. Especially under transient excitation (such as rapid start and shutdown or sudden load changes), it is difficult to accurately quantify the distortion of the vibration energy transfer path caused by the interaction of multiple physical fields, which can easily cause deviations in the vibration response prediction and affect the reliability of the gap vibration test under complex working conditions. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a unit gap vibration testing method based on transient numerical simulation, which is used to solve the problem that the existing transient numerical simulation method is insufficient in characterizing the non-steady-state coupling characteristics between the dynamic excitation source and the actual operating conditions of the unit.

[0005] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows: The present invention provides a method for testing unit clearance vibration based on transient numerical simulation, comprising: Step S1, obtaining structural parameters and operating parameters of the unit, wherein the structural parameters include blade clearance and bearing support stiffness, and the operating parameters include rotational speed and temperature field distribution; Step S2, constructing a multi-degree-of-freedom dynamic model of the unit based on the structural parameters and the operating parameters, performing global parameter optimization on the stiffness matrix and the damping matrix of the multi-degree-of-freedom dynamic model by using a genetic algorithm, and generating an optimized dynamic model; Step S3, integrating a thermal-fluid coupling module into the optimized dynamic model, simulating the multi-field coupled vibration response under unsteady boundary conditions in combination with a transient excitation input function, and dynamically correcting the thermal-fluid coupling weight factor using a particle swarm optimization algorithm to generate a multi-physics field unsteady coupled vibration response equation; Step S4, extracting the vibration energy distribution spectrum of the gap region based on the multi-physics field unsteady-state coupled vibration response equation, identifying the distortion region of the vibration energy transfer path through transfer path analysis, and correcting the transfer path distortion index of the distortion region in real time based on an adaptive fuzzy logic algorithm to generate a corrected vibration energy transfer characteristic; Step S5, establishing a sensitivity mapping relationship between gap geometric parameters and vibration amplitude based on the corrected vibration energy transfer characteristics, and solving the Pareto optimal solution set of gap design parameters and operation control parameters using a multi-objective genetic algorithm to generate a collaborative optimization strategy; Step S6, performing a bench experiment based on the collaborative optimization strategy to obtain measured vibration data, and using a time-frequency domain joint verification strategy to verify the generalization of the optimized dynamic model to generate a verified dynamic model; Step S7: embed the verified dynamic model into the online monitoring system, and use the Kalman filter algorithm to fuse the vibration data collected by the sensor with the prediction results of the dynamic model in real time, dynamically update the model parameters and output the real-time status evaluation and fault warning signal of the unit gap vibration.

[0006] Furthermore, in the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S2 includes: Screening an initial population based on a historical model parameter library, and setting a population size of the genetic algorithm in combination with the number of structural degrees of freedom of the multi-degree-of-freedom dynamics model; Iteratively optimizing the stiffness matrix and the damping matrix of the multi-degree-of-freedom dynamic model through adaptive crossover probability and mutation probability, wherein the mutation probability is graded and adjusted according to the sensitivity analysis results of the stiffness matrix and the damping matrix; When the amplitude of the continuous iterative decrease of the mean square error between the vibration energy predicted by the multi-degree-of-freedom dynamics model and the preset experimental data in the historical model parameter library is less than a preset threshold, the optimized dynamic model parameters are output.

[0007] Furthermore, the unit clearance vibration testing method based on transient numerical simulation of the present invention is characterized in that step S3 comprises: The particle swarm dimension is set according to the number of thermodynamic-fluid coupling fields, and the inertia weight of the particle swarm optimization is adjusted based on a nonlinear decreasing strategy; The vibration energy transfer error of the multi-physics field unsteady-state coupled vibration response equation is used as the optimization objective function, and the search space of the thermodynamic fluid coupling weight factor is dynamically updated through individual learning factors and social learning factors; When the thermodynamic-fluid coupling weight factor converges to a preset energy conservation threshold, a modified multi-physics field unsteady-state coupled vibration response equation is output.

[0008] Furthermore, in the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S4 includes: Define the membership functions and fuzzy rules of high-frequency resonance and low-frequency eddy distortion modes based on the historical fault case library; Calculating the transfer path distortion index of the distortion region according to the dynamic relationship between the frequency domain phase offset and the amplitude attenuation ratio of the multi-physics field unsteady-state coupled vibration response equation; Correcting the amplitude-frequency characteristics of the energy transfer path of the distorted region by a gradient descent method, wherein the correction is based on real-time feedback of the membership function and the fuzzy rule; When the transmission path distortion index is lower than a preset safety threshold, a corrected vibration energy transmission characteristic is output.

[0009] Furthermore, the unit clearance vibration testing method based on transient numerical simulation of the present invention, said step S5, includes: taking the minimum vibration amplitude and the highest operating stability as optimization objectives, combining the sensitivity mapping relationship with the penalty function method to incorporate the clearance design threshold and the operating safety boundary constraint; Generate non-dominated solution sets by simulating binary crossover and polynomial mutation, and select Pareto front solutions based on crowding ranking; When the Pareto front solution covers a preset engineering applicability range determined by the modified vibration energy transfer characteristics, a collaborative optimization strategy is output.

[0010] Furthermore, in the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S6 includes: The K-fold validation dataset is divided based on stratified sampling, covering cold start, load shedding and peak load regulation operation conditions; The frequency domain energy distribution consistency between the simulated data of the optimized dynamic model generated in step S2 and the measured data obtained from the current bench test is analyzed by using the spectral kurtosis difference index; When the time domain mean square error and the frequency domain spectral kurtosis difference of the simulation data are both lower than a preset verification threshold, it is determined that the dynamic model passes the generalization verification.

[0011] Furthermore, in the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S7 includes: According to the accuracy of the vibration displacement sensor and the temperature field infrared monitoring sensor, the process noise covariance matrix of the Kalman filter algorithm is set, and the observation noise covariance is dynamically updated through the prediction residual of the verified dynamic model; Performing covariance weighted fusion on the vibration displacement sensor data, the temperature field infrared monitoring data, and the output value of the verified dynamic model generated in step S6; When the parameter update period of the verified dynamic model is synchronized with the sampling period of the unit control system, a real-time status evaluation and fault warning signal of the unit gap vibration is output.

[0012] Beneficial effects of the present invention: The present invention significantly improves the accuracy and reliability of the unit gap vibration test through phased parameter optimization and dynamic coupling correction. Based on the global optimization mechanism of genetic algorithm and particle swarm optimization algorithm, combined with the historical model parameter library and sensitivity analysis, the stiffness matrix, damping matrix and thermal fluid coupling weight factor of the multi-degree-of-freedom dynamic model are optimized to accurately characterize the influence of multi-field interaction on the vibration energy transfer path under non-steady-state boundary conditions. The transmission path distortion index is corrected in real time through the adaptive fuzzy logic algorithm to suppress high-frequency resonance and low-frequency eddy distortion effects, and the Pareto front solution generated by the multi-objective genetic algorithm is combined to optimize the gap design parameters and operation control strategy. The time-frequency domain joint verification strategy enhances the generalization ability of the dynamic model under complex working conditions such as cold start and load shedding. The Kalman filter algorithm fuses sensor data and model prediction values to achieve real-time evaluation of vibration status and closed-loop output of fault warning signals. The above technical solution effectively solves the problem of insufficient characterization of transient excitation sources and multi-field coupling characteristics in existing methods, and provides high-precision technical support for online monitoring and maintenance of rotating machinery. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0014] Figure 1 A flow chart of a unit clearance vibration testing method based on transient numerical simulation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention are described in detail below in conjunction with the drawings. In order to better understand the purpose of the present invention, the present invention is further described in detail below.

[0016] See also Figure 1 The present invention provides a unit clearance vibration test method based on transient numerical simulation, comprising: Step S1, obtaining structural parameters and operating parameters of the unit, wherein the structural parameters include blade clearance and bearing support stiffness, and the operating parameters include rotational speed and temperature field distribution; In step S1, the turbine's structural parameters are acquired by measuring blade clearance using high-precision 3D scanning technology. This technology, combined with a laser displacement sensor, captures dynamic bearing support stiffness data in real time, providing geometric constraints and mechanical property input for the construction of a multi-degree-of-freedom dynamic model. Blade clearance measurement focuses on the radial and axial clearance distribution between the rotor and stator, using non-contact optical sensors to avoid mechanical interference. Bearing support stiffness is calibrated through loading tests and frequency response function inversion to reflect dynamic support characteristics at different speeds. Among the operating parameters, speed is acquired in real time using Hall sensors or photoelectric encoders, and temperature distribution is monitored using a distributed fiber-optic temperature sensor network. These sensors are arranged axially and circumferentially along the turbine, covering thermal gradient variations in high-temperature and high-pressure areas. The temperature field data is generated through thermal imager calibration and spatial interpolation algorithms to generate a continuous distribution map. This data is transmitted synchronously with the speed signal to the data processing module, forming a time-space-correlated operating parameter database. This parameter acquisition scheme provides high-fidelity input for subsequent model construction and optimization.

[0017] Step S2, constructing a multi-degree-of-freedom dynamic model of the unit based on the structural parameters and the operating parameters, performing global parameter optimization on the stiffness matrix and the damping matrix of the multi-degree-of-freedom dynamic model by using a genetic algorithm, and generating an optimized dynamic model; In step S2, when constructing the multi-degree-of-freedom dynamic model of the unit based on the structural parameters and operating parameters obtained in step S1, the number of degrees of freedom of the model is determined according to the support structure, blade distribution and bearing connection method of the unit rotor, and each degree of freedom corresponds to the translational and rotational motion components of the rotor. The initial parameters of the stiffness matrix are constructed based on the measured data of the bearing support stiffness, and the initial value of the damping matrix is calibrated through frequency domain attenuation experiments. The two together characterize the dynamic characteristics of the unit vibration energy transfer. The initial population of the genetic algorithm is selected from the historical model parameter library. The population size is positively correlated with the number of degrees of freedom of the model. The similarity matching algorithm is used to select historical parameter combinations with a high correlation with the current unit geometric characteristics and operating conditions to improve the efficiency of global optimization.

[0018] The adaptive crossover probability is dynamically adjusted based on the population diversity index. When the individual differences in the population are lower than the preset threshold, the crossover probability is increased to enhance the search breadth. The mutation probability is set in a hierarchical manner based on the sensitivity analysis results of the stiffness matrix and the damping matrix. Matrix elements that significantly affect vibration energy transfer use a low mutation probability to maintain stability, while minor elements have a higher mutation probability to expand the search space. The fitness function uses the mean square error between the model's predicted vibration energy and historical experimental data as an evaluation indicator. When the error decreases by less than the preset convergence threshold in consecutive iterations, the stiffness matrix and damping matrix parameter optimization is determined to be complete, and the optimized dynamic model is output. By correcting the distribution characteristics of stiffness and damping, the optimized model accurately characterizes the dynamic response of the unit under unsteady conditions, providing a high-precision basic model for subsequent multi-field coupling analysis.

[0019] Step S3, integrating a thermal-fluid coupling module into the optimized dynamic model, simulating the multi-field coupled vibration response under unsteady boundary conditions in combination with a transient excitation input function, and dynamically correcting the thermal-fluid coupling weight factor using a particle swarm optimization algorithm to generate a multi-physics field unsteady coupled vibration response equation; In step S3, when integrating the thermodynamic-fluid coupling module into the optimized dynamic model, the finite volume method is used to solve the interaction between the fluid and structural domains. The fluid domain parameters are defined based on the temperature field distribution and flow velocity characteristics in the operating parameters, while the structural domain parameters inherit the optimization results of the stiffness and damping matrices of the dynamic model. The transient excitation input function simulates unsteady boundary conditions (such as thermal shock during cold startup or sudden pressure changes during load rejection). The fluid pressure field, temperature gradient field, and mechanical vibration field are coupled through time stepping to generate transient vibration response data under multi-physics field interaction.

[0020] The particle swarm optimization algorithm's dimensionality is determined by the number of thermodynamic-fluid coupling fields, with each coupling field corresponding to an independent dimension of the particle position vector. Weight factors characterize the distribution of the fluid-structure energy transfer coefficient. A nonlinear decreasing strategy is employed for the inertia weights, with a high initial weight to enhance global search capabilities. This weight decreases exponentially with increasing iterations, gradually shifting to a more refined local search. The optimization objective is the vibration energy transfer error of the multi-physics unsteady coupled vibration response equation. This error is quantified by calculating the interaction loss rate between the kinetic energy in the fluid domain and the strain energy in the structural domain.

[0021] The individual learning factor controls the intensity with which particles track their own historical optimal positions, while the social learning factor guides particles toward the group optimal solution. The two factors work together to dynamically update the search space direction and step size of the weight factor. When the iterative update of the weight factor reduces the energy transfer error below a preset energy conservation threshold (satisfying the energy conservation law between physical fields), the optimization process is considered converged, and a revised multi-physics unsteady coupled vibration response equation is output. This equation accurately characterizes the nonlinear distortion characteristics of the vibration energy transfer path caused by fluid-structure interaction under transient excitation, providing high-precision input for subsequent transfer path analysis.

[0022] Step S4, extracting the vibration energy distribution spectrum of the gap region based on the multi-physics field unsteady-state coupled vibration response equation, identifying the distortion region of the vibration energy transfer path through transfer path analysis, and correcting the transfer path distortion index of the distortion region in real time based on an adaptive fuzzy logic algorithm to generate a corrected vibration energy transfer characteristic; In step S4, when extracting the vibration energy distribution spectrum of the gap region based on the multi-physics field unsteady-state coupled vibration response equation, frequency domain decomposition technology is used to perform wavelet packet transform on the vibration response data, separating the energy density distribution in different frequency bands and generating an energy distribution spectrum with time-frequency characteristics. Transfer path analysis uses an energy flow tracking algorithm to identify the transfer path of vibration energy between the blade gap and the bearing support structure. Combined with path contribution calculations, it locates distortion areas, which manifest as abnormal energy density accumulation or phase mutations within specific frequency bands.

[0023] The adaptive fuzzy logic algorithm, based on the statistical characteristics of high-frequency resonance and low-frequency eddy distortion patterns in a historical fault case database, defines a membership function to quantify the probability distribution of energy anomalies in different frequency bands. Fuzzy rules are then linked to experimental data through expert experience to establish a dynamic mapping relationship between the distortion index, phase offset, and amplitude attenuation ratio. The frequency-domain phase offset is extracted using a Hilbert transform to determine the instantaneous phase difference of the vibration waveform. The amplitude attenuation ratio is calculated based on the ratio of the attenuation rate to the steady-state amplitude. These two factors are combined to form a comprehensive evaluation metric for the transfer path distortion index.

[0024] The gradient descent method uses the distortion index as the optimization objective. Combining the weight coefficients output by the membership function with the correction direction constrained by fuzzy rules, iteratively adjusts the amplitude-frequency characteristic parameters of the distorted region. During the correction process, fuzzy rules provide physical constraints for gradient descent, preventing parameter adjustments from exceeding the feasible domain of actual operating conditions. Simultaneously, the membership function dynamically assigns correction priorities to different frequency bands. When the transfer path distortion index, after iterative optimization, falls below a preset safety threshold (set based on the statistical boundaries of historical safety conditions), the corrected vibration energy transfer characteristic is output. This characteristic characterizes the energy transfer path distribution after distortion suppression and provides a dynamic benchmark for gap parameter optimization.

[0025] Step S5, establishing a sensitivity mapping relationship between gap geometric parameters and vibration amplitude based on the corrected vibration energy transfer characteristics, and solving the Pareto optimal solution set of gap design parameters and operation control parameters using a multi-objective genetic algorithm to generate a collaborative optimization strategy; In step S5, when establishing a sensitivity mapping relationship between gap geometric parameters and vibration amplitude based on the corrected vibration energy transfer characteristics, a parameter perturbation method is used to apply small perturbations to geometric parameters such as blade clearance and bearing support stiffness. The response sensitivity values of each parameter change to the vibration amplitude are calculated through a dynamic model to generate a parameter sensitivity matrix. This matrix quantifies the influence weights of different design parameters on vibration energy and provides a priority basis for multi-objective optimization. The optimization objectives are set as minimum vibration amplitude and maximum operational stability. The penalty function method is combined to convert the clearance design threshold (such as the minimum allowable clearance value) and the operational safety boundary (such as the maximum allowable vibration amplitude) into constraints. The penalty coefficient is used to force the solution set to converge to the engineering feasible domain.

[0026] The multi-objective genetic algorithm maintains population diversity by simulating a binary crossover operation, exchanging parameter encoding segments of parent individuals. A polynomial mutation operation introduces random perturbations to prevent premature convergence and generate a non-dominated solution set that covers the multi-objective trade-offs. A crowding ranking algorithm selects Pareto front solutions based on the density of the solution set in the target space, prioritizing individuals with sparse distribution and wide coverage. The preset engineering applicability range is defined based on the modified vibration energy transfer characteristics, encompassing the vibration amplitude tolerance range and stability threshold under typical operating conditions such as cold start and load rejection. When the Pareto front solution covers all boundary conditions, a collaborative optimization strategy is output. The strategy includes recommended values for the gap geometry parameters and a priority sequence for adjusting the operational control parameters, such as prioritizing the reduction of gap parameters that are most sensitive to vibration amplitude. Through multi-objective optimization and constraint fusion, the above process balances the conflicting requirements of vibration suppression and operational stability, providing an implementable parameter combination solution for bench testing.

[0027] Step S6, performing a bench experiment based on the collaborative optimization strategy to obtain measured vibration data, and using a time-frequency domain joint verification strategy to verify the generalization of the optimized dynamic model to generate a verified dynamic model; In step S6, when conducting bench tests based on the collaborative optimization strategy, a K-fold validation dataset is divided into K-folds using a stratified sampling method, covering cold start, load rejection, and peak-shaving operating conditions. The cold start condition simulates the transient vibration characteristics of the unit in its initial low-temperature state, the load rejection condition corresponds to the dynamic response caused by sudden unloading, and the peak-shaving condition reflects the differences in vibration energy distribution under variable load factors. Stratified sampling ensures that each fold contains representative samples from all operating conditions, preventing the validation results from being biased towards a single operating condition.

[0028] In the time-frequency domain joint verification strategy, the spectral kurtosis difference index extracts the multi-resolution frequency band energy of the simulated data and the test bench measured data through wavelet packet decomposition, calculates the Euclidean distance of the kurtosis value of each frequency band, and quantifies the consistency of the frequency domain energy distribution. The time domain mean square error evaluates the prediction accuracy of the dynamic model in the time dimension by comparing the mean square of the point-to-point difference between the simulated vibration signal and the measured signal. The preset verification threshold is set according to the error statistical distribution of the historical safe operation model. When the time domain mean square error and the frequency domain spectral kurtosis difference of the simulated data are both below the threshold, the optimized dynamic model is judged to have the ability to generalize across operating conditions, and the verified dynamic model is output.

[0029] After verification, model parameters are fine-tuned through iterative calibration and residual feedback from measured data, enhancing the model's ability to capture nonlinear vibration characteristics under unsteady excitation and providing high-confidence input for the online monitoring system. This verification process, through stratified data sampling and the integration of multi-dimensional indicators, enhances the model's robustness and engineering applicability under complex operating conditions.

[0030] Step S7: embed the verified dynamic model into the online monitoring system, and use the Kalman filter algorithm to fuse the vibration data collected by the sensor with the prediction results of the dynamic model in real time, dynamically update the model parameters and output the real-time status evaluation and fault warning signal of the unit gap vibration.

[0031] In step S7, when the verified dynamic model is embedded in the online monitoring system, the process noise covariance matrix of the Kalman filter algorithm is set based on the measurement accuracy of the vibration displacement sensor and the temperature field infrared monitoring sensor. The micron-level resolution of the vibration displacement sensor corresponds to the noise parameters in the mechanical vibration dimension, while the temperature measurement error range of the temperature sensor affects the covariance components in the thermal field dimension. The observation noise covariance is dynamically updated using the model prediction residuals. The residuals are the real-time difference between the sensor's measured values and the model's predicted values. The noise parameters are adaptively adjusted based on the statistical characteristics of the residual sequence, enhancing the algorithm's ability to track non-steady-state conditions.

[0032] Vibration displacement sensor data and infrared temperature field monitoring data are integrated with the dynamic model output using a covariance-weighted fusion algorithm. Covariance weights are dynamically assigned based on the sensor measurement error range and the confidence level of the model's historical prediction residuals. The fused signals comprehensively reflect the interaction between the mechanical vibration state and the thermal field, generating high-confidence vibration amplitude, energy transfer path distortion index, and thermal coupling strength index. Model parameters are dynamically updated via the Kalman gain matrix based on the real-time data stream. This corrects for local deviations in the stiffness and damping matrices under unsteady conditions, ensuring synchronization of the model's prediction accuracy with real-time conditions.

[0033] The parameter update cycle is aligned with the sampling pulse signal of the unit control system through a clock synchronization module. When the model refresh frequency matches the control system sampling rate, real-time status assessment results are output. The status assessment results trigger a graded alarm mechanism based on preset safety thresholds. For example, when the vibration amplitude exceeds the operational safety boundary or the distortion index reaches the preset fault threshold, a warning signal of the corresponding level is generated. This technical solution achieves online monitoring and real-time warning functions for unit interstitial vibration through multi-source data fusion and closed-loop parameter updates.

[0034] In the unit clearance vibration test method based on transient numerical simulation provided by the present invention, step S1 provides basic data for subsequent modeling and optimization by collecting the structural parameters and operating parameters of the unit. Structural parameters include blade clearance and bearing support stiffness, which are obtained through three-dimensional scanning or high-precision displacement sensors; operating parameters include rotational speed and temperature field distribution, wherein the temperature field distribution is monitored in real time by distributed temperature sensors or infrared thermal imaging technology. Step S2 constructs a multi-degree-of-freedom dynamic model of the unit based on the above parameters, and performs global parameter optimization on the stiffness matrix and damping matrix of the model through a genetic algorithm. The initial population of the genetic algorithm is generated based on the screening of the historical model parameter library, and the population size matches the number of model degrees of freedom. The fitness function uses the mean square error between the model predicted vibration energy and the historical experimental data as the evaluation index. When the error reduction rate is lower than the preset threshold, the optimized dynamic model parameters are output.

[0035] Step S3 integrates the thermodynamic-fluid coupling module into the optimized dynamic model and simulates the multi-field coupled vibration response under unsteady boundary conditions using a transient excitation input function. The thermodynamic-fluid coupling module uses the finite volume method to solve the interaction between the fluid domain and the structural domain. The particle swarm optimization algorithm sets the particle dimensions based on the number of thermal fields. The inertia weight is dynamically adjusted based on a nonlinear decreasing strategy. The objective function is the vibration energy transfer error of the multi-physics field equation. The search space of the weight factor is updated through individual learning factors and social learning factors until the weight factor meets the energy conservation threshold, generating a revised multi-physics field unsteady coupled vibration response equation.

[0036] Step S4 extracts the vibration energy distribution spectrum of the gap region based on multi-physics equations and uses transfer path analysis to identify distortion regions within the vibration energy transfer path. The dynamic relationship between the frequency-domain phase offset and the amplitude attenuation ratio is used to calculate the transfer path distortion index. An adaptive fuzzy logic algorithm uses a gradient descent method to correct the amplitude-frequency characteristics of the distortion region based on membership functions and fuzzy rules defined in a historical fault case library. When the distortion index falls below a preset safety threshold, the corrected vibration energy transfer characteristics are output.

[0037] Step S5 establishes a sensitivity mapping between gap geometry parameters and vibration amplitude based on the corrected vibration energy transfer characteristics. Sensitivity analysis quantifies the impact of design parameters on vibration amplitude using a parameter perturbation method. A multi-objective genetic algorithm, with minimum vibration amplitude and maximum operational stability as its goals, incorporates gap design thresholds and operational safety boundary constraints using a penalty function method. Non-dominated solution sets are generated by simulating binary crossover and polynomial mutation. Pareto-front solutions that cover the engineering applicability range are selected based on congestion ranking, and a collaborative optimization strategy is output.

[0038] Step S6 conducts bench tests based on the collaborative optimization strategy, simulating cold start, load shedding, and peak-shaving operating conditions. A K-fold validation dataset is partitioned using stratified sampling. The time-frequency domain joint validation strategy uses wavelet transforms to analyze the time-domain mean square error (MSE). The spectral kurtosis difference metric assesses the consistency of the frequency-domain energy distribution between the simulated and measured data. When both values are below a preset validation threshold, the dynamic model is deemed to have passed generalization validation.

[0039] Step S7 embeds the verified dynamic model into the online monitoring system. The Kalman filter algorithm sets the process noise covariance matrix based on sensor accuracy and dynamically updates the observation noise covariance based on the model prediction residuals. Vibration displacement sensor data and infrared temperature field monitoring data are integrated with the model output using a covariance-weighted fusion algorithm to achieve real-time updating of model parameters. When the model update cycle is synchronized with the unit control system sampling cycle, real-time status assessment and fault warning signals are output, completing closed-loop monitoring and control.

[0040] The above steps form a complete unit gap vibration test and online monitoring system through progressive technical means of parameter optimization, multi-field coupling modeling, path correction and closed-loop verification.

[0041] Specifically, the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S2 includes: Screening an initial population based on a historical model parameter library, and setting a population size of the genetic algorithm in combination with the number of structural degrees of freedom of the multi-degree-of-freedom dynamics model; Iteratively optimizing the stiffness matrix and the damping matrix of the multi-degree-of-freedom dynamic model through adaptive crossover probability and mutation probability, wherein the mutation probability is graded and adjusted according to the sensitivity analysis results of the stiffness matrix and the damping matrix; When the amplitude of the continuous iterative decrease of the mean square error between the vibration energy predicted by the multi-degree-of-freedom dynamics model and the preset experimental data in the historical model parameter library is less than a preset threshold, the optimized dynamic model parameters are output.

[0042] In step S2, the initial population is selected based on a historical model parameter library containing verified stiffness matrix and damping matrix parameter combinations under different operating conditions. A similarity matching algorithm is used to select the historical parameters with the highest correlation with the current unit structural parameters as the initial population for the genetic algorithm. The population size is determined based on the number of structural degrees of freedom in the multi-degree-of-freedom dynamic model, with each degree of freedom corresponding to a certain proportion of individuals in the population to ensure the completeness of the parameter search space. In the adaptive crossover and mutation probability adjustment strategy, the crossover probability is dynamically adjusted based on the population diversity index, and the mutation probability is graded based on the sensitivity analysis results of the stiffness and damping matrices. Matrix elements with high sensitivity have lower mutation probabilities to maintain stability, while elements with low sensitivity have increased mutation probabilities to expand the search range. During the iterative optimization process, the fitness function uses the mean squared error (MSE) between the vibration energy predicted by the multi-degree-of-freedom dynamic model and the preset experimental data in the historical model parameter library as an evaluation metric. When the MSE decreases below a preset threshold in consecutive iterations, the parameters are considered converged and the optimized dynamic model parameters are output. The above process improves the global optimization efficiency through historical data drive and dynamic parameter adjustment mechanism.

[0043] Specifically, the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S3 includes: The particle swarm dimension is set according to the number of thermodynamic-fluid coupling fields, and the inertia weight of the particle swarm optimization is adjusted based on a nonlinear decreasing strategy; The vibration energy transfer error of the multi-physics field unsteady-state coupled vibration response equation is used as the optimization objective function, and the search space of the thermodynamic fluid coupling weight factor is dynamically updated through individual learning factors and social learning factors; When the thermodynamic-fluid coupling weight factor converges to a preset energy conservation threshold, a modified multi-physics field unsteady-state coupled vibration response equation is output.

[0044] In step S3, the number of thermodynamic-fluid coupling fields determines the dimensionality of the particle swarm optimization algorithm. Each coupling field corresponds to an independent optimization dimension in the particle swarm, allowing the particle position vector to fully represent the distribution of weight factors under the interaction of multiple fields. The inertia weight is adjusted using a nonlinear decreasing strategy. Initially, a higher inertia weight is set to enhance global search capabilities. As the number of iterations increases, the inertia weight decreases exponentially, gradually shifting to a localized, refined search to balance the algorithm's convergence speed and accuracy.

[0045] Individual and social learning factors control the intensity with which particles track their own historical optimal positions and the group's optimal position, respectively. By dynamically updating the search space of the thermodynamic-fluid coupling weight factors, particles are guided to converge in regions with minimal vibration energy transfer error. The optimization objective function is constructed based on the vibration energy transfer error of the multi-physics unsteady-state coupled vibration response equation, which is quantified by calculating the energy interaction loss rate between the fluid and structural domains. When the iterative update of the weight factors causes the energy transfer error to meet a preset energy conservation threshold, the particle swarm optimization process is considered converged, and the revised multi-physics unsteady-state coupled vibration response equation is output.

[0046] Specifically, the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S4 includes: Define the membership functions and fuzzy rules of high-frequency resonance and low-frequency eddy distortion modes based on the historical fault case library; Calculating the transfer path distortion index of the distortion region according to the dynamic relationship between the frequency domain phase offset and the amplitude attenuation ratio of the multi-physics field unsteady-state coupled vibration response equation; Correcting the amplitude-frequency characteristics of the energy transfer path of the distorted region by a gradient descent method, wherein the correction is based on real-time feedback of the membership function and the fuzzy rule; When the transmission path distortion index is lower than a preset safety threshold, a corrected vibration energy transmission characteristic is output.

[0047] In step S4, the historical fault case library defines the typical characteristics of high-frequency resonance and low-frequency eddy distortion modes by integrating vibration anomaly data recorded by the unit under various operating conditions. The membership function is constructed based on the distribution probability of vibration energy in a specific frequency band. The fuzzy rules are associated with historical data through expert experience to quantify the judgment logic of transmission path anomalies under different distortion modes. The frequency domain phase offset is extracted through the Fourier transform of the multi-physics field non-steady-state coupled vibration response equation, reflecting the phase lag characteristics during the vibration energy transfer process; the amplitude attenuation ratio is calculated based on the attenuation rate of the peak energy in the frequency domain response spectrum. The two are dynamically associated to form a comprehensive evaluation indicator of the transmission path distortion index.

[0048] The gradient descent method uses the distortion index as the optimization objective function and combines the membership function with the real-time feedback signal output by the fuzzy rules to adjust the amplitude-frequency characteristic parameters of the energy transfer path. During the correction process, the fuzzy rules provide constraints on the gradient direction to prevent the local optimal solution from deviating from the actual physical characteristics. At the same time, the membership function dynamically weights the correction weights of different frequency bands to improve the targetedness of the path correction. When the transfer path distortion index is lower than the preset safety threshold after iterative optimization, the transfer path is determined to be in a stable state, and the corrected vibration energy transfer characteristics are output. The preset safety threshold is set based on the statistical characteristics of safe operating conditions in the historical fault case library, which meets the safety boundary requirements of the actual operation of the unit.

[0049] The above technical solution accurately identifies and suppresses the distortion effect of the vibration energy transfer path through fault mode quantification, dynamic indicator calculation and adaptive correction mechanism.

[0050] Specifically, the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S5, includes: taking the minimum vibration amplitude and the highest operating stability as optimization objectives, combining the sensitivity mapping relationship with the penalty function method to incorporate the clearance design threshold and the operating safety boundary constraint; Generate non-dominated solution sets by simulating binary crossover and polynomial mutation, and select Pareto front solutions based on crowding ranking; When the Pareto front solution covers a preset engineering applicability range determined by the modified vibration energy transfer characteristics, a collaborative optimization strategy is output.

[0051] In step S5, minimum vibration amplitude and maximum operational stability are used as core indicators for multi-objective optimization. The dynamic correlation between gap geometry parameters and vibration amplitude is quantified through sensitivity mapping. Sensitivity analysis employs a parameter perturbation method, applying small perturbations to the gap design parameters and calculating the rate of change of the vibration response. This generates a parameter sensitivity matrix, which provides a basis for optimizing weight allocation. A penalty function method transforms the gap design threshold and operational safety boundary constraints into penalty terms in the objective function. When parameter combinations exceed the allowable range, a penalty coefficient is added to force the solution set to converge to the feasible region.

[0052] Simulated binary crossover and polynomial mutation operations generate non-dominated solution sets in a multi-objective genetic algorithm. Simulated binary crossover maintains population diversity by probabilistically exchanging parameter encoding segments of parent individuals. Polynomial mutation introduces local randomness based on probabilistic perturbations to prevent premature convergence. During the non-dominated solution set screening phase, a crowding ranking algorithm is used to calculate the density index of individuals in the solution set, prioritizing individuals with sparse distribution in the target space to form a Pareto front solution with broad coverage and uniform distribution.

[0053] The preset project applicability range is set based on the modified vibration energy transfer characteristics, encompassing the vibration amplitude and stability tolerance range under typical unit operating conditions. The Pareto front solution must cover all boundary conditions within this range. When the optimal solution set satisfies the coverage verification, a collaborative optimization strategy is output, including recommended value ranges for clearance design parameters and a priority sequence for adjusting operational control parameters. This process balances the conflicting requirements of vibration suppression and operational stability through multi-objective optimization and constraint fusion.

[0054] Specifically, the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S6 includes: The K-fold validation dataset is divided based on stratified sampling, covering cold start, load shedding and peak load regulation operation conditions; The frequency domain energy distribution consistency between the simulated data of the optimized dynamic model generated in step S2 and the measured data obtained from the current bench test is analyzed by using the spectral kurtosis difference index; When the time domain mean square error and the frequency domain spectral kurtosis difference of the simulation data are both lower than a preset verification threshold, it is determined that the dynamic model passes the generalization verification.

[0055] In step S6, a stratified sampling method divides the dataset based on the typical characteristics of the unit's operating conditions, with cold start, load rejection, and peak shaving conditions each serving as an independent subset. A K-fold validation dataset is constructed by proportionally sampling to ensure that each fold covers all operating conditions and avoid validation bias. K-fold cross-validation evaluates the generalization ability of the dynamic model under different data distributions by rotating the training and validation sets, improving the statistical significance of the validation results.

[0056] The spectral kurtosis difference metric is used to quantify the consistency of the frequency domain energy distribution between simulated data and bench test data. Spectral kurtosis characterizes the energy concentration characteristics of the signal by calculating the kurtosis value in a specific frequency band. The difference metric is constructed based on the Euclidean distance of the kurtosis values in each frequency band. Frequency domain energy distribution consistency analysis is combined with wavelet packet decomposition to extract multi-resolution frequency band energy. The energy contribution of each frequency band in the simulated and measured data is compared to verify the dynamic model's ability to capture nonlinear vibration characteristics under transient excitation.

[0057] The time-domain mean square error reflects the prediction accuracy of the dynamic model in the time dimension and is obtained by calculating the mean square of the point-to-point differences between the simulated vibration signal and the measured signal. The frequency-domain spectral kurtosis difference evaluates the model's accuracy in restoring complex vibration modes from the perspective of frequency-domain energy distribution. The preset verification threshold is set based on the error distribution of the safe operation model in historical verification cases. When both indicators are below the threshold, the dynamic model is judged to meet the engineering application requirements in terms of its time-frequency domain characteristics, and the model parameters are output as generalization-verified. The above process strengthens the model's robustness under complex working conditions through layered data verification and multi-dimensional indicator evaluation.

[0058] Specifically, the unit clearance vibration testing method based on transient numerical simulation of the present invention, step S7 includes: According to the accuracy of the vibration displacement sensor and the temperature field infrared monitoring sensor, the process noise covariance matrix of the Kalman filter algorithm is set, and the observation noise covariance is dynamically updated through the prediction residual of the verified dynamic model; Performing covariance weighted fusion on the vibration displacement sensor data, the temperature field infrared monitoring data, and the output value of the verified dynamic model generated in step S6; When the parameter update period of the verified dynamic model is synchronized with the sampling period of the unit control system, a real-time status evaluation and fault warning signal of the unit gap vibration is output.

[0059] In step S7, the process noise covariance matrix of the Kalman filter algorithm is set based on the measurement accuracy of the vibration displacement sensor and the temperature field infrared monitoring sensor. The accuracy of the vibration displacement sensor determines the covariance component of the process noise in the mechanical vibration dimension, while the accuracy of the temperature sensor affects the noise parameter allocation in the thermal field dimension. The observed noise covariance is dynamically updated using the prediction residuals of the verified dynamic model. The residuals are the real-time difference between the model predictions and the sensor measurements. The noise covariance matrix is adaptively adjusted based on the statistical characteristics of the residual sequence, improving the filter algorithm's ability to track non-steady-state conditions.

[0060] Vibration displacement sensor data and infrared temperature field monitoring data are integrated with the verified dynamic model output using a covariance-weighted fusion algorithm. Covariance weights are dynamically assigned based on the real-time confidence level of each data source. Confidence is determined by combining the sensor measurement error range with historical statistical results of the model's prediction residuals. The fused signal comprehensively reflects the interaction between the mechanical vibration state and the thermal field, generating a highly confident characteristic vector for the unit's clearance vibration.

[0061] The parameter update cycle of the verified dynamic model is aligned with the sampling cycle of the unit control system through a clock synchronization module. Model parameters are refreshed in real time based on the sensor data stream. When the model update cycle is synchronized with the control system sampling pulse signal, real-time state assessment results are output. State assessment results include vibration amplitude, energy transfer path distortion index, and thermal coupling strength index. Fault warning signals are dynamically generated by comparing the assessment results with preset safety thresholds, triggering a graded alarm mechanism. This technical solution achieves online monitoring and real-time warning of unit interstitial vibration through multi-source data fusion and closed-loop parameter updates.

[0062] Structural parameters and operating parameters: Structural parameters include blade clearance (the physical gap between the blades and the casing, which affects the airflow dynamics) and bearing support stiffness (the bearing's support stiffness for the rotor system, which determines the vibration transmission efficiency), which are measured by 3D scanning or high-precision displacement sensors.

[0063] Operating parameters include speed (rotor rotation frequency, which affects the main vibration frequency) and temperature field distribution (gap changes caused by thermal expansion, which are monitored in real time through distributed temperature sensors or infrared thermal imaging technology).

[0064] Multi-degree-of-freedom dynamic model: A mathematical model based on the unit's structural characteristics describes the system's vibration response through multiple degrees of freedom (such as translation and rotation). The stiffness matrix characterizes the system's ability to resist deformation, while the damping matrix reflects the vibration energy dissipation characteristics. Both are globally optimized using a genetic algorithm. The genetic algorithm uses a historical model parameter library to select an initial population and iteratively adjusts parameters through crossover and mutation operations. The fitness function uses the mean squared error between the model's predicted vibration energy and experimental data as the evaluation metric. Sensitivity analysis uses a hierarchical adjustment of mutation probabilities to balance search efficiency and stability.

[0065] Thermodynamic-Fluid Coupling Module simulates the interaction between fluid fields (such as steam or gas) and structural fields. Transient excitation input functions simulate unsteady boundary conditions (such as startup and shutdown, and sudden load changes). A particle swarm optimization algorithm sets dimensions based on the number of coupled fields. Inertia weights are adjusted using a nonlinear decreasing strategy (large-scale search in the early stages, followed by fine-grained convergence in the later stages). The objective function is vibration energy transfer error. The weights are dynamically adjusted through individual learning factors (particles' own experience) and social learning factors (the optimal solution of the swarm) until the energy conservation threshold is met (the energy interaction loss rate meets the target).

[0066] The Transfer Path Distortion Index (TDI) is calculated based on the frequency-domain characteristics extracted from the multi-physics field unsteady-state coupled vibration response equation. The TDI is dynamically correlated with the frequency-domain phase offset (phase lag of the vibration waveform) and the amplitude attenuation ratio (the rate of amplitude attenuation during energy transfer). An adaptive fuzzy logic algorithm uses a historical fault case library to define membership functions (quantifying the probability of energy anomalies in different frequency bands) and fuzzy rules (such as high-frequency resonance triggering conditions). This algorithm, combined with the gradient descent method, corrects the amplitude-frequency characteristics to suppress abnormal energy accumulation in distorted areas.

[0067] Multi-objective genetic algorithm and Pareto front solution: Sensitivity mapping relationships quantify the influence of gap geometry parameters on vibration amplitude through parameter perturbation. A penalty function method transforms design thresholds (such as the minimum allowable gap value) and safety margins (such as the maximum vibration amplitude) into constraints. Binary crossover (parameter encoding segment exchange) and polynomial mutation (local random perturbation) are simulated to generate a non-dominated solution set. Crowd-ranking is used to select evenly distributed Pareto front solutions that cover the engineering applicability range defined by modified energy characteristics (such as the stability tolerance range under peak load conditions).

[0068] A joint time-frequency domain validation strategy uses stratified sampling to create a K-fold data set covering typical operating conditions such as cold start (low-temperature initial state) and load rejection (sudden unloading). Spectral kurtosis difference is determined by extracting multi-band energy distribution through wavelet packet decomposition. The difference in frequency-domain kurtosis between simulated and bench test data is calculated, and the model's generalization is verified using time-domain mean square error (point-to-point signal difference).

[0069] Kalman filtering and real-time monitoring: The process noise covariance matrix is set based on sensor accuracy (e.g., micron-level error for displacement sensors), and the observed noise covariance is dynamically updated based on model prediction residuals (the difference between measured and predicted values). Covariance weighted fusion assigns weights based on sensor confidence (error range) and historical model residuals, integrating vibration displacement and temperature field data with model output values. Parameter update cycles are synchronized with the control system through a clock module, providing real-time output of vibration amplitude, distortion index, and early warning signals, triggering a tiered alarm mechanism.

[0070] The specific implementation of the present invention is based on the actual application scenario of the unit clearance vibration test, combined with the dynamic characteristics requirements of rotating equipment such as steam turbines and generators under cold start-up, load rejection and peak-shaving operation conditions, to construct a phased optimization and verification system. First, the structural parameters such as the unit blade clearance and bearing support stiffness are collected through three-dimensional scanning or high-precision displacement sensors, and the speed and temperature field distribution are monitored in real time in combination with distributed temperature sensors. The initial population that matches the current unit structure is screened based on the historical model parameter library, and the genetic algorithm population size is set according to the number of structural degrees of freedom of the multi-degree-of-freedom dynamic model. The stiffness matrix and damping matrix parameters are optimized through the adaptive crossover probability and the sensitivity graded adjustment of the mutation probability. When the decrease in the mean square error between the model-predicted vibration energy and the historical experimental data is lower than the preset threshold, the optimized dynamic model is output.

[0071] When integrating the thermal-fluid coupling module into the dynamic model, the particle swarm optimization algorithm sets the dimensions based on the number of thermal fields, adjusts the inertia weight according to a nonlinear decreasing strategy, and the objective function is the vibration energy transfer error of the multi-physics field unsteady-state coupled vibration response equation. The weight factor search space is dynamically updated through individual learning factors and social learning factors. When the energy transfer error meets the preset energy conservation threshold, a revised multi-physics field equation is generated. Based on this equation, the vibration energy distribution spectrum of the gap region is extracted. Combined with the membership functions and fuzzy rules of the high-frequency resonance and low-frequency eddy distortion modes in the historical fault case library, the transfer path distortion index is calculated. The amplitude-frequency characteristics are corrected under the constraints of the fuzzy rules using the gradient descent method, and the revised vibration energy transfer characteristics are output.

[0072] Based on the corrected characteristics, a sensitivity mapping relationship between gap geometry parameters and vibration amplitude was established. A multi-objective genetic algorithm, targeting minimum vibration amplitude and maximum operational stability, was combined with a penalty function method to incorporate design thresholds and operational safety constraints, generating a Pareto front solution covering the engineering applicability range. A K-fold validation dataset covering cold start, load rejection, and peak-shaving conditions was generated through stratified sampling. The spectral kurtosis difference metric was used to analyze the consistency between simulation data and bench test data. The model was deemed to have passed generalization verification when both the time-domain mean square error and frequency-domain difference were below preset thresholds. The validated model was embedded in the online monitoring system. The Kalman filter algorithm set the noise covariance matrix based on sensor accuracy, dynamically integrating vibration displacement sensor and temperature field infrared monitoring data with model predictions. Covariance weights were dynamically assigned based on confidence levels. Real-time status assessment and fault warning signals were output during the parameter update cycle, synchronized with the unit control system, forming a closed-loop monitoring and optimization system for non-steady-state conditions.

[0073] The present invention significantly improves the interactive characterization capability of dynamic excitation sources and non-steady-state working conditions through phased parameter optimization and multi-physics field coupling modeling. First, the stiffness matrix and damping matrix of the multi-degree-of-freedom dynamic model are globally optimized based on the genetic algorithm, and the initial population is screened in combination with the historical parameter library. The mutation probability is adjusted through sensitivity grading to optimize the accuracy and stability of the dynamic model. Secondly, a thermal fluid coupling module and a transient excitation input function are introduced, and the particle swarm optimization algorithm is used to dynamically correct the coupling weight factor. The multi-physics field non-steady-state coupling vibration response equation is generated based on the energy conservation threshold to accurately quantify the vibration energy transfer characteristics under fluid-structure interaction.

[0074] To address the problem of vibration energy transfer path distortion, an adaptive fuzzy logic algorithm is used to correct the transfer path distortion index in real time. Based on a historical fault case library, membership functions and fuzzy rules for high-frequency resonance and low-frequency eddy distortion modes are defined. The distortion index is calculated by combining the dynamic relationship between frequency-domain phase offset and amplitude attenuation ratio. A gradient descent method is used to correct the amplitude-frequency characteristics under the constraints of fuzzy rules, suppressing path distortion caused by unsteady-state excitation. The corrected vibration energy transfer characteristics provide a dynamic benchmark for gap parameter optimization, enhancing the closed-loop control capability of multi-physics field coupling effects.

[0075] The engineering applicability of the model is ensured through joint time-frequency domain validation and closed-loop integration with online monitoring. Bench tests utilize stratified sampling to partition the validation dataset into K-folds. The spectral kurtosis difference metric assesses the consistency of the frequency-domain energy distribution between simulated and measured data, and the time-domain mean square error (MSE) is combined to verify the model's generalizability. The validated dynamic model is embedded in the online monitoring system. A Kalman filter algorithm dynamically fuses sensor data with model predictions, assigning covariance weights based on confidence levels. Synchronous update cycles enable real-time vibration status assessment and fault warning, forming a closed-loop monitoring system for non-steady-state operating conditions.

Claims

1. A unit clearance vibration test method based on transient numerical simulation, characterized in that: include: Step S1, obtaining structural parameters and operating parameters of the unit, wherein the structural parameters include blade clearance and bearing support stiffness, and the operating parameters include rotational speed and temperature field distribution; Step S2, constructing a multi-degree-of-freedom dynamic model of the unit based on the structural parameters and the operating parameters, performing global parameter optimization on the stiffness matrix and the damping matrix of the multi-degree-of-freedom dynamic model by using a genetic algorithm, and generating an optimized dynamic model; Step S3, integrating a thermal-fluid coupling module into the optimized dynamic model, simulating the multi-field coupled vibration response under unsteady boundary conditions in combination with a transient excitation input function, and dynamically correcting the thermal-fluid coupling weight factor using a particle swarm optimization algorithm to generate a multi-physics field unsteady coupled vibration response equation; Step S4, extracting the vibration energy distribution spectrum of the gap region based on the multi-physics field unsteady-state coupled vibration response equation, identifying the distortion region of the vibration energy transfer path through transfer path analysis, and correcting the transfer path distortion index of the distortion region in real time based on an adaptive fuzzy logic algorithm to generate a corrected vibration energy transfer characteristic; Step S5, establishing a sensitivity mapping relationship between gap geometric parameters and vibration amplitude based on the corrected vibration energy transfer characteristics, and solving the Pareto optimal solution set of gap design parameters and operation control parameters using a multi-objective genetic algorithm to generate a collaborative optimization strategy; Step S6, performing a bench experiment based on the collaborative optimization strategy to obtain measured vibration data, and using a time-frequency domain joint verification strategy to verify the generalization of the optimized dynamic model to generate a verified dynamic model; Step S7: embed the verified dynamic model into the online monitoring system, and use the Kalman filter algorithm to fuse the vibration data collected by the sensor with the prediction results of the dynamic model in real time, dynamically update the model parameters and output the real-time status evaluation and fault warning signal of the unit gap vibration.

2. The unit clearance vibration testing method based on transient numerical simulation according to claim 1 is characterized in that: The step S2 includes: Screening an initial population based on a historical model parameter library, and setting a population size of the genetic algorithm in combination with the number of structural degrees of freedom of the multi-degree-of-freedom dynamics model; Iteratively optimizing the stiffness matrix and the damping matrix of the multi-degree-of-freedom dynamic model through adaptive crossover probability and mutation probability, wherein the mutation probability is graded and adjusted according to the sensitivity analysis results of the stiffness matrix and the damping matrix; When the amplitude of the continuous iterative decrease of the mean square error between the vibration energy predicted by the multi-degree-of-freedom dynamics model and the preset experimental data in the historical model parameter library is less than a preset threshold, the optimized dynamic model parameters are output.

3. The unit clearance vibration testing method based on transient numerical simulation according to claim 2 is characterized in that: The step S3 comprises: The particle swarm dimension is set according to the number of thermodynamic-fluid coupling fields, and the inertia weight of the particle swarm optimization is adjusted based on a nonlinear decreasing strategy; The vibration energy transfer error of the multi-physics field unsteady-state coupled vibration response equation is used as the optimization objective function, and the search space of the thermodynamic fluid coupling weight factor is dynamically updated through individual learning factors and social learning factors; When the thermodynamic-fluid coupling weight factor converges to a preset energy conservation threshold, a modified multi-physics field unsteady-state coupled vibration response equation is output.

4. The unit clearance vibration testing method based on transient numerical simulation according to claim 3 is characterized in that: The step S4 comprises: Define the membership functions and fuzzy rules of high-frequency resonance and low-frequency eddy distortion modes based on the historical fault case library; Calculating the transfer path distortion index of the distortion region according to the dynamic relationship between the frequency domain phase offset and the amplitude attenuation ratio of the multi-physics field unsteady-state coupled vibration response equation; Correcting the amplitude-frequency characteristics of the energy transfer path of the distorted region by a gradient descent method, wherein the correction is based on real-time feedback of the membership function and the fuzzy rule; When the transmission path distortion index is lower than a preset safety threshold, a corrected vibration energy transmission characteristic is output.

5. The unit clearance vibration testing method based on transient numerical simulation according to claim 4 is characterized in that: The step S5 includes: taking the minimum vibration amplitude and the highest operating stability as optimization objectives, combining the sensitivity mapping relationship with the penalty function method to incorporate the clearance design threshold and the operating safety boundary constraint; Generate non-dominated solution sets by simulating binary crossover and polynomial mutation, and select Pareto front solutions based on crowding ranking; When the Pareto front solution covers a preset engineering applicability range determined by the modified vibration energy transfer characteristics, a collaborative optimization strategy is output.

6. The unit clearance vibration testing method based on transient numerical simulation according to claim 5 is characterized in that: The step S6 comprises: The K-fold validation dataset is divided based on stratified sampling, covering cold start, load shedding and peak load regulation operation conditions; The frequency domain energy distribution consistency between the simulated data of the optimized dynamic model generated in step S2 and the measured data obtained from the current bench test is analyzed by using the spectral kurtosis difference index; When the time domain mean square error and the frequency domain spectral kurtosis difference of the simulation data are both lower than a preset verification threshold, it is determined that the dynamic model passes the generalization verification.

7. The unit clearance vibration testing method based on transient numerical simulation according to claim 6 is characterized in that: The step S7 includes: According to the accuracy of the vibration displacement sensor and the temperature field infrared monitoring sensor, the process noise covariance matrix of the Kalman filter algorithm is set, and the observation noise covariance is dynamically updated through the prediction residual of the verified dynamic model; Performing covariance weighted fusion on the vibration displacement sensor data, the temperature field infrared monitoring data, and the output value of the verified dynamic model generated in step S6; When the parameter update period of the verified dynamic model is synchronized with the sampling period of the unit control system, a real-time status evaluation and fault warning signal of the unit gap vibration is output.

Images