Nuclear power equipment progressive failure event prediction method based on parallel feedback Kalman filtering model
By adopting a parallel feedback Kalman filtering model in nuclear power equipment, combining nonlinear Wiener process and physical mechanism model, effective prediction of progressive failure events of nuclear power equipment is achieved, the accuracy and reliability of prediction are improved, and the safe operation of nuclear power is ensured.
Patent Information
- Application Number
- CN202510647495.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-20
- Publication Date
- 2025-06-17
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art is difficult to effectively predict the progressive failure event of nuclear power equipment, and traditional physical mechanism models are difficult to cope with uncertainty under complex operating conditions, while data-driven models have shortcomings in small samples and model interpretability.
Using a method based on the parallel feedback Kalman filtering model, the state model of the nonlinear Wiener process and the observation model of the physical mechanism model are constructed. The mechanism and data perception are fusion through the Kalman filter, the degradation state is estimated in real time, and the parameters are adaptively adjusted through the traceless Kalman filter.
It improves the accuracy and reliability of the prediction of progressive failure events of nuclear power equipment, solves the problems of model mismatch, noise sensitivity, and long-term prediction in traditional methods, ensures the safe operation of nuclear power and reduces operation and maintenance costs.
Smart Images
Figure CN120163072A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of nuclear power equipment progressive failure event prediction, and in particular to a nuclear power equipment progressive failure event prediction method based on a parallel feedback Kalman filter model. Background Art
[0002] With the continuous development of modern industrial technology, some core components of many large-scale unit equipment are closely related to the reliability of the unit. Once these components fail mechanically, it will directly lead to equipment failure, which will in turn cause significant losses. Therefore, real-time status monitoring and performance degradation assessment of these key components are of vital importance to ensure the reliability of equipment operation.
[0003] Performance degradation analysis of high-reliability, long-life components is crucial for the healthy operation of the unit. The degradation paths of large unit components generally show dynamic and random characteristics, and exhibit different degradation patterns under different operating conditions and working conditions. Traditional physical mechanism models and artificial intelligence methods have certain limitations when dealing with such complex problems. For example, physical mechanism models have difficulties in expressing uncertainty, while artificial intelligence methods have high requirements on the quality and quantity of data, and the model has poor interpretability. At present, there is little research on the integration of multiple methods for degradation modeling, and the existing hybrid methods are mostly limited to the combination of similar methods, lacking the complementary advantages of cross-methods.
[0004] Component-level reliability analysis is the basic link to ensure the healthy operation of the unit. The reliability assessment at the physical level above the component level depends on the failure rate of key components. Large units in the industrial field usually face harsh operating environments. For example, nuclear power units. The components of these equipment operate under high temperature, high pressure and strong radiation conditions, and are subjected to high speed, heavy load and long-term cyclic working pressure. The degradation phenomena such as wear, fatigue and corrosion caused by long-term operation, progressive failure has no obvious impact on the overall performance of the unit in the early stage, but as time goes by, the gradual accumulation of degradation may cause serious failures. Traditional reliability analysis methods have certain limitations in dealing with the problem of progressive failure of nuclear power equipment components. For example, a simple physical mechanism model is difficult to cope with the uncertainty under complex working conditions, and the data-driven model has deficiencies in small samples and model interpretability. Therefore, it is particularly urgent to develop a method that can integrate multiple information and effectively predict the progressive failure events of nuclear power equipment, so as to timely maintain or replace potential risk components and ensure the safe operation of nuclear power equipment. Summary of the invention
[0005] Aiming at the deficiencies of the existing technology, the technical problem solved by the present invention is to provide a method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model, which solves the problem that the existing simple physical mechanism model is difficult to cope with the uncertainties under complex working conditions, while the data-driven model has deficiencies in small samples and model interpretability, resulting in the inability to meet the prediction of progressive failure events of nuclear power equipment.
[0006] To solve the above problems, the technical solution adopted by the present invention is: a method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model, including the following steps; S100: Construct a state model of component progressive failure based on a non-linear Wiener process; S200: Construct an observation model of component progressive failure based on physical mechanism according to the Archard model and crack propagation theory; S300: Use the observation model of progressive failure based on physical mechanism as the observation equation of the Kalman filter, and the state model of component progressive failure based on the non-linear Wiener process as the state equation of the Kalman filter to form a complete prediction model structure; S400: Initialize the Kalman filter to set the initial state estimate and covariance estimate; and carry out real-time estimation of the degradation state by integrating mechanism and data perception based on the unscented Kalman filter; S500: Improve the real-time estimation of the degradation state of the UKF in step S400 based on the parameter adjustment strategy of the observation innovation, so that the data-driven state model can evolve adaptively online; S600: Use the updated state model and the preset failure threshold to measure the reliability of the component, and judge whether the component is progressively failing.
[0007] The technical principle and beneficial effects of this solution are as follows: In this solution, the non-linear drift Wiener process is used as the degradation state equation, and the physical mechanism model is used as the observation equation in the state space model to carry out degradation estimation driven by both mechanism and data. The physical mechanism model constructs the mathematical relationship between the observation vector and the degradation state, making up for the limitation that the degradation state is unobservable. The time-varying non-linear characteristics of the degradation process are described by the non-linear drift Wiener process, making up for the insufficient adaptability of the traditional linear model to complex degradation trajectories. And based on the unscented Kalman filter algorithm for state prediction and update, and then through the feedback fusion mechanism to obtain a comprehensive degradation state evaluation. Finally, according to the relevant reliability theory, the progressive failure events of nuclear power equipment components are calculated. The present invention effectively improves the accuracy and reliability of prediction, provides an important basis for the maintenance decision-making of nuclear power equipment, solves the pain points such as model mismatch, noise sensitivity, and inaccurate long-term prediction in traditional methods, and effectively ensures the safe operation of nuclear power and reduces the operation and maintenance costs.
[0008] Furthermore, the specific steps for constructing the state model in step S100 are as follows; S101: The state model of the Wiener process with non-linear drift is expressed as follows:
[0009] In the formula, is the initial state, is the drift term, μ is the drift coefficient, reflecting the average degradation rate, is the average degradation path, which is in the form of a power function of time t; is the diffusion term, σ is the diffusion coefficient, and B(t) is a standard Brownian motion, whose mathematical definition is: , , ; S102: Fit the non-linear Wiener process according to the historical data samples, optimize the parameters with the goal of minimizing the fitting mean square error, and determine the exponential coefficient , drift coefficient μ and diffusion coefficient σ in the non-linear Wiener process; S103: Convert the non-linear Wiener process in continuous time into a discrete-time state equation; the degradation process is monitored in chronological order t 1 <t 2, and the discrete-time state variable is denoted as x k =x (t k ), and the observed variable is denoted as y k =y (t k ). At time t k , the goal is to estimate the current degradation state value x k-1 based on the previous state estimate value x k , and so on; The discrete-time state equation is expressed as: , where υ k is related to the diffusion term, .
[0010] Furthermore, the specific steps for constructing the observation model in step S200 are as follows; S201: For wear progressive failure, use the Archard model to derive the relationship expression between the wear volume and physical parameters such as the coating particle radius, initial crack length, wear distance, and load, and take the wear distance as the observed variable; the relationship expression of the physical mechanism model of wear progressive failure is:
[0011] Wherein, K' is the corrected wear coefficient, r is the average radius of the cemented carbide phase on the surface of the positioning mechanism, d0 is the initial crack length, ζ1 is the multiple relationship of the load, F is the load, and L is the wear distance in 60 years; S202: For impact fatigue progressive failure, based on the Paris formula and crack propagation theory, combined with the stress conditions of the component, the relationship expression between the crack length and physical parameters such as the number of impact cycles and impact load is obtained, and the number of impact cycles is used as the observed variable; the relationship expression of the physical mechanism model of impact fatigue progressive failure is;
[0012] Wherein, a is the crack length, N is the number of stress cycles, represents the crack propagation rate, C and m are constants related to the coating material, the integral constant C1 is determined by the initial conditions, and ζ2 is the multiple relationship of the impact load on the mechanism.
[0013] Furthermore, the physical mechanism model of wear progressive failure constructed in step S201 and the physical mechanism model of impact fatigue progressive failure constructed in step 202 are respectively used as the observation equations of parallel sub-Kalman filters. The synchronous observation of different failure parameters of nuclear power equipment is realized, and the multi-dimensional prediction of the progressive failure of nuclear power equipment is realized.
[0014] Furthermore, the specific steps of the said step S400 are as follows; Step S401: The unscented Kalman filter performs state prediction according to the prediction model structure established in step S300; Step S402: Update the state model according to the corresponding observation equation and the observed value obtained from actual measurement, and calculate key parameters such as the cross-covariance matrix and Kalman gain.
[0015] Furthermore, the said step S500 updates the state model parameters through the following steps, Step S501: Use the observation innovation to adaptively adjust the state model parameters and add the new observed value to the model evolution process; assume that the state model parameter vector at the current moment is , multiply the set update step size k by the observation innovation , and then add it to to update and obtain the state model parameters at the next moment; the formula is as follows;
[0016] Among them, refers to the parameter exponential coefficient , drift coefficient μ and diffusion coefficient σ in the state model; Step S502: Set the innovation threshold, and make adjustments and evolutions after the innovation at a certain moment reaches the threshold.
[0017] After reaching the set threshold, update and adjust the parameters of the state model to avoid unstable model predictions caused by frequent model parameter updates. At the same time, the state equation can be updated regularly at a certain frequency to save computing resources.
[0018] Furthermore, step S100 of the method further includes step S104; Step S104: Based on the characteristic that the first passage time of the Wiener process follows an inverse Gaussian distribution, establish the following reliability metric formula for the failure mode:
[0019] In step S600, the reliability metric formula in step S104 is used to calculate the reliability of the component; and when it is determined that the reliability is lower than the set safety threshold, it is determined that a progressive failure event of the component has occurred, and a warning signal is sent to the operation and maintenance terminal.
[0020] Through the constructed reliability metric formula, accurate calculation of the component reliability based on the failure threshold is realized, so as to accurately judge the reliability of the equipment, and further judge whether a progressive failure event of the component has occurred; thus, operation and maintenance technicians can take corresponding technical measures in a timely manner according to the warning information received by the operation and maintenance terminal, such as replacing components or adjusting operating parameters, so as to ensure the safe and stable operation of nuclear power equipment. Brief Description of the Drawings
[0021] Figure 1 It is a schematic diagram of the step flow of a method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model.
[0022] Figure 2 It is a structural block diagram of a data-driven model and a physical mechanism model fused in two stages of UKF state prediction and observation update. Detailed Description of the Specific Embodiment
[0023] The following is a more detailed description through specific embodiments: Basically as shown in the appendix Figure 1 、 Figure 2 shown: A method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model includes the following steps; S100: Construct a state model for the progressive failure of components based on a non-linear Wiener process; the specific steps for constructing the state model are as follows; S101: The state model of the Wiener process with non-linear drift is expressed as follows:
[0024] In the formula, is the initial state, is the drift term, μ is the drift coefficient, reflecting the average degradation rate, is the average degradation path, which is in the form of a power function of time t; is the diffusion term, σ is the diffusion coefficient, and B(t) is a standard Brownian motion, whose mathematical definition is: , , ; S102: Fit the nonlinear Wiener process according to the historical data samples, optimize the parameters with the goal of minimizing the fitting mean square error, and determine the exponential coefficient , drift coefficient μ, and diffusion coefficient σ in the nonlinear Wiener process; the objective optimization function is expressed as follows:
[0025] S103: Convert the nonlinear Wiener process in continuous time into a discrete-time state equation; the degradation process is monitored in chronological order t 1 <t 2 ...<t n The discrete-time state variable is denoted as x k =x (t k ), and the observed variable is denoted as y k =y (t k ). At time t k , the goal is to estimate the current degradation state value x k-1 based on the previous state estimate value x k , and so on; The discrete-time state equation is expressed as: , where υ k is related to the diffusion term, ; S104; Based on the characteristic that the first passage time of the Wiener process follows an inverse Gaussian distribution, establish the reliability metric formula for the failure mode as follows:
[0026] S200: Construct an observation model for the progressive failure of components based on physical mechanisms according to the Archard model and crack propagation theory; the specific steps for constructing the observation model are as follows; S201: For progressive wear failure, the relationship expression between the wear volume and physical parameters such as the coating particle radius, initial crack length, wear distance, and load is derived using the Archard model, and the wear distance is taken as the observation variable; the relationship expression of the physical mechanism model of progressive wear failure is:
[0027] In the formula, K' is the modified wear coefficient, r is the average radius of the cemented carbide phase on the surface of the positioning mechanism, d0 is the initial crack length, ζ1 is the multiple relationship of the load, F is the load, and L is the wear distance in 60 years; S202: For impact fatigue progressive failure, based on the Paris formula and crack propagation theory, combined with the stress conditions of the component, the relationship expression between the crack length and physical parameters such as the impact cycle number and impact load is obtained, and the impact cycle number is taken as the observation variable; the relationship expression of the physical mechanism model of impact fatigue progressive failure is;
[0028] In the formula, a is the crack length, N is the stress cycle number, represents the crack propagation rate, C and m are constants related to the coating material, the integral constant C1 is determined by the initial conditions, and ζ2 is the multiple relationship of the impact load on the mechanism.
[0029] S300: The physical mechanism model of progressive wear failure constructed in step S201 and the physical mechanism model of impact fatigue progressive failure constructed in step 202 are respectively used as the observation equations of parallel sub-Kalman filters, and the component progressive failure state model based on the nonlinear Wiener process is used as the state equation of the Kalman filter to form a complete prediction model structure; The complete prediction model structure of this solution is as Figure 2 shown. In the state space of the prediction model, the upper circle and the lower circle represent the state variable and the observation variable respectively, where the upper dotted circle and solid circle represent the prior estimate value and the posterior estimate value respectively, the lower dotted circle and solid circle represent the estimated observation value and the true observation value respectively, and the black dotted arrow represents the information transfer process in a single time step of state estimation. The data-driven nonlinear Wiener process is used as the state transition equation in the UKF to provide state prior prediction. The physical mechanism model is used as the observation equation in the UKF update stage.
[0030] S400: Initialize the Kalman filter to set the initial state estimate and covariance estimate; and perform real-time estimation of the degradation state of mechanism and data perception fusion based on the unscented Kalman filter; S401: The unscented Kalman filter performs state prediction according to the prediction structure established in step S300; S402: Update the state of the state model according to the corresponding observation equation and the observed values obtained from the actual measurement, and calculate key parameters such as the cross-covariance matrix and the Kalman gain.
[0031] In the prediction stage, the Kalman filter performs state prediction based on its own state equation. Considering the non-linear characteristics of the model, the unscented Kalman filter (UKF) technology or its improved methods are used to handle it. In the update stage, the state is updated according to the corresponding observation equation and the observed values obtained from the actual measurement, and key parameters such as the cross-covariance matrix and the Kalman gain are calculated, thereby improving the accuracy of state estimation.
[0032] S500: Improve the real-time estimation of the degraded state of UKF in step S400 based on the parameter adjustment strategy of the observation innovation, so that the data-driven state model can evolve online adaptively; S501: Adaptively adjust the state model parameters using the observation innovation, and add the new observed values to the model evolution process; Let the state model parameter vector at the current moment be , multiply the set update step size k by the observation innovation , and then add it to to update and obtain the state model parameters at the next moment; The formula is as follows;
[0033] Among them, refers to the parameter exponential coefficient , the drift coefficient μ, and the diffusion coefficient σ in the state model; S502: Set the innovation threshold, and perform adjustment and evolution after the innovation at a certain moment reaches the threshold.
[0034] S600: Use the updated state model and the preset failure threshold to measure the reliability of the component through the measurement formula in step S104, and when it is judged that the reliability is lower than the set safety threshold, determine that the component has an incipient failure event and send a warning signal to the operation and maintenance terminal.
[0035] Operation and maintenance technicians can take corresponding technical measures in a timely manner according to the warning information received by the operation and maintenance terminal, such as replacing components or adjusting operating parameters, so as to ensure the safe and stable operation of nuclear power equipment.
[0036] The above are only embodiments of the present invention, and common general knowledge such as specific structures and characteristics known in the art are not described in detail herein. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several modifications and improvements can be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope claimed in this application shall be subject to the content of its claims, and the specific implementation manners and the like recorded in the specification can be used to interpret the content of the claims.
Claims
1. A method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model, characterized in that: The steps include: S100: Construct a state model for component progressive failure based on nonlinear Wiener process; S200: Based on the Archard model and crack propagation theory, an observation model for progressive failure of components based on physical mechanisms is constructed; S300: The observation model of progressive failure based on physical mechanism is used as the observation equation of Kalman filter, and the component progressive failure state model based on nonlinear Wiener process is used as the state equation of Kalman filter to form a complete prediction model structure; S400: Initialize the Kalman filter to set the initial state estimation and covariance estimation; and perform real-time estimation of degradation state based on the fusion of mechanism and data perception based on the unscented Kalman filter; S5 00: Improve the real-time estimation of the degradation state of UKF in step S400 based on the parameter adjustment strategy of the observed new information, so that the data-driven state model can be adaptively evolved online; S600: Measure the reliability of the component using the updated state model and a preset failure threshold to determine whether the component is progressively failing.
2. According to claim 1, a method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model is characterized in that: The specific steps of constructing the state model in step S100 are as follows; S101: The state model of the Wiener process with nonlinear drift is expressed as follows: In the formula, is the initial state, is the drift term, μ is the drift coefficient, reflecting the average degradation speed, is the average degradation path, which is a power function of time t; is the diffusion term, σ is the diffusion coefficient, B(t) is the standard Brownian motion, and is mathematically defined as: , , ; S102: Fit the nonlinear Wiener process based on historical data samples, optimize the parameters with the goal of minimizing the mean square error of the fitting, and determine the exponential coefficient in the nonlinear Wiener process , drift coefficient μ and diffusion coefficient σ; S103: Convert the continuous-time nonlinear Wiener process into a discrete-time state equation; the degradation process is carried out in time order t 1 <t 2 monitoring, the discrete time state variable is recorded as x k =x (t k ), the observed variable is recorded as y k =y (t k ), at t k At this moment, the goal is to estimate the value x based on the previous state k-1 Estimate the current degradation state value x k , and so on; The discrete-time state equation is expressed as: , where υ k Related to the diffusion term, .
3. The method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model according to claim 1, characterized in that: The specific steps of constructing the observation model in step S200 are as follows: S201: For wear progressive failure, the Archard model is used to derive the relationship between the wear volume and physical parameters such as coating particle radius, initial crack length, wear distance and load, and the wear distance is used as the observed variable; the relationship expression of the physical mechanism model of wear progressive failure is obtained as follows: Where K' is the corrected wear coefficient, r is the average radius of the cemented carbide phase on the surface of the positioning mechanism, d0 is the initial crack length, ζ1 is the multiple relationship of the load, F is the load, and L is the 60-year wear distance; S202: For impact fatigue progressive failure, based on the Paris formula and crack propagation theory, combined with the stress conditions of the components, the relationship expression between crack length and physical parameters such as impact cycle number and impact load is obtained, and the impact cycle number is used as the observed variable; the relationship expression of the physical mechanism model of impact fatigue progressive failure is obtained as follows; Where a is the crack length, N is the number of stress cycles, indicating the crack growth rate, C and m are constants related to the coating material, the integral constant C1 is determined by the initial conditions, and ζ2 is the multiple of the impact load on the mechanism.
4. The method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model according to claim 3 is characterized in that: The physical mechanism model of progressive failure due to wear constructed in step S201 and the physical mechanism model of progressive failure due to impact fatigue constructed in step 202 are respectively used as observation equations of parallel sub-Kalman filters.
5. A method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model according to claim 4, characterized in that: The specific steps of step S400 are as follows: Step S401: The unscented Kalman filter performs state prediction according to the prediction model structure established in step S300; Step S402: Update the state of the state model according to the corresponding observation equation and the observation value obtained by actual measurement, and calculate the cross covariance matrix, Kalman gain and observation information according to the predicted value.
6. The method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model according to claim 1, characterized in that: The step S500 updates the state model parameters through the following steps: Step S501: Adaptively adjust the state model parameters using the new observation information and add the new observation value to the model evolution process; let the state model parameter vector at the current moment be , set the update step size k and observe the new information Multiply and then Add and update to get the state model parameters of the next moment ;The formula is as follows; in, Refers to the parameter exponential coefficient in the state model , drift coefficient μ and diffusion coefficient σ; Step S502: Set a new information threshold, and adjust and evolve after the new information at a certain moment reaches the threshold.
7. A method for predicting progressive failure events of nuclear power equipment based on a parallel feedback Kalman filter model according to claim 2, characterized in that: The step S100 also includes step S104; Step S104: Based on the characteristic that the first arrival time of the Wiener process obeys the inverse Gaussian distribution, the reliability measurement formula of the failure mode is established as follows: In the step S600, the reliability of the component is calculated using the reliability metric formula of step S104; and when it is determined that the reliability is lower than the set safety threshold, it is determined that a progressive failure event has occurred in the component, and an early warning signal is sent to the operation and maintenance terminal.