Test method for failure mechanism of node corrosion fatigue of floating structure under multiple working conditions

Abstract

The application discloses a test method for node corrosion fatigue failure mechanism of a floating structure considering multiple working conditions, and comprises the following steps: S1, constructing an intelligent verification and regulation digital twin parallelly running with a physical test system, wherein the digital twin comprises a multi-scale-multi-fidelity digital twin model library; S2, calculating a comprehensive equivalent deviation in real time, wherein the comprehensive equivalent deviation fuses macroscopic performance deviation and microscopic and mesoscopic feature similarity; S3, constructing an adaptive acceleration ratio scheduler, wherein the comprehensive equivalent deviation and a test progress are input, and adjustment instructions for time-varying acceleration factors of each working condition, working condition residence time and whether to insert a high-fidelity verification cycle are decided and output; S4, receiving the adjustment instructions and executing an updated test plan; and S5, using multi-source monitoring data collected in real time to perform online fine tuning and parameter updating on key sub-process models in the multi-scale-multi-fidelity digital twin model library.

Description

Technical Field

[0001] This invention relates to the field of testing technology for nodes of floating structures, and in particular to a test method for corrosion fatigue failure mechanism of nodes of floating structures considering multiple working conditions. Background Technology

[0002] Floating structures in marine environments serve as core engineering carriers for offshore resource development, offshore wind power, and marine ranching. They constantly face complex and variable coupled environments including wave impact, wind loads, temperature fluctuations, and salt spray corrosion. Their nodal points, as critical areas of the structural force transmission path, are susceptible to corrosion fatigue failure under the combined effects of corrosive media erosion and alternating cyclic loads, seriously threatening the overall structural safety and service reliability. Therefore, this study systematically investigates the corrosion fatigue failure mechanism of nodal points under multiple operating conditions. The aim is to reveal the effects of different operating parameters (such as load amplitude, frequency, corrosive medium concentration, temperature gradient, and flow velocity distribution) on the corrosion rate, crack initiation life, and other parameters of the nodal materials. This study investigates the influence of propagation rate and final fracture mode, verifies and optimizes the applicability and accuracy of existing experimental methods in multi-condition environment simulation, quantitative characterization of corrosion-fatigue coupling effects, and real-time monitoring of local stress-strain fields at nodes. This provides a scientific basis for improving safety design standards for floating structures, constructing remaining service life prediction models, optimizing corrosion protection strategies, and supporting maintenance decisions. It has irreplaceable theoretical innovation value and engineering practical significance for enhancing the corrosion fatigue resistance of marine engineering structures, extending service life, reducing life-cycle costs, ensuring safety in offshore operations, promoting the reliable application of floating structures in extreme marine environments, and advancing technology in the field of marine engineering.

[0003] In existing technologies, floating structures exhibit strong nonlinear and time-varying coupling relationships with mechanical loads, corrosive environments, and structural responses under different operating conditions. Traditional controllers with fixed parameters struggle to accurately track complex multiphysics targets. Therefore, this paper proposes an experimental method for understanding the corrosion fatigue failure mechanism of floating structure nodes under multiple operating conditions. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an experimental method for the corrosion fatigue failure mechanism of floating structure nodes that considers multiple working conditions.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An experimental method for assessing the corrosion fatigue failure mechanism of nodal nodes in floating structures under multiple operating conditions includes the following steps: S1: Construct an intelligent verification and control digital twin that runs in parallel with the physical test system. The digital twin includes a multi-scale and multi-fidelity digital twin model library, which is used to simulate the multi-scale structural evolution and performance degradation process of the node specimen under the current accelerated test conditions and equivalent real service conditions in parallel in virtual space. S2: Based on the simulation results output by the multi-scale-multi-fidelity digital twin model library, the comprehensive equivalence deviation is calculated in real time. The comprehensive equivalence deviation integrates macroscopic performance deviation and micro-mesoscopic feature similarity to characterize the fidelity of accelerating the test to reproduce the actual service failure mechanism. S3: Construct an adaptive speedup scheduler, with the comprehensive equivalence deviation and test progress as inputs, and make decisions and output adjustment instructions for the time-varying acceleration factor, dwell time of each working condition, and whether to insert a high-fidelity verification loop for each working condition. S4: The control architecture adopts generalized predictive control, receives the adjustment command, performs coordinated tracking control of mechanical loading, electrochemical environment and temperature field, and executes the updated test plan; S5: During the experiment, the key sub-process models in the multi-scale-multi-fidelity digital twin model library are fine-tuned and their parameters are updated online using real-time multi-source monitoring data, so as to realize the autonomous evolution of the model's predictive ability.

[0006] The above further includes: Furthermore, a multi-scale, multi-fidelity digital twin model library is constructed, specifically including: It integrates historical service data from the service history of the target floating structure, multi-scale numerical simulation results based on computational fluid dynamics and finite element methods, and basic material constitutive models and physicochemical mechanism models covering macroscopic to microscopic scales; Establish a cross-scale parameter transfer interface and a two-way data exchange protocol between the macroscopic continuum mechanical model, the microscopic phase field or crystal plasticity model and the atomic-scale proxy model; Before the experiment began, the integrated historical service data and multi-scale numerical simulation results were used to jointly pre-train and calibrate the multi-scale model library with the constructed coupled architecture. During the experiment, the initialized digital twin model library was deployed as a real-time simulation engine. The real-time simulation engine synchronously received real-time monitoring data from the physical test system and used it as boundary conditions and stimuli to run the first set of tasks and the second set of tasks in parallel in the virtual space. The first set of tasks simulated the multi-scale evolution of the test specimen in the accelerated environment using the accelerated test conditions parameters executed by the current physical test system. The second set of tasks simulated the multi-scale evolution of the test specimen in the real environment using the real service conditions parameters after equivalence mapping. The real-time simulation engine outputs the multi-scale simulation results generated from the derivation of the first set of tasks and the second set of tasks to the comprehensive equivalence deviation calculation module in real time.

[0007] Furthermore, the specific steps for calculating the comprehensive equivalence deviation are as follows: Calculate the deviation between the real-time measured values ​​of macroscopic performance parameters and the actual service simulation values ​​simulated by the digital twin; High-dimensional features are extracted from real-time in-situ monitoring data and corresponding simulation data, and the dynamic time warping similarity between the two is calculated. The macroscopic performance deviation and the feature similarity are weighted, fused, and normalized to generate a scalar value between 0 and 1 as the comprehensive equivalence deviation.

[0008] Furthermore, the specific steps for the adaptive speedup scheduler to perform adaptive speedup scheduling are as follows: Configure an initial time-varying acceleration factor and dwell time for each typical working condition in the current test mission, and set warning thresholds and action thresholds for the deviation of comprehensive equivalence. During the test run, the comprehensive equivalence deviation calculated in real time is continuously received and the test progress is monitored. The test progress includes the current working condition indicator, the time already executed, and the planned remaining time. Based on the received comprehensive equivalence deviation sequence and its changing trend, the corresponding decision logic is triggered. Based on the decision results, specific scheduling control instructions are generated and output. These instructions include adjustments to the time-varying acceleration factor, revisions to the dwell time of the working condition, and instructions on whether to initiate a high-fidelity verification cycle or microscopic morphology scan. While deciding to adjust the time-varying acceleration factor or insert a high-fidelity verification cycle, the scheduler will dynamically extend the dwell time of the current working condition based on the adjustment time or the expected duration of the verification cycle to ensure that the total cumulative damage equivalent under this working condition is equivalent to the original test plan. The revised time parameters will also be synchronized to the test progress management module. The generated scheduling instructions are sent to the lower-level generalized predictive control module and test equipment for execution. After the instructions are executed, the new comprehensive equivalence deviation is monitored to evaluate the scheduling effect and complete the decision-making closed loop.

[0009] After completing one round of scheduling instruction output, the adaptive speedup scheduler will enter a waiting and evaluation period. During this period, it suspends new major decisions, continues to collect test response data after the lower-level control system executes instructions and newly calculated comprehensive equivalence deviation, and evaluates the effectiveness of previous scheduling instructions; The next round of decision-making is only triggered when the assessment confirms that the scheduling effect has not met expectations or that the deviation has shown a new trend of deterioration, in order to avoid decision oscillation.

[0010] Furthermore, the decision logic includes: First-level decision logic: When the real-time comprehensive equivalence deviation is detected to exceed the warning threshold for the first time but not to reach the action threshold, the adaptive acceleration ratio scheduler decides to automatically reduce the load or environmental acceleration factor of the current working condition. The reduction magnitude is determined by calculating the proportion exceeding the warning threshold and the value of the current acceleration factor through a preset lookup table or attenuation function. Second-level decision logic: When it is detected that the overall equivalence deviation continues to rise after taking measures to reduce the acceleration factor, or when the overall equivalence deviation directly exceeds the action threshold, the adaptive speedup scheduler decides to insert a high-fidelity verification loop. The instructions of the high-fidelity verification loop are to control the physical test system to briefly switch to a set of predefined settings that are closer to the real service environment parameters and run for a short period of time to obtain high-confidence data for calibrating the digital twin model. The third-level decision logic is as follows: based on the preset test stage, the micro-verification instruction is triggered. The adaptive speedup scheduler, according to the specific stage nodes planned by the test progress, or when the comprehensive equivalence deviation is fluctuating at a high level for a long time, decides to trigger the micro-sampling or high-resolution in-situ scanning instruction. The scanning instruction will coordinate the external micro-analysis equipment to sample or scan the specific parts of the specimen online to obtain the actual corrosion product morphology and crack tip micro-morphology data for direct comparison and verification.

[0011] Furthermore, the specific steps for the control architecture to perform cooperative tracking control are as follows: S4.1: Receive and parse adaptive scheduling instructions to generate a sequence of underlying control setpoints; More specifically, it receives adjustment instructions from the upper-level adaptive speedup scheduler. These instructions include the updated time-varying acceleration factor, dwell time for each operating condition, and possible high-fidelity verification loop insertion points. The control architecture parses these instructions and combines them with pre-stored reference load spectra and environmental spectra. Through a time-amplitude scaling algorithm, it generates in real-time a high-resolution target setpoint sequence for the mechanical loading system (multi-axis load), electrochemical environmental chamber (potential, solution concentration, pH value), and temperature control system within a future prediction time domain. This sequence serves as the tracking target for the generalized predictive controller. S4.2: Perform real-time multi-physics data acquisition. Through the sensor network integrated on the experimental device, synchronously and at high speed acquire data of mechanical (multi-channel strain, displacement), electrochemical (working electrode potential, current density, electrochemical impedance) and temperature fields, and estimate the current state vector in real time. S4.3: Perform multi-step prediction and rolling optimization calculations for generalized predictive control. The generalized predictive controller takes the current state estimate and the future sequence of underlying control setpoints as inputs. At each sampling time, the generalized predictive controller predicts the future output behavior within the set prediction time domain. At the same time, within the control time domain, it calculates the optimal future control input increment by solving an optimization cost function with the goal of the sum of a weighted quadratic form of the future tracking error and the change in control input. S4.4: Output the collaborative control signal and execute closed-loop feedback. The optimal control increment calculated in S4.3 at the current moment is superimposed with the control quantity at the previous moment to generate the collaborative control signal actually sent to each actuator (servo actuator, potentiostat, temperature controller). This signal drives the physical system to achieve precise control of mechanical loading, electrochemical environment and temperature field. Subsequently, the system returns to step S4.2 to perform state perception based on a new round of sensor data.

[0012] Furthermore, the specific method for online fine-tuning of the key sub-process model is as follows: To address the physicochemical processes of corrosion product film growth and hydrogen diffusion-induced cracking, a physical information neural network or deep generative model with embedded physical mechanism constraints is constructed as a sub-model. During the experiment, the parameters of the sub-model were continuously updated in an online learning manner using real-time collected multi-source data of current, potential, and hydrogen permeation signals, so that the digital twin could adaptively characterize the strong nonlinear coupling effect caused by damage accumulation and operating condition switching.

[0013] Furthermore, during the experiment, a cross-scale validation data and decision knowledge base is constructed to store and mine the association rules between multi-scale validation data, experimental parameters, the comprehensive equivalence deviation, and scheduling decisions.

[0014] The present invention has the following beneficial effects: In this invention, through the collaboration of an adaptive speedup scheduler and an online evolution mechanism for sub-models, the system possesses dynamic adaptability. The adaptive speedup scheduler dynamically adjusts the acceleration strategy based on the real-time equivalence deviation. Meanwhile, the online evolution mechanism continuously updates the key sub-models in the twin using real-time data, enabling the system to identify and compensate for nonlinear dynamic changes caused by operating condition switching and damage accumulation in real time, thus maintaining the accuracy and stability of the control system. Attached Figure Description

[0015] Figure 1 This is a step diagram of the experimental method for assessing the corrosion fatigue failure mechanism of nodal nodes in floating structures, which takes into account multiple working conditions, as proposed in this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 As shown, this invention provides an experimental method for assessing the corrosion fatigue failure mechanism of nodal corrosion in floating structures under multiple operating conditions, comprising the following steps: S1: Construct an intelligent verification and control digital twin that runs in parallel with the physical test system. The digital twin includes a multi-scale and multi-fidelity digital twin model library, which is used to simulate the multi-scale structural evolution and performance degradation process of the node specimen under the current accelerated test conditions and equivalent real service conditions in parallel in virtual space. S2: Based on the simulation results output by the multi-scale-multi-fidelity digital twin model library, the comprehensive equivalence deviation is calculated in real time. The comprehensive equivalence deviation integrates macroscopic performance deviation and micro-mesoscopic feature similarity to characterize the fidelity of accelerating the test to reproduce the actual service failure mechanism. S3: Construct an adaptive speedup scheduler, with the comprehensive equivalence deviation and test progress as inputs, and make decisions and output adjustment instructions for the time-varying acceleration factor, dwell time of each working condition, and whether to insert a high-fidelity verification loop for each working condition. S4: The control architecture adopts generalized predictive control, receives the adjustment command, performs coordinated tracking control of mechanical loading, electrochemical environment and temperature field, and executes the updated test plan; S5: During the experiment, the key sub-process models in the multi-scale-multi-fidelity digital twin model library are fine-tuned and their parameters are updated online using real-time multi-source monitoring data, so as to realize the autonomous evolution of the model's predictive ability.

[0018] In one embodiment, constructing a multi-scale, multi-fidelity digital twin model library specifically includes: It integrates historical service data from the service history of the target floating structure, multi-scale numerical simulation results based on computational fluid dynamics and finite element methods, and basic material constitutive models and physicochemical mechanism models covering macroscopic to microscopic scales; More specifically, the integrated model includes at least a macroscopic continuum mechanics model for simulating the overall stress-strain response of the structure, a microscopic phase-field model or crystal plasticity model for simulating grain-scale plastic deformation and microcrack initiation and propagation, and an atomic-scale surrogate model for describing hydrogen atom diffusion and trapping behavior. Establish a cross-scale parameter transfer interface and a two-way data exchange protocol between the macroscopic continuum mechanical model, the microscopic phase field or crystal plasticity model and the atomic-scale proxy model; More specifically, the hydrogen effective diffusion coefficient and trap binding energy distribution parameters calculated by the atomic-scale proxy model are passed as key input parameters to the microscopic phase field or crystal plasticity model. Meanwhile, the microcrack distribution density and local plastic strain field predicted by the micro-phase field or crystal plasticity model are mapped to the initial damage field and anisotropic stiffness reduction coefficient in the macro-continuous medium mechanics model, realizing the mechanism correlation and coupling calculation from the atomic, micro-scale to the macro-scale. Before the experiment began, the integrated historical service data and multi-scale numerical simulation results were used to jointly pre-train and calibrate the multi-scale model library with the constructed coupled architecture. More specifically, the calibration process adjusts the undetermined parameters in the models at each scale to minimize the deviation between the model library's simulation predictions of structural response, corrosion rate and crack propagation behavior under known service conditions and historical service data or high-confidence numerical simulation benchmarks, thereby obtaining a high-fidelity initialized digital twin model library. During the experiment, the initialized digital twin model library was deployed as a real-time simulation engine. The real-time simulation engine synchronously received real-time monitoring data from the physical test system and used it as boundary conditions and stimuli to run the first set of tasks and the second set of tasks in parallel in the virtual space. The first set of tasks simulated the multi-scale evolution of the test specimen in the accelerated environment using the accelerated test conditions parameters executed by the current physical test system. The second set of tasks simulated the multi-scale evolution of the test specimen in the real environment using the real service conditions parameters after equivalence mapping. The real-time simulation engine outputs the multi-scale simulation results generated by the derivation of the first set of tasks and the second set of tasks to the comprehensive equivalence deviation calculation module in real time. More specifically, the simulation results include at least the crack length evolution curve and residual stiffness value at the macroscopic scale, the local corrosion current density distribution and microcrack morphology feature vector at the microscopic scale, and the spatiotemporal distribution data of stress field, strain field and hydrogen concentration field at the mesoscopic scale; at the same time, the deviation data between the simulation prediction value and the actual monitoring value are fed back to the online evolution module of the sub-model to drive the online update of the model.

[0019] In one embodiment, the specific steps for calculating the overall equivalence deviation are as follows: Calculate the deviation between the real-time measured values ​​of macroscopic performance parameters and the actual service simulation values ​​simulated by the digital twin; It should be noted that the specific analytical process for calculating the deviation is as follows: From the multi-scale-multi-fidelity digital twin model library, obtain the simulation sequence of macroscopic performance parameters obtained from simulations under equivalent real service conditions; The corresponding measured sequence of macroscopic performance parameters is obtained from the real-time monitoring data of the physical test system. The macroscopic performance parameters include at least the crack length propagation rate, the local strain energy density of the specimen, and the corrosion current density monitored by the electrochemical workstation. The deviation between the measured sequence and the simulated sequence at the same time or under the same load cycle is calculated, and the deviations of multiple macroscopic performance parameters are normalized and weighted to obtain macroscopic performance deviation components. The deviation values ​​are calculated using the relative error or root mean square error method, and the weights of the weighted summation are determined based on the sensitivity analysis of the contribution of each macroscopic performance parameter to node failure. High-dimensional features are extracted from real-time in-situ monitoring data and corresponding simulation data, and the dynamic time warping similarity between the two is calculated. It should be noted that the specific analytical process for calculating dynamic time-warped similarity is as follows: Extract microscopic or mesoscopic physical field data generated from simulations of equivalent real service conditions from the multi-scale-multi-fidelity digital twin model library; The high-resolution in-situ monitoring data acquired by the physical experiment system is preprocessed. The high-resolution in-situ monitoring data includes the local electrochemical impedance spectral distribution map acquired by the scanning galvanometer system, the time spectrum of the acoustic emission signal acquired by the acoustic emission sensor array, and the full-field strain distribution cloud map acquired by the digital image correlation system. The preprocessing includes noise reduction, alignment, and format normalization operations to ensure that it is comparable to the simulated physical field data in the spatial and temporal dimensions. The simulated physical field data and the preprocessed in-situ monitoring data are respectively input into a pre-trained deep learning feature extraction network to obtain the corresponding high-dimensional feature vectors. The pre-trained deep learning feature extraction network is a convolutional neural network or a visual Transformer model. It is trained in a supervised manner using paired simulation data and actual observation image data obtained from historical experiments. The training objective is to enable the network to extract robust high-dimensional features that characterize typical corrosion fatigue damage patterns (such as the plastic zone at the crack tip, the morphology of corrosion product film, and hydrogen-induced microcracks). Calculate the dynamic time warping distance between two sets of high-dimensional feature vectors, and map this distance to micro- and meso-level feature similarity components between 0 and 1. The specific function for mapping the dynamic time warping distance to micro- and meso-level feature similarity components is an exponential decay function, which is used to control the sensitivity of the distance to similarity decay. The macroscopic performance deviation and the feature similarity are weighted, fused, and normalized to generate a scalar value between 0 and 1 as the comprehensive equivalence deviation.

[0020] It should be noted that the specific analysis process for weighted fusion and normalization is as follows: The calculated macroscopic performance deviation component and the calculated micro-mesoscopic feature similarity component are weighted and fused, wherein the weight coefficients are pre-set according to the primary and secondary relationship of the failure mechanisms of interest in the experiment or are dynamically recommended by the cross-scale verification data and decision knowledge base. The weighted fusion result is normalized to generate a scalar value between 0 and 1, which serves as the real-time deviation of the overall equivalence.

[0021] In one embodiment, the specific steps for the adaptive speedup scheduler to perform adaptive speedup scheduling are as follows: Configure an initial time-varying acceleration factor and dwell time for each typical working condition in the current test mission, and set warning thresholds and action thresholds for the deviation of comprehensive equivalence. During the test run, the comprehensive equivalence deviation calculated in real time is continuously received and the test progress is monitored. The test progress includes the current working condition indicator, the time already executed, and the planned remaining time. Based on the received comprehensive equivalence deviation sequence and its changing trend, the corresponding decision logic is triggered. Based on the decision results, specific scheduling control instructions are generated and output. These instructions include adjustments to the time-varying acceleration factor, revisions to the dwell time of the working condition, and instructions on whether to initiate a high-fidelity verification cycle or microscopic morphology scan. While deciding to adjust the time-varying acceleration factor or insert a high-fidelity verification cycle, the scheduler will dynamically extend the dwell time of the current working condition based on the adjustment time or the expected duration of the verification cycle to ensure that the total cumulative damage equivalent under this working condition is equivalent to the original test plan. The revised time parameters will also be synchronized to the test progress management module. The generated scheduling instructions are sent to the lower-level generalized predictive control module and test equipment for execution. After the instructions are executed, the new comprehensive equivalence deviation is monitored to evaluate the scheduling effect and complete the decision-making closed loop.

[0022] After completing one round of scheduling instruction output, the adaptive speedup scheduler will enter a waiting and evaluation period. During this period, it suspends new major decisions, continues to collect test response data after the lower-level control system executes instructions and newly calculated comprehensive equivalence deviation, and evaluates the effectiveness of previous scheduling instructions; The next round of decision-making is only triggered when the assessment confirms that the scheduling effect has not met expectations or that the deviation has shown a new trend of deterioration, in order to avoid decision oscillation.

[0023] In one embodiment, the decision logic includes: First-level decision logic: When the real-time comprehensive equivalence deviation is detected to exceed the warning threshold for the first time but not to reach the action threshold, the adaptive acceleration ratio scheduler decides to automatically reduce the load or environmental acceleration factor of the current working condition. The reduction magnitude is determined by calculating the proportion exceeding the warning threshold and the value of the current acceleration factor through a preset lookup table or attenuation function. Second-level decision logic: When it is detected that the overall equivalence deviation continues to rise after taking measures to reduce the acceleration factor, or when the overall equivalence deviation directly exceeds the action threshold, the adaptive speedup scheduler decides to insert a high-fidelity verification loop. The instructions of the high-fidelity verification loop are to control the physical test system to briefly switch to a set of predefined settings that are closer to the real service environment parameters and run for a short period of time to obtain high-confidence data for calibrating the digital twin model. The third-level decision logic is as follows: based on the preset test stage, the micro-verification instruction is triggered. The adaptive speedup scheduler, according to the specific stage nodes planned by the test progress, or when the comprehensive equivalence deviation is fluctuating at a high level for a long time, decides to trigger the micro-sampling or high-resolution in-situ scanning instruction. The scanning instruction will coordinate the external micro-analysis equipment to sample or scan the specific parts of the specimen online to obtain the actual corrosion product morphology and crack tip micro-morphology data for direct comparison and verification.

[0024] In one embodiment, the specific steps for the control architecture to perform cooperative tracking control are as follows: S4.1: Receive and parse adaptive scheduling instructions to generate a sequence of underlying control setpoints; More specifically, it receives adjustment instructions from the upper-level adaptive speedup scheduler. These instructions include the updated time-varying acceleration factor, dwell time for each operating condition, and possible high-fidelity verification loop insertion points. The control architecture parses these instructions and combines them with pre-stored reference load spectra and environmental spectra. Through a time-amplitude scaling algorithm, it generates in real-time a high-resolution target setpoint sequence for the mechanical loading system (multi-axis load), electrochemical environmental chamber (potential, solution concentration, pH value), and temperature control system within a future prediction time domain. This sequence serves as the tracking target for the generalized predictive controller. S4.2: Perform real-time multi-physics data acquisition. Through the sensor network integrated on the experimental device, synchronously and at high speed acquire data of mechanical (multi-channel strain, displacement), electrochemical (working electrode potential, current density, electrochemical impedance) and temperature fields, and estimate the current state vector in real time. S4.3: Perform multi-step prediction and rolling optimization calculations for generalized predictive control. The generalized predictive controller takes the current state estimate and the future sequence of underlying control setpoints as inputs. At each sampling time, the generalized predictive controller predicts the future output behavior within the set prediction time domain. At the same time, within the control time domain, it calculates the optimal future control input increment by solving an optimization cost function with the goal of the sum of a weighted quadratic form of the future tracking error and the change in control input. S4.4: Output coordinated control signal and execute closed-loop feedback: The optimal control increment calculated in S4.3 at the current moment is superimposed with the control quantity at the previous moment to generate a coordinated control signal that is actually sent to each actuator (servo actuator, potentiostat, temperature controller). This signal drives the physical system to achieve precise control of mechanical loading, electrochemical environment and temperature field. Then, the system returns to step S4.2 to perform state perception based on a new round of sensor data.

[0025] In one embodiment, the specific method for online fine-tuning of the key subprocess model is as follows: To address the physicochemical processes of corrosion product film growth and hydrogen diffusion-induced cracking, a physical information neural network or deep generative model with embedded physical mechanism constraints is constructed as a sub-model. It should be noted that the specific analysis process for constructing the sub-model is as follows: For the two key physicochemical sub-processes of corrosion product film growth and hydrogen diffusion-induced cracking in the corrosion fatigue coupling process, dedicated sub-models based on physical information neural networks or deep generative models are constructed respectively. The sub-model is embedded with corresponding physical mechanism constraints during initial training. These physical mechanism constraints include partial differential equations describing membrane growth dynamics or control equations describing hydrogen concentration diffusion and trapping processes. During the experiment, the parameters of the sub-model were continuously updated in an online learning manner using real-time collected multi-source data of current, potential, and hydrogen permeation signals, so that the digital twin could adaptively characterize the strong nonlinear coupling effect caused by damage accumulation and operating condition switching.

[0026] It should be noted that the specific analysis process for online fine-tuning and parameter updates is as follows: During the experiment, multi-source real-time data streams from the electrochemical workstation, hydrogen permeation sensor, acoustic emission sensor and strain measurement device were simultaneously acquired. The data streams included at least current, potential, hydrogen permeation transient current and acoustic emission event signals. The multi-source data is time-stamp aligned and features are extracted to form a high-dimensional temporal feature vector for sub-model updates; An online learning process for the dedicated sub-model is established. The temporal feature vector obtained in step S5.2 is used as an incremental input. Gradient-based optimization algorithms or sequential Bayesian inference methods are used to perform small-batch, continuous incremental updates on the network weights or generator parameters of the sub-model, thereby enabling the model parameters to evolve autonomously with the experimental process. The sub-models, after online fine-tuning, are re-integrated into the multi-scale-multi-fidelity digital twin model library, replacing the original modules; The updated digital twin was then subjected to forward simulation verification using the latest collected measured data. The consistency between its predicted output and the actual monitoring results was compared to evaluate and confirm the effectiveness of the model's improved predictive capabilities. The key information of each online update of the sub-model, including the data features that triggered the update, the amount of parameter update, and the change in model prediction accuracy before and after the update, is recorded and stored in the cross-scale validation data and decision knowledge base to form a model evolution history chain, providing data support for subsequent model optimization and scheduling decisions.

[0027] In one embodiment, during the experiment, a cross-scale validation data and decision knowledge base is constructed to store and mine the association rules between multi-scale validation data, experimental parameters, the comprehensive equivalence deviation, and scheduling decisions; Systematically store full-spectrum verification data, from macroscopic mechanical response and electrochemical signals to microscopic morphology images, along with their corresponding experimental conditions and overall equivalence deviations; By utilizing graph neural networks to mine the mapping relationship between "experimental parameter combinations - multi-scale response characteristics - comprehensive equivalence deviation", an interpretable decision rule graph is formed, which is used to provide priority strategy recommendations based on historical experience when the adaptive speedup scheduler faces multiple options.

[0028] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An experimental method for assessing the corrosion fatigue failure mechanism of nodal nodes in floating structures under multiple operating conditions, characterized by: Includes the following steps: S1: Construct an intelligent verification and control digital twin that runs in parallel with the physical test system. The digital twin includes a multi-scale and multi-fidelity digital twin model library, which is used to simulate the multi-scale structural evolution and performance degradation process of the node specimen under the current accelerated test conditions and equivalent real service conditions in parallel in virtual space. S2: Based on the simulation results output by the multi-scale-multi-fidelity digital twin model library, the comprehensive equivalence deviation is calculated in real time. The comprehensive equivalence deviation integrates macroscopic performance deviation and micro-mesoscopic feature similarity to characterize the fidelity of accelerating the test to reproduce the actual service failure mechanism. S3: Construct an adaptive speedup scheduler, with the comprehensive equivalence deviation and test progress as inputs, and make decisions and output adjustment instructions for the time-varying acceleration factor, dwell time of each working condition, and whether to insert a high-fidelity verification loop for each working condition. S4: The control architecture adopts generalized predictive control, receives the adjustment command, performs coordinated tracking control of mechanical loading, electrochemical environment and temperature field, and executes the updated test plan; S5: During the experiment, the key sub-process models in the multi-scale-multi-fidelity digital twin model library are fine-tuned and their parameters are updated online using real-time multi-source monitoring data.

2. The test method for corrosion fatigue failure mechanism of floating structure nodes considering multiple working conditions according to claim 1, characterized in that: Constructing a multi-scale, multi-fidelity digital twin model library, specifically including: It integrates historical service data from the service history of the target floating structure, multi-scale numerical simulation results based on computational fluid dynamics and finite element methods, and basic material constitutive models and physicochemical mechanism models covering macroscopic to microscopic scales; Establish a cross-scale parameter transfer interface and a two-way data exchange protocol between the macroscopic continuum mechanical model, the microscopic phase field or crystal plasticity model and the atomic-scale proxy model; Before the experiment began, the integrated historical service data and multi-scale numerical simulation results were used to jointly pre-train and calibrate the multi-scale model library with the coupled architecture. During the experiment, the initialized digital twin model library was deployed as a real-time simulation engine. The real-time simulation engine synchronously received real-time monitoring data from the physical test system and used it as boundary conditions and stimuli to run the first set of tasks and the second set of tasks in parallel in the virtual space. The first set of tasks simulated the multi-scale evolution of the test specimen in the accelerated environment using the accelerated test conditions parameters executed by the current physical test system. The second set of tasks simulated the multi-scale evolution of the test specimen in the real environment using the real service conditions parameters after equivalence mapping. The real-time simulation engine outputs the multi-scale simulation results generated from the derivation of the first set of tasks and the second set of tasks to the comprehensive equivalence deviation calculation module in real time.

3. The test method for corrosion fatigue failure mechanism of floating structure nodes considering multiple working conditions according to claim 1, characterized in that: The specific steps for calculating the overall equivalence deviation are as follows: Calculate the deviation between the real-time measured values ​​of macroscopic performance parameters and the actual service simulation values ​​simulated by the digital twin; High-dimensional features are extracted from real-time in-situ monitoring data and corresponding simulation data, and the dynamic time warping similarity between the two is calculated. The macroscopic performance deviation and the feature similarity are weighted, fused, and normalized to generate a scalar value between 0 and 1 as the comprehensive equivalence deviation.

4. The test method for corrosion fatigue failure mechanism of floating structure nodes considering multiple working conditions according to claim 1, characterized in that: The specific steps for the adaptive speedup scheduler to perform adaptive speedup scheduling are as follows: Configure an initial time-varying acceleration factor and dwell time for each typical working condition in the current test mission, and set warning thresholds and action thresholds for the deviation of comprehensive equivalence. During the test run, the comprehensive equivalence deviation calculated in real time is continuously received and the test progress is monitored. The test progress includes the current working condition indicator, the time already executed, and the planned remaining time. Based on the received comprehensive equivalence deviation sequence and its changing trend, the corresponding decision logic is triggered. Based on the decision results, specific scheduling control instructions are generated and output. The scheduling control instructions include the adjustment amount of the time-varying acceleration factor, the revision value of the working condition dwell time, and instructions on whether to start a high-fidelity verification cycle or micro-morphology scan. The generated scheduling instructions are sent to the lower-level generalized predictive control module and test equipment for execution, and after the instructions are executed, the new comprehensive equivalence deviation is monitored to evaluate the scheduling effect.

5. The test method for corrosion fatigue failure mechanism of floating structure nodes considering multiple working conditions according to claim 4, characterized in that: The decision-making logic includes: First-level decision logic: When the real-time comprehensive equivalence deviation is detected to exceed the warning threshold for the first time but not to reach the action threshold, the adaptive acceleration ratio scheduler decides to automatically reduce the load or environmental acceleration factor of the current working condition. The reduction magnitude is determined by calculating the proportion exceeding the warning threshold and the value of the current acceleration factor through a preset lookup table or attenuation function. Second-level decision logic: When it is detected that the overall equivalence deviation continues to rise after the acceleration factor reduction measures are taken, or the overall equivalence deviation directly exceeds the action threshold, the adaptive speedup scheduler decides to insert a high-fidelity verification loop. The instructions of the high-fidelity verification loop are to control the physical test system to briefly switch to a set of environmental parameter settings and run for a short period of time to obtain high-confidence data for calibrating the digital twin model. The third-level decision logic is as follows: based on the preset test stage, the micro-verification instruction is triggered. The adaptive speedup scheduler, according to the specific stage nodes planned by the test progress, or when the comprehensive equivalence deviation is fluctuating at a high level for a long time, decides to trigger the micro-sampling or high-resolution in-situ scanning instruction. The scanning instruction will coordinate the external micro-analysis equipment to sample or scan the specific parts of the specimen online to obtain the actual corrosion product morphology and crack tip micro-morphology data for direct comparison and verification.

6. The test method for corrosion fatigue failure mechanism of floating structure nodes considering multiple working conditions according to claim 1, characterized in that: The specific steps for the control architecture to perform collaborative tracking control are as follows: S4.1: Receive and parse adaptive scheduling instructions to generate a sequence of underlying control setpoints; S4.2: Perform real-time multi-physics data acquisition. Through the sensor network integrated on the experimental device, synchronously and at high speed acquire data of mechanical, electrochemical and temperature fields, and estimate the current state vector in real time. S4.3: Perform multi-step prediction and rolling optimization calculations for generalized predictive control. The generalized predictive controller takes the current state estimate and the future sequence of underlying control setpoints as inputs. At each sampling time, the generalized predictive controller predicts the future output behavior within the set prediction time domain. At the same time, within the control time domain, it calculates the optimal future control input increment by solving an optimization cost function with the goal of the sum of a weighted quadratic form of the future tracking error and the change in control input. S4.4: The optimal control increment calculated in S4.3 at the current moment is superimposed with the control quantity at the previous moment to generate the actual coordinated control signal sent to each actuator.

7. The test method for corrosion fatigue failure mechanism of floating structure nodes considering multiple working conditions according to claim 1, characterized in that: The specific method for online fine-tuning of the key subprocess model is as follows: To address the physicochemical processes of corrosion product film growth and hydrogen diffusion-induced cracking, a physical information neural network or deep generative model with embedded physical mechanism constraints is constructed as a sub-model. During the experiment, the parameters of the sub-model were continuously updated using real-time multi-source data and online learning, enabling the digital twin to adaptively characterize the strong nonlinear coupling effect caused by damage accumulation and operating condition switching.

8. The test method for corrosion fatigue failure mechanism of floating structure nodes considering multiple working conditions according to claim 1, characterized in that: During the experiment, a cross-scale validation data and decision knowledge base was constructed to store and mine the association rules between multi-scale validation data, experimental parameters, the comprehensive equivalence deviation, and scheduling decisions.

Images