Variable-working-condition multi-stage deep-sea pressure-resistant cabin fatigue crack propagation prediction method

Abstract

The invention discloses a variable-working-condition multi-stage deep-sea pressure-resistant cabin fatigue crack propagation prediction method, and relates to the technical field of deep-sea pressure-resistant cabin fatigue detection. A complete machine finite element model of a deep-sea pressure-resistant cabin is established, weak links are analyzed, and fine modeling is performed; establishing a stress intensity factor-Gaussian process agent model; and based on the dynamic Bayesian best-worst network model, predicting the fatigue crack of the deep sea pressure-resistant cabin. Performing sensitivity analysis on uncertainty parameters in the dynamic Bayesian network model based on a fatigue life prediction model of a small time scale; according to a parameter reasoning algorithm based on hybrid adaptive resampling intelligent particle filtering, the influence of uncertain parameters in the crack expansion process is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fatigue detection of deep - sea pressure - resistant cabins, and specifically to a method for predicting fatigue crack propagation of deep - sea pressure - resistant cabins with variable working conditions and multiple stages. Background Technique

[0002] Deep - sea pressure - resistant cabins are key equipment to ensure the safety of deep - sea operation personnel. During the diving process, the load on them will increase, while during the floating process, the load will decrease. Therefore, they will bear complex alternating loads. These loads will not only cause significant stress concentration but also may trigger creep effects. In this extreme environment, the microstructure inside the material may change. Especially at the welded joints of the pressure - resistant cabin, cracks are likely to initiate and continue to expand. Therefore, determining the crack position and its propagation trend of deep - sea pressure - resistant cabins has great economic and safety significance.

[0003] CN118536404A, in order to predict the life of key components such as aero - engines that bear complex alternating loads, presents a method for predicting fatigue crack propagation under variable - amplitude loads based on dynamic Bayesian update. It combines the Huang model and the dynamic Bayesian network model for crack propagation prediction, uses the Kalman filtering theory formula to calculate the theoretical optimal estimate by combining observation information and model information, and then corrects the model according to the optimal estimate for crack prediction, improving the calculation accuracy. However, this method does not involve the training of the prediction model, making it inapplicable to fatigue crack propagation prediction under other models.

[0004] CN114282709B, in order to accurately know the current crack state at the key structural positions of complex equipment and analyze and predict its propagation trend, proposes a method for predicting fatigue crack propagation of structural components based on digital twin. It establishes and trains a Gaussian process surrogate model, combines the dynamic Bayesian network model and the particle filter algorithm, and can realize crack prediction of complex equipment and reduce the influence of uncertain factors during the crack propagation process. However, this method does not consider the influence of changing uncertain factors and load characteristics in the working condition environment of deep - sea pressure - resistant cabins, and is very limited in its application in this field.

[0005] (1) Poor prediction accuracy. There are influencing factors such as material properties, load conditions, and initial cracks in the fatigue crack prediction process. Especially under the working characteristics of deep-sea pressure cabins, during the loading and overloading stages, the size of the plastic zone and the change of material yield strength need to be mainly considered, while during the bearing stage, its creep characteristics need to be noted. As time accumulates, considering only a single situation will lead to the prediction result deviating from the actual. The dynamic Bayesian best-worst method adopted in this patent is applicable to dealing with uncertainty problems in complex systems, analyzing the uncertain influencing factors at each stage and assigning different weights, obtaining the dominant influencing factors at different stages after prediction and comparison, and using the data-model fusion method to implement model correction in different working stages on the premise that the crack detection or monitoring data is assumed to be correct, so as to improve the prediction accuracy.

[0006] (2) Low simulation efficiency. At present, the training data has high redundancy and poor representativeness, resulting in a reduction in the training accuracy of the model. For the training of the model, this patent proposes a method of nearest neighbor sampling based on unilateral limitation, which can generate representative sample data and can quickly and accurately complete the training of the fatigue crack prediction model; and divides the particle generation range according to the weights of uncertain factors, which can effectively improve the efficiency and accuracy of particle filtering.

[0007] In summary, regarding the problem of fatigue crack growth prediction around deep-sea pressure cabins, although the improved Paris formula can consider the case of variable amplitude loads, for the research object of pressure cabins, the load change amplitude is relatively slow, making the load characteristics inconsistent with the working conditions assumed by the pairs formula, and coupled with the influence of creep, the traditional method is not applicable. Therefore, a fatigue crack growth prediction technology at a small time scale is proposed. By considering the crack growth process in any small interval, it can effectively cope with the problems of short-term overload and a certain depth of load maintenance in the load cycle of the pressure cabin lasting for several hours. However, when this model is used, it does not consider the prediction deviation affected by multiple uncertain factors such as materials, loads, and models during the service process, resulting in many differences between the crack prediction result and the actual situation, thus affecting the safety and reliability assessment of the pressure cabin. Summary of the Invention

[0008] At present, most of the predictions of fatigue cracks are based on the number of load cycles. However, under the working conditions of a deep-sea pressure-resistant cabin, due to the complex load and short time period, the traditional prediction method will be computationally complex and the accuracy will be reduced. The present invention takes a small time scale as the unit, integrates various loading conditions into a small time surrogate model, continuously trains the surrogate model with data, and then corrects the parameters of the model after comparing with the actual crack condition, so as to improve the prediction accuracy of the model; on the other hand, instead of calculating the crack propagation condition based on the number of load cycles, it calculates based on time, so as to efficiently and real-time master the crack propagation of the deep-sea pressure-resistant cabin, and then judge whether its working condition is safe.

[0009] A method for predicting fatigue crack propagation of a deep-sea pressure-resistant cabin with variable working conditions and multiple stages according to the present invention includes: S1. Establish a finite element model of the whole pressure-resistant cabin by using the shell hypothesis, import the operating conditions to draw up boundary conditions, analyze the stress distribution characteristics and perform fine modeling on the weak links S2. Construct a sample for the stress intensity factor - Gaussian process surrogate model, bring the generated sample into the finite element model, calculate the stress intensity factor at the crack tip, and establish a stress intensity factor sample database; S3. Based on the Gaussian process theory, establish a stress intensity factor - Gaussian process surrogate model according to the stress intensity factor sample database; S4. Based on the dynamic Bayesian best-worst network model, predict the fatigue cracks of the deep-sea pressure-resistant cabin; S5. Based on the fatigue life calculation model on a small time scale, perform sensitivity analysis on the uncertain parameters in the dynamic Bayesian best-worst network model; S6. Based on the parameter inference algorithm of the hybrid adaptive resampling intelligent particle filter, reduce the influence of uncertain parameters in the crack propagation process.

[0010] In the preferred embodiment, in step S2, the spatial format of the sample defining the stress intensity factor - Gaussian process surrogate model is: ; Among them, is the load magnitude in the loading and overloading stages, is the load magnitude in the hold load stage, , are the creep constant and the secondary creep stage exponent respectively, is the stress-strain state constraint factor, is the stress intensity factor threshold value, is the crack propagation value in the loading and overloading stages, is the crack propagation value in the hold load stage.

[0011] In a preferred embodiment, a method of generating a sample database using nearest neighbor sampling based on unilateral restriction is adopted. Based on simulating or predicting cracks, a sample generation space with a distance d is taken above and below it, and it is expanded outward in sequence with as the unit. The distance between each data point selected each time and the sample is a multiple of . There are N groups of data above and below, for a total of 2N groups of data.

[0012] In a preferred embodiment, in step S3, a squared exponential covariance function is selected as the covariance kernel function of the stress intensity factor - Gaussian process surrogate model : ; where is the kernel function signal variance that controls the local correlation of the input variables; is the length scale parameter, x i is the i-th input data, and x j is the j-th input data; Use the selected sample data to train the stress intensity factor - Gaussian process surrogate model, estimate the hyperparameters of the covariance kernel function, and optimize and adjust the hyperparameters of the covariance kernel function through maximum likelihood estimation to minimize the prediction error; use the trained stress intensity factor - Gaussian process surrogate model as the surrogate model to predict the stress intensity factor according to the input crack length and load conditions.

[0013] In a preferred embodiment, step S4 includes: S41. Determine the weights of each influencing factor through BWM, manually score the influencing factors according to prior experience, and finally determine the weights; S42. Construct a DBN model according to the influencing factors and their weights. Each node represents a risk factor, and the arrow represents a causal relationship; S43. Dynamically update the weight distribution in the DBN model through monitoring data; S44. When there is a deviation in the prediction data of the dynamic Bayesian best - worst network model, quickly judge the working state of the pressure - resistant cabin through the backward inference of the DBN model, so as to adjust the parameters of the dynamic Bayesian best - worst network model to adapt to the crack propagation characteristics in the current stage.

[0014] In a preferred embodiment, in step S5, the fatigue life calculation model on a small time scale is: ; ; ; Among them, is the crack increment in the holding stage during a loading cycle, is the crack increment in the loading stage during a loading cycle, is the initial crack length, , , are respectively functions of the secondary creep exponent , the creep constant, and the second creep stage exponent, θ maxo is the maximum crack opening angle, is the stress-strain state constraint factor, is the elastic modulus of the material, is the yield strength of the material, is the maximum stress intensity factor, is the stress intensity factor threshold value; , , , the stress-strain state constraint factor and the stress intensity factor threshold value are regarded as uncertain nodes, and other variables are regarded as deterministic fixed nodes.

[0015] In a preferred embodiment, after setting an initial crack length in the dynamic Bayesian best-worst network model, the actually detected load data is input into the dynamic Bayesian best-worst network model, so as to calculate the crack propagation situation of the structure on the basis of time series; When the observed value of the crack is detected at time t-1, then the crack length at the next moment is the optimal estimated value calculated by combining the observed crack length and the prior value of the crack length; if the observed value of the crack is not detected at time t-1, then the crack length .

[0016] In a preferred embodiment, after completing the state estimation of the particles, according to the small time scale prediction model, the state equation of the fatigue crack propagation process is: ; Among them is the crack at time t-1, and The parameters are determined by resampling. Through the Gaussian process surrogate model, the stress intensity factor at any time can be accurately predicted, which is a key parameter for crack propagation analysis. Finally, the predicted crack at the moment can be output.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: A method for generating sample data based on nearest neighbor sampling with unilateral restriction is proposed, which can generate high-quality sample data, avoid redundancy, cover key areas, improve the efficiency of model training and prediction accuracy, provide a good foundation for model optimization, and enable the model to predict crack propagation more stably and accurately in complex environments, significantly improving the reliability and efficiency of crack prediction for deep-sea pressure cabins; a dynamic Bayesian best-worst network model with variable working conditions and multiple stages is constructed. By analyzing the influence degrees of various uncertain factors under different working conditions, the most accurate prediction method for each stage is determined, which is more in line with the actual working conditions of deep-sea pressure cabins compared with traditional prediction methods; a hybrid adaptive resampling intelligent particle filter is used as the inference algorithm of the dynamic Bayesian best-worst. According to the real-time situation of the system state and particle distribution, by adaptively adjusting the resampling strategy, it can better cope with the problem of particle degradation, has strong robustness, can improve the accuracy of state estimation, and is not prone to significant performance degradation.

[0018] The dominant influencing factors in different stages of deep-sea pressure cabins are different. By using the dynamic Bayesian best-worst network model to find out the dominant influencing factors in each stage, the error range can be adjusted according to the weights in the subsequent particle filter. The particle generation range of the dominant factors is appropriately increased, and the particle generation range of the secondary factors is appropriately reduced to improve efficiency and increase accuracy, so as to determine the prediction criteria that best fit the fatigue cracks in different time periods.

[0019] By using the hybrid adaptive resampling intelligent particle filter to monitor indicators such as the degree of particle degradation, the effective sample size, and the diversity of particles in real time, the resampling process is automatically optimized, and the parameters and method combinations of resampling are dynamically changed. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is the generation method based on nearest neighbor sampling with unilateral restriction of the present invention; Figure 2 is the crack calculation flow chart of the stress intensity factor - Gaussian process surrogate model; Figure 3 is the flow chart of the hybrid adaptive resampling intelligent particle filter algorithm; DETAILED DESCRIPTION OF THE INVENTION

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] Deep-sea pressure chambers are important equipment for developing marine resources and conducting underwater operations. Fatigue cracks are prone to initiation at the connections of their hulls, which is very detrimental to the safety of personnel and equipment. In order to solve this problem, this patent proposes a variable working condition multi-stage fatigue crack prediction method for pressure chambers.

[0023] In order to obtain the key parameter of stress intensity factor in fatigue crack prediction, while improving the calculation efficiency and reducing the calculation cost, the present invention establishes a proxy model based on the stress intensity factor of nearest neighbor sampling-Gaussian process, realizes the calculation of the stress intensity factor at the crack tip, distinguishes the sample space format according to the time points of transition between different loading stages, adopts the nearest neighbor sampling method based on unilateral restriction to generate the sample data space, and uses typical representative data to train the proxy model to improve its accuracy and credibility.

[0024] The specific steps are: S1. Use the shell assumption to establish the finite element model of the pressure cabin, import the operating conditions to formulate boundary conditions, analyze stress and deformation to determine the weak links. Use solid unit modeling for the weak links, divide the fine grid and perform unit coupling, combine Solidworks software to establish an accurate geometric model, and then import it into Abaqus for meshing.

[0025] S2. Construct samples for the stress intensity factor-Gaussian process proxy model, bring the generated samples into the finite element model, calculate the stress intensity factor at the crack tip, and establish a stress intensity factor sample database.

[0026] According to the working environment characteristics of the deep-sea pressure chamber, the crack propagation characteristics remain unchanged when it is in the respective loading, load-keeping and overloading stages. In order to better construct and apply the Gaussian process proxy model to efficiently predict the distribution and changes of the stress intensity factor (SIF), the spatial format of the sample of the stress intensity factor-Gaussian process proxy model is defined as: ; in, is the load magnitude during the loading and overloading phases, is the load size in the load-holding stage, , are creep constant, second creep stage index, is the stress-strain state constraint factor, is the stress intensity factor threshold value, is the crack extension value in the loading and overloading stages, is the crack extension value during the load-keeping stage.

[0027] Bring the generated samples into the finite element model, calculate the stress intensity factor at the crack tip, and establish a stress intensity factor sample database.

[0028] In this step, in order to obtain sample data that conforms to the actual crack propagation and improve the credibility of the surrogate model, the method shown in Figure 1 is proposed to generate the sample database.

[0029] Select reference points: First, select multiple reference points from the existing crack paths. These reference points can be specific positions on the crack path, such as the starting point, inflection point, or termination point of the crack.

[0030] Determine the center and radius of the circle: Take each reference point as the center of the circle and draw a circle with a predetermined length above or below it as the radius. This length can be set according to actual needs and the expected pattern of crack propagation.

[0031] Generate new sample data: Select a point uniformly or randomly on each circle as a new sample data point. These points represent the possible crack propagation positions near the current reference point.

[0032] Connect the new sample data: Connect these newly generated sample data points from the reference point to the new points to form a new crack sample path. These paths can be straight lines, curves, or other shapes, depending on the physical mechanism and expected pattern of crack propagation.

[0033] Expand the sample data: Take each reference point as the center of the circle in turn and continue to draw circles by expanding a set length outward to obtain more new sample data. This process can be repeated multiple times to generate enough sample data for training and validating the crack prediction model.

[0034] Form a crack sample set: Through the above steps, we can generate a set of crack samples that contain the crack propagation conditions at different positions and directions. These samples can be used to construct and train a Gaussian process surrogate model to predict the crack propagation path and behavior.

[0035] To improve the accuracy of the surrogate model, the method of nearest neighbor sampling based on unilateral restriction is used to generate the sample database. Based on simulating or predicting cracks, the method of nearest neighbor sampling based on unilateral restriction is used to generate the sample database. Taking d as the distance above and below it as the sample generation space, as the unit, expand outward in turn. The distance between each selected data point and the sample is a multiple of . There are N groups of data above and below, a total of 2N groups of data. Among them, d can be taken as 1 - 2 mm, can be taken as 0.01 - 0.02 mm, and N can be taken as 50 - 60.

[0036] In a preferred embodiment, 1 mm is taken as the sample generation space above and below it, and it is expanded outward in turn in units of 0.02 mm. There are 50 groups of data above and below, for a total of 100 groups of data.

[0037] S3. Based on the Gaussian process theory, establish a stress intensity factor - Gaussian process surrogate model according to the stress intensity factor sample database. Specifically, it includes: S31. Select the squared exponential covariance function as the covariance kernel function of the stress intensity factor - Gaussian process surrogate model : ; Among them, is the kernel function signal variance that controls the local correlation of the input variables; is the length scale parameter, x i is the i-th input data, x j is the j-th input data.

[0038] S32. Use the selected sample data to train the stress intensity factor - Gaussian process surrogate model (GP model), estimate the hyperparameters of the covariance kernel function, and optimize and adjust the hyperparameters of the covariance kernel function through maximum likelihood estimation to minimize the prediction error.

[0039] S33. Use the trained GP model as the surrogate model, and this model can predict the stress intensity factor according to the input crack length and load conditions.

[0040] Traditional finite element analysis requires repeated calculations for each crack state when predicting crack propagation, which is very time-consuming computationally. While the SIF-GP surrogate model can quickly predict the SIF value in a new crack state through machine learning methods, significantly improving the computational efficiency. The deep-sea pressure hull will be affected by various uncertain factors in the actual working environment. The SIF-GP surrogate model can integrate these uncertainty factors, provide more accurate crack propagation predictions, and dynamically update and predict the crack propagation state according to real-time monitoring data, which is crucial for structural health management and maintenance decisions.

[0041] S4. Based on the dynamic Bayesian network model, predict the fatigue cracks of the deep-sea pressure hull.

[0042] The fatigue life assessment of the deep-sea pressure hull needs to reasonably consider its load characteristics, including short-term overloads during load cycles lasting several hours and load maintenance at a certain depth. If the influencing factors are not analyzed specifically for the characteristics of each stage, over time, the predicted results of crack propagation will gradually deviate significantly from the actual crack situation. Therefore, as Figure 2A dynamic Bayesian best-worst network model is proposed as shown below. Based on the traditional Bayesian method, it can compare influencing factors and achieve accurate prediction of fatigue cracks in deep-sea pressure cabins. The key steps for establishing this dynamic Bayesian best-worst network model are as follows: S41. Determine the weights of each influencing factor through BWM. Manually score the influencing factors according to prior experience and finally determine the weights.

[0043] In crack prediction, it is first necessary to determine the weights of each influencing factor through BWM to provide initial parameter estimates for the model. These parameters will affect the prediction accuracy of the model. Manually score these uncertain parameters according to prior experience. For example, in the loading and overloading stages, the stress-strain state constraint factor is generally regarded as the main influencing factor, while in the holding stage, the creep constant is regarded as the main influencing factor. Combine historical data analysis to determine the relative importance of each factor.

[0044] S42. Construct a dynamic Bayesian network model (DBN) according to the influencing factors and their weights. Each node represents a risk factor, and the arrow represents a causal relationship.

[0045] Construct a DBN model according to the influencing factors and their weights. Each node represents a risk factor, and the arrow represents a causal relationship. Create a model framework that can represent the interaction of various factors during crack propagation. The arrow indicates how these factors affect the formation and propagation of cracks.

[0046] S43. Dynamically update the weight distribution in the dynamic Bayesian network model through monitoring data.

[0047] In practical applications, dynamically update the weight distribution in the DBN through monitoring data. When the load changes greatly, it indicates that it is in the loading or overloading stage. When the load change is not obvious, it indicates that it is in the holding stage. The effect of this step is to enable the model to adapt to the changing working environment and improve the prediction accuracy and reliability. For example, if it is monitored that the load suddenly increases, the model can adjust the weights to reflect the impact of this change on crack propagation.

[0048] S44. When there is a deviation in the prediction data of the dynamic Bayesian network model, quickly judge the working state of the pressure cabin through the reverse inference of the DBN, and then adjust the parameters of the dynamic Bayesian network model to adapt to the crack propagation characteristics of the current stage.

[0049] When there is a deviation in the predicted data of the dynamic Bayesian network model, the working state of the pressure-resistant cabin can be quickly judged through the reverse reasoning of the DBN, so as to adjust the parameters of the dynamic Bayesian network model to adapt to the crack growth characteristics in the current stage. The effect of this step is to improve the adaptability and accuracy of the model and ensure that the model can accurately reflect the actual crack growth process. For example, if the crack growth rate predicted by the model does not match the actual observed data, the model can be optimized by adjusting the weights of relevant nodes to better conform to the actual situation.

[0050] S5. Conduct a sensitivity analysis of the uncertain parameters in the dynamic Bayesian network model based on the fatigue life calculation model on a small time scale.

[0051] In order to obtain the uncertain parameters in the dynamic Bayesian network model, a sensitivity analysis of the uncertain parameters is carried out on its fatigue life calculation model based on a small time scale. This module can not only accurately calculate the crack growth under long-term cyclic loads, considering the overloading and sustained load effects, but also comprehensively consider the plastic zone and creep effects, and can comprehensively evaluate the crack propagation behavior of the deep-sea pressure-resistant cabin.

[0052] Fatigue life calculation model based on a small time scale: ; ; ; where is the crack increment during the load-holding stage in a loading cycle, t is the time increment at any moment, is the crack increment during the loading stage in a loading cycle (the overloading stage is included in the loading stage), is the initial crack length, , , are functions of the secondary creep exponent , creep constant, second creep stage exponent, , is a constant coefficient determined by the material itself and the load ratio effect, is the crack opening angle, is the stress-strain state constraint factor, is the elastic modulus of the material, is the yield strength of the material, is the maximum stress intensity factor, is the stress intensity factor threshold.

[0053] In the dynamic Bayesian best-worst network model, we determined several key uncertain parameters: , , , Stress-strain state constraint factor and stress intensity factor threshold value . These parameters are highly sensitive in the dynamic Bayesian network model of crack propagation and are regarded as uncertain nodes. Other variables in the model are regarded as deterministic fixed nodes.

[0054] After setting an initial crack length in the dynamic Bayesian best-worst network model, the actually detected load data is input into the dynamic Bayesian best-worst network model, so as to calculate the crack propagation of the structure on the basis of time series.

[0055] When the observed crack length detected at time t-1 is, then the crack length at the next moment is the optimal estimated value calculated by combining the observed crack length and the prior value of crack length ; if no crack observation value is detected at time t-1, then the crack length at the next moment .

[0056] S6. Parameter inference algorithm based on hybrid adaptive resampling intelligent particle filter to reduce the influence of uncertain parameters in the crack propagation process.

[0057] To reduce the influence of uncertain parameters in the crack propagation process, as Figure 3 shown, a parameter inference algorithm based on hybrid adaptive resampling intelligent particle filter is adopted. Through the dynamic inference mechanism of particle filter and combined with the adaptive resampling strategy, this algorithm can effectively address the problem of lack of particle diversity.

[0058] The following are the specific steps of this algorithm: 1) Prior distribution Extract N initial particles, where N is the number of particles; 2) Extract particles from the importance probability density function , calculate the particle weights according to Equation , and perform normalization processing on the weights to obtain the particle set , where is the particle weight, represents the conditional probability of the observed value under the condition of the known state , is the state value of the particle, is the normalized particle weight; 3) Divide the particle set into a high-weight particle set and the set of low-weight particles , retain the set of high-weight particles, and resample each particle in the set of low-weight particles according to steps (3)-(6); ; In the formula: the weight threshold , represents rounding up, is the weight set arranged in descending order, is the number of effective samples, which can be calculated by the following formula: ; 4) With probability (0 < < 1), select the adaptive Gaussian mutation strategy, and with probability select the crossover strategy. λ is calculated by the following formula: ; Generate a random number r ∈ [0,1]. When r ≤ , execute step (5), otherwise go to step (6), where is the probability of selecting the adaptive Gaussian mutation strategy; 5) According to , generate particle with the adaptive Gaussian mutation strategy, and calculate the weight , where is the particle after Gaussian mutation, , is the number of particles in the set of low-weight particles , is the particle before mutation, , is the number of particles in the set of high-weight particles , is the mean vector and , is the covariance matrix; 6) According to , generate particle with the adaptive crossover strategy, and calculate the weight ;: is the particle after crossover, , is the number of particles in the set of low-weight particles, is the particle in the set of low-weight particles, is the particle in the set of high-weight particles, , is the number of particles in the set of high-weight particles, the random number is the crossover rate, can be calculated using ; 7) Accept or reject the newly generated particles to obtain particles , where , are the particles after acceptance and rejection after resampling, is the particle weight before normalization, is the particle weight before normalization; 8) Put the new particles into the high-weight particle set to obtain a new particle set , when the number of resampled particles is less than , where is the number of particles in the low-weight particle set, go to step (4), otherwise go to step (9); 9) Calculate the final state estimate according to .

[0059] (8) After completing the state estimation of the particles, the parameters of the small time-scale prediction model are obtained, and the state equation of the fatigue crack growth process is: ; where is the crack length at time t - 1, which can be determined by the hybrid intelligent resampling and Gaussian process surrogate model and of each parameter and the key factor of the stress intensity factor, and finally the predicted crack length at time t can be output.

[0060] If there is a difference between the predicted crack growth data and the actual monitored data, the DBN-BWM model will correct the prediction by adjusting its parameters. This adjustment is based on the best-worst method, that is, according to the difference between the prediction and the actual monitoring results, the optimal and worst parameter combinations are selected for adjustment. In this way, the DBN-BWM model can accurately track the crack growth in real time and provide decision support for the maintenance and safe operation of the pressure hull. This dynamic adjustment mechanism enables the model to adapt to the changing working environment and improve the accuracy and reliability of the prediction It is obvious to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Accordingly, all changes that fall within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claims concerned.

Claims

1. A fatigue crack growth prediction method for a deep-sea pressure-resistant cabin with variable operating conditions and multiple stages, characterized in that, Including: S1. Establish a finite element model of the whole pressure-resistant cabin, import the boundary conditions determined by the operating conditions, analyze the characteristics of stress distribution, and determine the weak links; S2. Construct samples for the stress intensity factor - Gaussian process surrogate model, input the generated samples into the finite element model, calculate the stress intensity factor at the crack tip, and establish a stress intensity factor sample database; S3. Based on the Gaussian process theory, establish a stress intensity factor - Gaussian process surrogate model according to the stress intensity factor sample database; S4. Based on the dynamic Bayesian network model, predict the fatigue cracks of the deep-sea pressure-resistant cabin; S5. Based on the fatigue life calculation model on a small time scale, conduct a sensitivity analysis of the uncertain parameters in the dynamic Bayesian network model; S6. Based on the parameter inference algorithm of the hybrid adaptive resampling intelligent particle filter, reduce the influence of uncertain parameters during the crack propagation process.

2. The fatigue crack growth prediction method for a deep-sea pressure-resistant cabin with variable operating conditions in multiple stages according to claim 1, wherein In step S2, the spatial format of the samples defining the stress intensity factor - Gaussian process surrogate model is: ； Among them, is the load magnitude in the loading and overloading stages, is the load magnitude in the hold stage, , are the creep constant and the secondary creep stage exponent respectively, is the stress-strain state constraint factor, is the stress intensity factor threshold value, is the crack propagation value in the loading and overloading stages, is the crack propagation value in the hold stage, and m is the number of rows in the spatial format.

3. The fatigue crack growth prediction method for a deep-sea pressure-resistant cabin with variable operating conditions and multiple stages according to claim 2, wherein A method of generating a sample database using nearest neighbor sampling based on unilateral constraints is adopted. Based on simulating or predicting cracks, d is taken above and below it as the sample generation space, and is used as the unit to expand outward in sequence. The distance between each data point selected and the sample is a multiple of . There are N groups of data above and below, for a total of 2N groups of data.

4. The fatigue crack growth prediction method for a deep-sea pressure-resistant cabin with variable operating conditions and multiple stages according to claim 2, wherein, In step S4, it includes: S41. Determine the weights of various influencing factors, score the influencing factors according to prior experience, and determine the final weights; S42. According to the influencing factors and their weights, construct a dynamic Bayesian network model, where each node represents a risk factor and the arrow represents a causal relationship; S43. Dynamically update the weight distribution in the dynamic Bayesian network model through monitoring data; S44. When there is a deviation in the prediction data of the dynamic Bayesian network model, quickly judge the working state of the pressure-resistant cabin through the reverse inference of the dynamic Bayesian network model, so as to adjust the parameters of the dynamic Bayesian network model to adapt to the crack propagation characteristics in the current stage.

5. The fatigue crack growth prediction method for a deep-sea pressure-resistant cabin with variable operating conditions and multiple stages according to claim 1, characterized in that, In step S5, the fatigue life calculation model on a small time scale is: ； ； ； wherein, is the crack increment during the hold stage in a loading cycle, is the crack increment during the loading stage in a loading cycle, is the initial crack length, , , are functions of the secondary creep exponent respectively, the creep constant and the secondary creep stage exponent, θ maxo is the maximum crack opening angle, is the stress-strain state constraint factor, is the elastic modulus of the material, is the yield strength of the material, is the maximum stress intensity factor, is the stress intensity factor threshold value.

6. The fatigue crack growth prediction method for a deep-sea pressure-resistant cabin with variable working conditions and multiple stages according to claim 1, wherein After setting an initial crack length in the dynamic Bayesian network model, input the actually detected load data into the dynamic Bayesian network model, so as to calculate the crack propagation of the structure on the basis of the time series; When the observed crack length detected at time t-1 , the crack length at the next moment is the optimal estimated value calculated by combining the observed value and the prior value of the crack length ; if no observed value of the crack is detected at time t-1, the crack length at the next moment .

7. The method for predicting fatigue crack growth of a deep-sea pressure-resistant cabin with variable operating conditions and multiple stages according to claim 6, wherein Based on the parameter inference algorithm of the hybrid adaptive resampling intelligent particle filter, conduct state estimation of the particles, and then calculate the state equation of the fatigue crack propagation process as: ； Among them is the crack length at time t-1, which can be determined by hybrid intelligent resampling and Gaussian process surrogate model and parameters of and stress intensity factor, and finally output the predicted crack length at time t .

Images