Carbon steel corrosion modeling method based on finite element

Abstract

The invention discloses a finite element-based carbon steel corrosion modeling method, which relates to the technical field of computational mechanics and finite element analysis, and comprises the following steps: S1, acquiring real-time data from an environment sensor, acquiring an initial parameter set containing temperature, salinity and material attributes, and determining initial corrosion rate distribution; s3, if the deviation value exceeds a preset threshold value, a parameter coupling relation is processed through a genetic algorithm, and an optimized parameter set is obtained and used for correcting model input; according to the finite element-based carbon steel corrosion modeling method, the precision and reliability of corrosion prediction in a complex environment are remarkably improved, and a scientific basis is provided for carbon steel structure maintenance.

Description

Technical Field

[0001] This invention relates to the fields of computational mechanics and finite element analysis, specifically to a finite element-based method for modeling carbon steel corrosion. Background Technology

[0002] Corrosion modeling, a crucial research area in materials science and engineering, is essential for predicting the performance of metallic structures in harsh environments, particularly in industries such as aerospace, energy, and marine engineering, where it directly impacts equipment safety and lifespan. Accurate corrosion prediction can significantly reduce maintenance costs and prevent catastrophic failures, thus becoming a core element driving industrial technological progress. However, the limitations of existing methods are becoming increasingly apparent, especially their insufficient prediction accuracy in complex environments, making it difficult to adapt to varying working conditions and material properties, leading to significant deviations between predicted results and actual corrosion behavior. This deviation not only affects model reliability but can also lead to misjudgments in engineering design, thereby increasing safety risks. Current research primarily relies on the finite element method to simulate corrosion processes, but its core challenge lies in accurately capturing the sources of systematic bias. Systematic biases typically stem from the inaccuracies of model parameters, including material properties, environmental factors, and corrosion rates. Since these parameters are often coupled in real-world environments—for example, temperature changes affect corrosion rates, which in turn alter the chemical properties of the material surface—it is difficult for the model to maintain consistency in dynamic environments. This coupling effect makes it difficult for initial parameter settings to accurately reflect the real-world scenario, leading to a continuous accumulation of discrepancies between predicted and measured values. Furthermore, the bias caused by this parameter inaccuracy is particularly significant in complex geometries or non-uniform environments. For example, in marine environments, the corrosion rate variation on a ship's surface due to local salinity differences is difficult to accurately describe using a single set of parameters. Models often cannot adaptively adjust parameters to cope with such local variations, thus amplifying the bias when predicting the corrosion behavior of complex structures. This lack of dynamic adaptability makes it difficult for models to meet high-precision requirements in practical applications. Therefore, how to effectively identify and correct systematic biases caused by parameter coupling and dynamic changes in finite element corrosion modeling by optimizing parameter adjustment mechanisms has become a key issue in improving prediction accuracy. Specifically, in real-world engineering scenarios, such as the steel structure of a cross-sea bridge, the corrosion rate varies significantly in different parts of the surface due to regional differences in seawater salinity and temperature. Existing finite element models are often based on uniform initial parameter settings and cannot reflect these local differences in real time. For example, the part of the bridge pier closer to the water surface corrodes faster due to higher salinity, while the corrosion rate of the higher parts is lower, but the model may incorrectly predict a uniform corrosion distribution. This deviation stems not only from the static nature of the parameter settings but also from the model's lack of responsiveness to dynamic environmental changes, making it difficult to adjust parameters based on measured data during the prediction process. Consequently, the discrepancy between predicted and actual corrosion depths can lead to errors in maintenance planning and even increase the risk of structural failure. Summary of the Invention

[0003] The purpose of this invention is to provide a carbon steel corrosion modeling method based on the finite element method, thereby solving the problems existing in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a carbon steel corrosion modeling method based on finite element method, comprising: S1, acquiring an initial parameter set including temperature, salinity, and material properties by collecting real-time data from environmental sensors, and determining a preliminary corrosion rate distribution; S2, generating a corrosion depth map of the metal structure using finite element simulation based on the initial parameter set, and obtaining the deviation value between the preliminary simulation results and the measured corrosion data; S3, if the deviation value exceeds a preset threshold, processing the parameter coupling relationship through a genetic algorithm to obtain an optimized parameter set for correcting the model input; S4, rerunning the finite element simulation using the optimized parameter set to determine the local variation trend of the corrosion rate in the complex structure, and obtaining an updated predicted depth map; S5, obtaining the residual deviation caused by dynamic changes by comparing the updated predicted depth map with historical measured data, and determining the environmental factor weights; S6, using a genetic algorithm iterative update adaptive mechanism based on the environmental factor weights to determine whether the parameter optimization has converged, and obtaining the stable model correction coefficients; S7, generating the final corrosion prediction distribution by applying the stable model correction coefficients to the finite element simulation, and determining the improved prediction accuracy level in the environment.

[0005] Preferably, step S1 includes collecting real-time data through environmental sensors to obtain an initial parameter set including temperature, salinity, and material properties, and determining a preliminary corrosion rate distribution; based on the preliminary corrosion rate distribution, obtaining corrosion trend indicators from historical data, which are obtained by comparing the differences between historical corrosion records and the current distribution, to obtain an extended parameter set; using the extended parameter set, integrating humidity influencing factors, which are collected in real-time from auxiliary sensors, to determine a corrosion acceleration threshold, which is determined by the product of humidity and salinity in the extended parameter set exceeding a preset threshold, to obtain an adjusted distribution; and for the adjusted distribution, integrating pH monitoring data, which are collected from environmental sensors, to determine the final corrosion rate distribution.

[0006] Preferably, step S2 includes: constructing a three-dimensional mesh model of the metal structure using the finite element method based on an initial parameter set; extracting node coordinates and element connections from the mesh model to obtain the corrosion simulation base structure; importing material properties and boundary conditions into the corrosion simulation base structure; applying loads to environmental factors; using a finite element solver with node coordinates and loads as inputs to calculate the corrosion rate at each node; solving the corrosion diffusion equation based on the finite element discretization equation to determine the corrosion depth distribution map; obtaining the corrosion depth distribution map and comparing it with measured corrosion data; if the deviation exceeds a preset threshold, adjusting the material property parameters by changing the elastic modulus and Poisson's ratio parameters to determine the optimized depth map; calculating the sum of squares of the differences between the simulated and measured values ​​based on the optimized depth map; using a preliminary deviation quantification index, iteratively updating the time step control, and performing local mesh refinement to obtain the final deviation value between the simulation result and the measured data.

[0007] Preferably, step S3 includes calculating the difference between the current model output and the expected value to obtain a deviation value, and determining whether the deviation value exceeds a preset threshold; if the deviation value exceeds the preset threshold, extracting coupling relationships from the parameter set, using a genetic algorithm to select the coupling relationships, and obtaining initial optimized parameters; based on the initial optimized parameters, using a fitness function to evaluate the parameter fit and determine the iteration direction of the parameter set; updating the parameter set through crossover and mutation operations to obtain converged optimized parameters; and correcting the model input for the optimized parameters to obtain an adjusted input set.

[0008] Preferably, step S4 includes obtaining an optimized parameter set from a preset parameter library, selecting an environmental corrosion factor for the structure, and determining preliminary simulation input data; rerunning the finite element simulation using the input data, calculating the corrosion rate distribution for the complex structure, and obtaining a local rate matrix; judging the changing trend in the local rate matrix, and if the trend exceeds a preset threshold, adjusting the structural mesh division and obtaining a trend correction value through mesh refinement; fusing the material corrosion resistance properties with the trend correction value to generate a corrosion depth prediction model, and obtaining an intermediate depth distribution map through attribute weighted summation; integrating stress influence factors based on the intermediate depth distribution map, and updating the predicted depth map through factor superposition calculation.

[0009] Preferably, step S5 includes: obtaining the residual deviation value caused by dynamic changes by comparing the updated predicted depth map with historical measured data, and determining the environmental factor weights that need further adjustment; processing the residual deviation value using a deviation comparison method, extracting the source of dynamic change from the historical measured data, and obtaining the factor influence degree, wherein the deviation comparison method obtains the average deviation index by calculating the sum of the absolute values ​​of the differences between each point in the predicted depth map and the corresponding points in the historical measured data, divided by the number of points; if the factor influence degree exceeds a preset threshold, then executing a weight adjustment strategy for the source of dynamic change to obtain the adjusted environmental factor weights, wherein the weight adjustment strategy updates the weight value by multiplying the factor influence degree by the current environmental factor weight and normalizing it; generating a data update stream based on the adjusted environmental factor weights, and judging the change trend analysis, wherein the data update stream is formed by continuously recording the sequence of adjusted environmental factor weights; extracting the measured verification step from the change trend analysis to complete the adjustment and optimization loop, wherein the measured verification step iterates the residual deviation value by matching the change trend analysis with a new round of historical measured data.

[0010] Preferably, step S6 includes obtaining the weights of environmental factors, using a genetic algorithm to iteratively update the adaptive mechanism, where the genetic algorithm uses environmental factors as the initial population, generates new solutions through crossover and mutation operations, determines whether the parameter optimization has converged, and obtains preliminary correction coefficients; for the preliminary correction coefficients, the weights are dynamically adjusted, and the environmental impact assessment results are determined from the mechanism's adaptation to changes, and the environmental impact assessment results are obtained by comparing the weight differences before and after the iteration.

[0011] Preferably, step S6 further includes obtaining an iterative convergence judgment value from the environmental impact assessment results, stabilizing the optimized parameters, and obtaining a coefficient correction basis, which is formed by weighted averaging of the judgment value; for the coefficient correction basis, integrating algorithm efficiency to determine the model correction coefficient extension value, which is expanded by multiplying by an efficiency factor; obtaining a stable processing output from the model correction coefficient extension value to obtain the stabilized model correction coefficient.

[0012] Preferably, step S7 includes obtaining the initial corrosion prediction distribution by multiplying the model correction coefficients by material property parameters after stabilization using the model correction coefficients; integrating environmental factors by superimposing environmental factors into the distribution grid to determine the simulation framework after environmental variable integration, and obtaining the adjusted distribution generation process; if the generation process meets the preset threshold, then simulation data calibration is used, parameters are adjusted by comparing with historical corrosion data, the corrosion rate calculation result is judged, and the rate value is obtained by dividing the result by the time interval to obtain the accuracy evaluation index.

[0013] Preferably, step S7 further includes applying a stable model to the evaluation index, obtaining the prediction accuracy level by applying an updated coefficient matrix, determining the final value of the accuracy improvement determination, obtaining the corrosion prediction distribution from the final value, mapping the distribution to the environmental grid, and generating the improved prediction accuracy level in the environment.

[0014] As can be seen from the above technical solution, the present invention has the following beneficial effects: This finite element method for carbon steel corrosion modeling collects initial parameters such as temperature, salinity, and material properties in real time. It then uses finite element simulation to generate a corrosion depth map and compares the deviation with measured data. If the deviation exceeds a threshold, the invention utilizes a genetic algorithm to optimize the parameter coupling relationship, generating a corrected parameter set. The finite element simulation is then rerun to obtain the local corrosion trend and an updated predicted depth map. By comparing historical measured data, the weights of environmental factors are dynamically adjusted, and the adaptive mechanism is iteratively optimized until the parameters converge, obtaining stable correction coefficients and ultimately generating a high-precision corrosion prediction distribution. This invention significantly improves the accuracy and reliability of corrosion prediction in complex environments by integrating environmental data, finite element simulation, and genetic algorithm optimization, providing a scientific basis for the maintenance of carbon steel structures. Attached Figure Description

[0015] Figure 1 This is a flowchart of the carbon steel corrosion modeling method based on the finite element method of the present 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] like Figure 1As shown, this invention provides a technical solution: a carbon steel corrosion modeling method based on finite element method, comprising: S1, acquiring an initial parameter set including temperature, salinity, and material properties by collecting real-time data from environmental sensors, and determining a preliminary corrosion rate distribution; S2, generating a corrosion depth map of the metal structure using finite element simulation based on the initial parameter set, and obtaining the deviation value between the preliminary simulation result and the measured corrosion data; S3, if the deviation value exceeds a preset threshold, processing the parameter coupling relationship through a genetic algorithm to obtain an optimized parameter set for correcting the model input; S4, rerunning the finite element simulation using the optimized parameter set to determine the local variation trend of the corrosion rate in the complex structure, and obtaining an updated predicted depth map; S5, obtaining the residual deviation caused by dynamic changes by comparing the updated predicted depth map with historical measured data, and determining the environmental factor weights; S6, using a genetic algorithm iterative update adaptive mechanism based on the environmental factor weights to determine whether the parameter optimization has converged, and obtaining the stable model correction coefficients; S7, generating the final corrosion prediction distribution by applying the stable model correction coefficients to the finite element simulation, and determining the improved prediction accuracy level in the environment.

[0018] This implementation method is based on multi-source sensor data fusion and finite element modeling technology. It utilizes environmental sensors to collect information such as temperature, salinity, and material properties to construct an initial parameter set, which is then used to determine the initial distribution of corrosion rates. A corrosion depth map of the metal structure is established through finite element simulation to quantify the deviation between the initial simulation and measured data. When the deviation exceeds a set threshold, a genetic algorithm is introduced to analyze the coupling relationship between parameters and optimize the input parameter set, thereby improving model accuracy. The optimized parameters are then reused in finite element analysis to simulate the local changes in corrosion rate of carbon steel under complex structures. Subsequently, by comparing with historical measured data, residual deviations caused by dynamic environmental factors are extracted, and the weight of each environmental variable is determined accordingly. Furthermore, the system iteratively updates the parameter adjustment mechanism using a genetic algorithm to achieve adaptive model correction until the parameter optimization process converges and outputs stable correction coefficients. Finally, these correction coefficients are applied to the finite element simulation to generate corrosion prediction results that conform to the changing trends under real-world conditions, improving prediction accuracy.

[0019] This implementation method enables dynamic optimization of the corrosion prediction model, significantly improving the accuracy and robustness of corrosion depth prediction. By introducing a genetic algorithm to efficiently optimize multi-parameter coupling relationships, it effectively overcomes the problems of high subjectivity in parameter setting and low simulation accuracy in traditional methods. Simultaneously, an environmental factor weight identification mechanism enhances the model's adaptability to changes in different environmental conditions, achieving a more intelligent and automated model calibration process. The combination of finite element simulation and historical data verification enhances the model's practicality in complex structural scenarios. Furthermore, this method possesses good scalability, allowing for flexible adjustment of parameter optimization strategies according to different working conditions, thus improving the model's versatility and reliability in various application fields.

[0020] S1 includes collecting real-time data through environmental sensors to obtain an initial parameter set containing temperature, salinity, and material properties, and determining a preliminary corrosion rate distribution; based on the preliminary corrosion rate distribution, obtaining corrosion trend indicators from historical data, which are obtained by comparing the differences between historical corrosion records and the current distribution, to obtain an extended parameter set; using the extended parameter set, integrating humidity influencing factors, which are collected in real-time from auxiliary sensors, to determine the corrosion acceleration threshold, which is determined by the product of humidity and salinity in the extended parameter set exceeding a preset threshold, to obtain an adjusted distribution; and for the adjusted distribution, integrating pH monitoring data, which are collected from environmental sensors, to determine the final corrosion rate distribution.

[0021] In one possible implementation, environmental sensors continuously collect temperature and salinity data in the environment where the object under test is located, and pH data is also collected in addition. Humidity data is continuously collected by auxiliary sensors, and material property values ​​are provided by material records. The sampling period is fixed at 1 minute, and all channels complete sampling at the same time. At the end of each sampling period, the system performs missing data checks and range checks on the temperature, salinity, humidity, and pH monitoring data. The range consists of the upper and lower limits in the device calibration file. If a missing data or data exceeds the range, the median of the valid values ​​of the same channel within the most recent 5 minutes is used for a one-time replacement. If replacement is not possible, that period is not included in the calculation. When the data is valid, the system generates an initial parameter set. The parameter set includes temperature, salinity, and material property values. The material property values ​​are derived from existing records of the carbon steel component under test and mapped to discrete levels via a lookup table before deployment. The system determines a preliminary corrosion rate distribution based on the initial parameter set. Specifically, at each location within the monitoring area, the temperature, salinity, and material property values ​​are read and their influence mapped in a fixed order: temperature influence mapping, salinity influence mapping, and material property influence mapping. Each mapping step uses a pre-calibrated monotonic relation table for lookup output, and only one lookup is performed without other calculations. The lookup results for all locations constitute the preliminary corrosion rate distribution. The system then obtains corrosion trend indicators from historical data, using the same bit depth as the current data. The system uses the same time granularity for storage, with the historical comparison period fixed at the most recent 30 days. At each location, the system calculates the absolute value of the difference between the average corrosion rate during the historical comparison period and the initial corrosion rate for the current period. This absolute value is then compared to the median of the daily variation range for that location during the historical comparison period. If the absolute value is greater than the median, the trend category is recorded as upward, and the trend strength is recorded as the absolute value. If the absolute value is not greater than the median, the trend category is recorded as stable, and the trend strength is recorded as the absolute value. The trend category and trend strength constitute the corrosion trend index. The system incorporates the corrosion trend index into the initial parameter set to form an extended parameter set, which is used for subsequent fusion of humidity influencing factors. The system uses auxiliary transmission... The humidity value provided by the sensor and the salinity value provided by the environmental sensor are multiplied and compared at the same location to determine whether the corrosion acceleration threshold has been triggered. The corrosion acceleration threshold is determined only by historical data and does not use empirical constants. The determination steps are as follows: search for time periods marked as showing obvious corrosion acceleration in the historical records of the same object, take the start time of these time periods as the positioning point, collect the product of humidity value and salinity value at each positioning point, sort them in ascending order, and take the value at the middle position of the sequence as the threshold candidate. Then, use whether the candidate causes more than half of the positions to continue to show the acceleration phenomenon in the following 7 days as the verification criterion. If the criterion is met, the candidate remains unchanged; if the criterion is not met, the candidate is moved up 3 positions in the sorted sequence to obtain the final threshold.The system calculates the product of humidity and salinity values ​​for each location during the current sampling period and compares it with a final threshold. If the product exceeds the threshold, the location is considered to be in an accelerated state, and a one-time amplification is applied to the initial corrosion rate value at that location. This amplification is the average value of the corrosion rate increments recorded at that location during the historical accelerated period. The system directly adds this average value to the initial corrosion rate value at that location to obtain the value for that location in the adjusted distribution. If the product does not exceed the final threshold, the value at that location remains unchanged and is directly written into the adjusted distribution. After obtaining the adjusted distribution, the system integrates pH monitoring data for each location to determine the final corrosion rate distribution. The integration step involves combining the pH monitoring data from the historical comparison period for each location... The pH monitoring data at the location is paired with residual deviation, defined as the average difference between the measured corrosion rate and the initial corrosion rate at that location under the same weather type and tidal period. The historical pH monitoring data for that location is then sorted in ascending order, and a moving average sequence of residual deviation is calculated using five adjacent samples as windows. The first inflection point where the residual deviation changes from negative to positive is identified as the acidic influence boundary, and the second inflection point where the residual deviation changes from positive to negative is identified as the alkaline influence boundary. If only one inflection point appears, that inflection point is used as the sole boundary, and the other side is considered stable. If no inflection point appears, the entire range is considered stable. The system reads the pH monitoring data for each location during the current sampling period. The data is then corrected once according to boundary rules. If the pH value is below the acidic influence boundary, the absolute average of the historical residual deviations within the same window is added to the adjusted value at that location. If the pH value is above the alkaline influence boundary, the absolute average of the historical residual deviations within the same window is subtracted from the value at that location. If the pH value is between the two boundaries or no boundary is detected, the value at that location remains unchanged. After all locations have been corrected, the set of values ​​for all locations constitutes the final corrosion rate distribution and is written into storage at the end of the sampling cycle for subsequent steps. All parameters involved in the above process and their determination methods are as follows: the initial parameter set consists of temperature values, salinity values, and material property values, and is directly converted to standard parameters after quality checks at the end of each sampling cycle. The initial corrosion rate distribution is directly composed of the results of a three-step lookup table mapping at each location; the historical corrosion record is a time series of the measured corrosion rate or corrosion depth of the same object at each location in the past period, and is saved with the same time granularity and spatial location as the current one; the corrosion trend index is composed of trend category and trend intensity and is output according to the comparison rule of the median of the daily variation range of the most recent 30 days; the extended parameter set is the union of the initial parameter set and the corrosion trend index; the humidity influencing factor is the humidity value of the auxiliary sensor in the current sampling period; the corrosion acceleration threshold is a single value determined based on the median of the product sequence of historical acceleration time periods and the verification rule of maintaining acceleration for most of the subsequent 7 days and the correction rule of shifting up 3 positions;The adjusted distribution is a set of values ​​formed by summing and correcting the values ​​at each location according to whether the corrosion acceleration threshold is exceeded; the pH monitoring data are real-time values ​​collected by environmental sensors; the acidic and alkaline influence boundaries are determined by two inflection points in the moving average sequence of historically paired pH values ​​and residual deviations; the final corrosion rate distribution is a set formed by adding and subtracting the adjusted distribution according to the boundary rules.

[0022] The corrosion acceleration threshold is determined by multiplying the humidity and salinity values ​​and using the median of historical acceleration time periods, shifting upwards by three positions if verification fails. This method stably characterizes the triggering conditions for accelerated corrosion within the historical data range of the same object, maintaining consistency and traceability in the judgment: using the product as a single trigger value reflects the synergistic amplification effect of humidity and salinity, thus avoiding misjudgments caused by high values ​​of a single factor; sampling only the starting moments of time periods recorded as showing significant acceleration in historical data ensures that the samples correspond to the triggering boundary of acceleration rather than peak values ​​during the acceleration process, reducing the risk of the threshold being driven up by excessively high values ​​from the source; and sorting the products of these samples from smallest to largest and using the median as a candidate threshold.

[0023] S2 includes constructing a three-dimensional mesh model of the metal structure using the finite element method based on an initial parameter set; extracting node coordinates and element connections from the mesh model to obtain the basic structure for corrosion simulation; importing material properties and boundary conditions into the corrosion simulation basic structure; applying loads to environmental factors; using a finite element solver with node coordinates and loads as inputs to calculate the corrosion rate at each node; solving the corrosion diffusion equation based on the finite element discretization equation to determine the corrosion depth distribution map; obtaining the corrosion depth distribution map and comparing it with measured corrosion data; if the deviation exceeds a preset threshold, adjusting the material property parameters by changing the elastic modulus and Poisson's ratio parameters to determine the optimized depth map; calculating the sum of squares of the differences between the simulated and measured values ​​based on the optimized depth map; using a preliminary deviation quantification index, iteratively updating the time step control, and performing local mesh refinement processing to obtain the final deviation value between the simulation result and the measured data.

[0024] In one possible implementation, a three-dimensional mesh model is first established using the three-dimensional geometry of the metal structure under test as input, employing the finite element method. Before deployment, the edge length of the mesh elements is fixed at one-tenth of the minimum geometric feature size and no greater than one-fifth of the thinnest wall thickness. The mesh transition ratio is fixed at 1.2. Mesh quality is checked using element shape indices, and local re-meshing is triggered directly when the mesh falls below a set lower limit. The system extracts node coordinates from the generated mesh, connects them to the elements, and saves them as the corrosion simulation foundation structure. This foundation structure serves as the sole data carrier for subsequent solutions. The system then imports material property parameters and boundary conditions and applies environmental factor loads, where the material property parameters... The system includes only two parameters: elastic modulus and Poisson's ratio. The initial values ​​for both are determined before deployment by taking the median of the incoming inspection records for the same material batch and fixing it as the initial material property. The outer surface of the boundary conditions is the exposed surface, with equivalent corrosion flux or equivalent corrosion rate boundary values ​​set. The inner non-exposed surface is the flux-free boundary, and the component's symmetry plane is the symmetry boundary. Environmental factor loads are determined by a one-to-one lookup table rule established before deployment, based on temperature, salinity, humidity, and pH values, and are used as equivalent boundary values ​​on the exposed surface, remaining constant throughout the time step. The finite element solver uses nodal coordinates, element connections, material properties, boundary conditions, and environmental factor loads as the sole inputs. The system progresses from an initial time step, fixed at 1 hour, with a maximum time step of 4 hours and a minimum time step of 15 minutes. At the end of each time step, the corrosion rate of all nodes is output, and the corrosion depth distribution map is obtained by accumulating the time. The time accumulation adopts a stepwise summation method, adding the corrosion rate of the current time step to the historical accumulated depth to update the current depth. After obtaining the corrosion depth distribution map of the current round, the system performs point-by-point registration with spatial locations and time nodes consistent with the field measured corrosion data. For each registration point, the difference between the simulated value and the measured value is calculated, and the sum of the squared differences of all registration points is used as the partial difference. The differential index is calculated by taking the maximum value of the absolute deviation and the maximum value of the relative deviation as the criteria for exceeding the limit. The preset threshold is determined by setting it offline using the historical comparison dataset of the same object. The steps are as follows: select 13 sets of actual and simulated comparison records from the historical data of 90 consecutive days, calculate the sum of squares of the difference for each set, sort them from smallest to largest, and take the value in the middle position as the threshold candidate. Then check whether more than half of the comparison records in each solution of the candidate in the last 30 days meet the threshold. If it meets the threshold, the candidate is used as the final threshold. If it does not meet the threshold, the candidate is moved up 3 positions in the sorting sequence as the final threshold and fixed in the deployment document.If the maximum absolute deviation or the maximum relative deviation exceeds the final threshold in the current comparison round, the system executes a material property parameter adjustment process. Material property parameter adjustment is strictly limited to changing the values ​​of the elastic modulus and Poisson's ratio, and is performed only once per round. The adjustment range is determined by retrospectively analyzing the decrease in the sum of squared differences from the last three comparison rounds. If the decrease is less than 10%, the elastic modulus is decreased by 10% and the Poisson's ratio is increased by 5%. If the decrease exceeds 30%, the material parameters are kept unchanged to avoid overcorrection. If the decrease falls between these ranges, only the elastic modulus is decreased by 5%, and the Poisson's ratio remains unchanged. Immediately after the parameter update, the optimized depth is re-solved using the same mesh and boundary values. The system recalculates the sum of squared differences after obtaining the optimized depth map and uses its ratio to the sum of squared differences from the previous round as the sole basis for adaptive time step control. If the current sum of squared differences decreases by less than 10% compared to the previous round, the next time step is reduced to half of the current sum but not less than the minimum time step. If the decrease exceeds 30%, the next time step is increased to twice the current sum but not greater than the maximum time step. If it falls between these values, it remains unchanged. Simultaneously, the system performs local mesh refinement on the local regions that contribute the most to the sum of squared differences. The contribution is calculated by averaging the absolute values ​​of the differences between the nodes within each cell, sorting them from largest to smallest, and selecting the top 10% of cells as local refinement targets. The target side length of the element is halved and a one-time local re-mesh is performed. After re-meshment, the historical cumulative depth and current corrosion rate are interpolated to the nodes of the new mesh through a node mapping table to maintain the continuity of the field. The system iterates alternately according to the aforementioned three rules of material parameter adjustment, time step control, and local refinement until the change of the sum of squares of the differences in three consecutive rounds is less than 10%, which is considered to have reached stability. At this time, the deviation value between the final simulation result and the measured data is output and the current solution loop ends. All parameters in the above process and their determination methods are as follows: the element side length, mesh transition ratio, and quality lower limit of the 3D mesh are fixed at one-tenth before deployment, 1.2 and the project quality standard lower limit, and the node coordinates and element connections are directly derived from the mesh. The generation results are as follows: initial material property parameters are determined by the median of batch inspection records; boundary conditions and environmental factor loads are uniquely determined by lookup rules before deployment; the initial, maximum, and minimum time steps are fixed at 1 hour, 4 hours, and 15 minutes, respectively; preset thresholds are determined once using the 90-day historical median and a 30-day majority pass and shift-up rule, and remain unchanged during operation; the adjustment range of material property parameters in each round is determined by the decrease in the sum of squared differences of the most recent three rounds and is executed only once per round; time step control depends only on the ratio of the sum of squared differences of adjacent rounds and scales at a fixed ratio; local densification only applies to the top 10% of units, and the densification ratio is fixed at 1 / 2.

[0025] In this implementation, the preset threshold is determined by selecting 13 sets of simulated and measured comparison records from 90 consecutive days of historical data, each weekend, calculating the sum of squared differences, taking the median of the sorted values, and then verifying the results by the majority of the most recent 30 days passing the verification; if a result fails, the threshold is shifted up 3 places in the sorted sequence. This method is chosen because it provides a stable, verifiable, and directly relevant approach to the actual purpose of this step in determining when material property parameter adjustments need to be triggered: a 90-day sampling period covers fluctuations in environmental factors such as temperature, salinity, humidity, and pH within a distinct climatic cycle, ensuring sufficient temporal representativeness of the threshold; selecting 13 sets of records each weekend reduces the stacking effect of highly correlated data within the same week, preventing the threshold from being biased by short-term continuous anomalies, while ensuring the sample size is not too large and avoids introducing redundant calculations; and using the sum of squared differences... Using the median as the sole comparison metric assigns higher weight to large local deviations, aligning with engineering concerns regarding high-gradient corrosion regions in 3D meshes, thus making the threshold more sensitive to critical mismatches. Taking the median of the sorted values ​​naturally suppresses the impact of extreme values ​​on the threshold, ensuring its representativeness of the overall data distribution and its robustness to noise. Verifying the threshold based on the majority of successful tests over the last 30 days allows for testing its usability from the perspective of the current season and operating conditions, ensuring that when the difference exceeds the threshold, it corresponds to a stable deviation requiring material property parameter adjustments, rather than an occasional measurement disturbance. Moving the threshold up three positions when verification fails is a conservative correction strategy with a fixed step size, which can quickly increase the trigger threshold to reduce false alarms without changing the sample set and computational scope, avoiding solution oscillations and convergence delays caused by frequent parameter adjustments.

[0026] S3 includes calculating the deviation value by comparing the current model output with the expected value, and determining whether the deviation value exceeds a preset threshold. If the deviation value exceeds the preset threshold, coupling relationships are extracted from the parameter set, and a genetic algorithm is used to select the coupling relationships to obtain initial optimized parameters. Based on the initial optimized parameters, a fitness function is used to evaluate the degree of parameter fit and determine the iteration direction of the parameter set. The parameter set is updated through crossover and mutation operations to obtain converged optimized parameters. For the optimized parameters, the model input is corrected to obtain the adjusted input set.

[0027] In one possible implementation, the corrosion depth distribution obtained from S2 is first used as the current model output, and the measured corrosion data at the same spatial location and time point is used as the expected value. The absolute values ​​of the two are subtracted point by point and then summed over the entire domain to obtain the deviation value. At the same time, a preset threshold fixed before deployment is read and a single comparison is performed. The preset threshold is determined by sampling 13 sets of simulated and measured paired data at the end of each week in the historical records of 90 consecutive days, calculating the corresponding deviation value for each, sorting the 13 deviation values ​​in ascending order, and taking the value in the middle position as the threshold candidate. Then, it is checked whether each solution in the most recent 30 days exceeds one. Half of the paired records are below the candidate. If the condition is met, the candidate is used as the final threshold; otherwise, the candidate is shifted up 3 positions in the sorting sequence as the final threshold and fixed in the deployment document. When the current deviation value is not greater than the preset threshold, the optimization loop terminates and the S2 result is directly passed to the subsequent steps. When the deviation value is greater than the preset threshold, the parameter set coupling relationship extraction and genetic algorithm optimization process is started. The parameter set consists of entries that are allowed to participate in the adjustment and have a direct impact on the model output. These entries are fixed item by item according to the lookup table range and step size before deployment, including temperature weight, salinity weight, humidity weight, acid-base weight, and material property correction. The coefficients are categorized into five types, each with a set minimum, maximum, and step size that are not exceeded during runtime. Coupling relationships are extracted based on the solution records of the same object over the past 30 days. The system statistically analyzes the synchronous change direction and magnitude of other entries when any entry changes its step size. When two or more entries change in the same direction in most solution records and cause a synchronous decrease in the deviation value, the system groups these entries into a co-directional coupling group. When one entry increases while another decreases, causing a decrease in the deviation value, the system groups these two entries into an anti-coupling group. An upper limit for the relative change ratio of each group is set at one step size per iteration, thus forming a fixed... A fixed and read-only coupling relationship table is used. The genetic algorithm uses a candidate parameter set with a population size of 50 as the initial population. The first individual is the current parameter set, and the remaining 49 individuals are selected at equal intervals between the minimum and maximum values ​​of each entry. During generation, the coupling relationship is checked one by one. Any individual that violates the coupling relationship immediately backs to the boundary value that satisfies the coupling relationship. The selection operation adopts a comparison method with a fixed size of 3. That is, each time, 3 individuals are randomly selected from the population with equal probability, and their deviation values ​​are compared. The one with the smallest deviation value enters the next generation. This process is repeated until a new population with a size of 50 is formed. The selection result is the initial optimization parameters.The system then evaluates the initial optimization parameters using a fitness function. The fitness function is constructed by ranking all individuals in descending order of their deviation values ​​and mapping the rankings to 100 equal levels, with higher rankings resulting in higher levels. The system reads the top 20 individuals and calculates the median trend for each of the five categories. If the majority of values ​​for a given item in the top 20 individuals are higher than the corresponding value in the current parameter set, the iteration direction for that item is determined to be one step upward. If the majority of values ​​are lower than the corresponding value in the current parameter set, the iteration direction is determined to be one step downward. If the values ​​are evenly split, the iteration direction remains unchanged. The system remains unchanged. After clarifying the iteration direction, the system performs crossover and mutation operations to update the parameter set. The crossover operation uses the adjacent pairing method, that is, the first individual after fitness sorting is paired with the second individual, the third individual with the fourth individual, and so on. For each pair, two offspring are generated. The generation method is that the first half of the entry is taken from the pair with the higher fitness, and the second half of the entry is taken from the pair with the lower fitness. Immediately after generation, a consistency check is performed according to the coupling relationship table, and any entry that exceeds the coupling limit is backed to the nearest boundary that meets the limit. The mutation operation is performed uniformly after the crossover of the entire generation is completed, and the mutation probability is... The mutation rate is 5%, with a step size of one. In each generation, the system alternately selects coupled core entries and coupled dependent entries for single-point mutation to avoid synchronous mutation of entries in the same group within the same generation. After all offspring are generated, the system merges the parent and offspring generations, recalculates fitness based on the deviation value, and retains the top 50 individuals as the new generation population. The system uses the change in the optimal deviation value over five consecutive generations as the convergence criterion. Convergence is determined when the change in the optimal deviation value over five consecutive generations is less than 1%, or when the cumulative number of generations reaches 100, the current best individual is taken as the convergence result and the optimization is terminated. The converged optimization parameters are the values ​​of the five categories of items for the first ranked individual. The system writes these values ​​into the model input item by item to generate the adjusted input set. The writing rules are as follows: the material property correction coefficient is applied to the nominal values ​​of elastic modulus and Poisson's ratio used in S2 in a multiplicative manner; the temperature weight, salinity weight, humidity weight, and acid-base weight are applied to the lookup table entries in S2 used to generate equivalent boundary values ​​in an additive manner. All writing is required to be within the minimum and maximum values ​​of the entries and the step size must be strictly aligned. After successful writing, the system ends S3 and submits the adjusted input set to S4 for use.The parameters involved in the above process and their determination methods are as follows: the deviation value is directly obtained in the current calculation by summing the absolute differences between the current model output and the expected value point by point; the preset threshold is determined once before deployment and remains unchanged during operation, based on the rule of the 90-day historical median, passing the majority test over 30 days, and shifting up by 3 places if necessary; the parameter set contains five categories of entries, and the range and step size of each entry are fixed by table lookup before deployment; the coupling relationship table is obtained by statistical analysis of the solution records of the most recent 30 days and is read-only during operation; the population size is 50, the comparison method size is 3, the fitness level is 100, the number of top-ranked statistics is 20, the mutation probability is 5%, the mutation amplitude is one step, the convergence window is 5 generations, the convergence threshold is 1%, and the maximum number of generations is 100, all of which are set before deployment and remain unchanged during operation.

[0028] In this implementation, the preset threshold is determined by calculating the deviation value and taking the median of the sorted data by sampling 13 pairs of paired data from the last 90 days of historical records at the end of each week. This is then verified by the majority of the most recent 30 days; if a pair fails, the sorted sequence is shifted up 3 places. The reason for this is that the threshold must only be triggered when the deviation reaches a level requiring the genetic algorithm to be activated, and the triggering criteria must be representative, robust, and verifiable. A 90-day sampling period covers fluctuations in environmental factors such as temperature, salinity, humidity, and pH within a clear climatic cycle, allowing the threshold to reflect long-term background rather than short-term noise. Sampling 13 pairs per week compresses strongly correlated samples within the same week into a single observation, avoiding high-frequency autocorrelation exaggeration. Large deviations are avoided while ensuring sufficient sample size and controllable computational load. Using the median of the deviation values ​​as a candidate threshold naturally suppresses the pull of extreme values ​​on the threshold, making the threshold representative of the overall distribution and robust to outliers. The majority of the candidate thresholds have been validated in the last 30 days, which can verify their usability under the current season and operating conditions. This ensures that if the threshold is exceeded, it means that there is a stable mismatch between the model and the actual measurement that requires parameter optimization, rather than a one-time measurement disturbance. When the validation fails, the conservative correction strategy of shifting the threshold order up by 3 positions with a fixed step size can increase the trigger threshold without changing the sample set and calculation method, reducing the risk of frequent optimization and solution oscillations caused by false alarms.

[0029] S4 includes obtaining an optimized parameter set from a preset parameter library, selecting environmental corrosion factors for the structure, and determining preliminary simulation input data; rerunning the finite element simulation using the input data, calculating the corrosion rate distribution for complex structures, and obtaining a local rate matrix; judging the changing trend in the local rate matrix, and if the trend exceeds a preset threshold, adjusting the structural mesh division and obtaining a trend correction value through mesh refinement; integrating the material corrosion resistance properties with the trend correction value to generate a corrosion depth prediction model, and obtaining an intermediate depth distribution map through attribute weighted summation; integrating stress influence factors based on the intermediate depth distribution map, and updating the predicted depth map through factor overlay calculation.

[0030] In one possible implementation, the system first reads the optimized parameter set from a preset parameter library according to the structure number and the latest timestamp. The optimized parameter set is a unique copy of the S3 convergence result and includes five categories of items: temperature weight, salinity weight, humidity weight, acid-base weight, and material corrosion resistance property correction coefficient. Each item has a minimum value, maximum value, and step size set before deployment and does not exceed these limits during operation. The system then selects an environmental corrosion factor for the current structure. The environmental corrosion factor consists of four items: temperature value, salinity value, humidity value, and pH value. The system reads these four values ​​from the monitoring records at the same time point and checks them one by one through a table lookup and the optimized parameter set. A one-to-one mapping is established for the corresponding weights, generating equivalent boundary inputs for the exposed surface and equivalent volume inputs for the internal region. The lookup table rules are fixed before deployment and remain unchanged during operation. The system combines the above inputs with the 3D mesh and nominal material parameters generated in stage S2 to form the initial simulation input data and performs consistency verification. The verification includes whether the weights and values ​​are between the minimum and maximum values, whether the step size is aligned, and whether the mesh and material names are consistent with the structure number. After the verification is passed, the finite element solver is started to rerun the simulation. The time step adopts the initial time step, maximum time step, and minimum time step set in stage S2 and follows a fixed convergence criterion. The system then proceeds by outputting the corrosion rate at the unit or node scale at each time step and assembling it into a local rate matrix according to spatial and temporal indices. Each row of the matrix corresponds to a spatial location, each column corresponds to a time step, and each element is the corrosion rate value at that location at that time step. After obtaining the local rate matrix, the system calculates the trend of change position by position. The trend is measured by the cumulative sum of the absolute values ​​of the rate differences between adjacent time steps within a fixed window, plus the absolute values ​​of the rate differences at the beginning and end of the window. The window length is fixed at 5 before deployment and does not change during operation. The system uses this metric as the single trend value. The trend preset threshold is set at the same value before deployment. The historical local rate matrix of an object over the past 90 days is used as a sample for calculation. The specific process is as follows: on each historical day, the trend values ​​of all locations are calculated and summarized with the same window length. The median is taken as the candidate threshold and sorted from smallest to largest. The records of the most recent 30 days are then used for verification. The verification rule is that during the period marked by the field as having obvious acceleration, the trend values ​​of more than half of the known acceleration locations are greater than the candidate and the trend values ​​of more than half of the non-acceleration locations are not greater than the candidate. If this double majority rule is satisfied, the candidate is used as the final threshold. If it is not satisfied, the sorting sequence is shifted up by 3 positions to obtain the final threshold and it is fixed in the deployment document.The system performs a single comparison between the trend value and the final threshold at each location in the current loop. If the trend value is not greater than the final threshold, the existing mesh and rate value are retained at that location. If the trend value is greater than the final threshold, mesh refinement is performed on that location and its adjacent cells. The mesh refinement rule is to halve the target cell size and fix the transition ratio at 1.2. After refinement, the system performs rapid recalculation on the local subdomain using the same time step and boundary input as the global domain and outputs the refined local corrosion rate. The system uses the absolute value of the rate difference at the same location and time step before and after refinement as the trend correction value and writes this value to the corresponding location. The system then performs material corrosion resistance property fusion on all locations. The depth is determined by a combination of nominal material parameters and material corrosion resistance correction coefficients. This combination rule is fixed as a single weighted rule before deployment. The system integrates the trend correction value and material corrosion resistance properties in one step and outputs the local depth increment at that location. Simultaneously, the cumulative result of the equivalent boundary input and equivalent volume input is used as the base depth term. The system obtains the intermediate depth distribution map by summing the base depth term and the local increment term using attribute weights. The weights are allocated by temperature, salinity, humidity, and acid / alkali weights from the optimization parameter set according to a fixed ratio given by a lookup table before deployment. This ratio is set once before deployment and remains unchanged during operation. The system integrates stress influence factors based on the intermediate depth distribution map. The stress influence factor is the equivalent stress quantity of the structure under a given load condition. The numerical source is the mechanical field solution result of stage S2 or the nominal stress field record consistent with S2. The system converts the stress influence factor into depth increment or decrement according to a fixed ratio superposition rule and adds it to the intermediate depth distribution map position by position to obtain the updated predicted depth map as the final output of S4. The parameter definition and determination method in the above process is as follows: the set of optimization parameters is given by the convergence result of S3 and the minimum value, maximum value and step size boundary are retained for consistency verification. The environmental corrosion factor is the measured value of temperature, salinity, humidity and pH at the same time point and is formed by equivalent input by looking up the table and mapping it one by one with the weight. The local rate matrix is ​​a table of rate values ​​organized by spatial location and time step; the trend measure is the sum of the cumulative differences between adjacent values ​​and the cumulative differences between the first and last values ​​within a fixed window of length 5; the preset trend threshold is the median of the past 90 days, validated by a double majority vote over 30 days, and shifted up by 3 places if necessary; the target size for mesh refinement is half the original size with a transition ratio of 1.2; the trend correction value is the absolute value of the rate difference before and after refinement; the material corrosion resistance property is a single corrosion resistance value resulting from the combination of nominal parameters and correction coefficients; the weights for the weighted summation of properties are fixed proportions of temperature weight, salinity weight, humidity weight, and acid / alkali weight; the stress influence factor is an equivalent stress quantity superimposed at a fixed proportion as a depth increment or decrement.

[0031] The trend preset threshold is determined by using the historical local rate matrix of the past 90 days as a sample, calculating the trend value with a fixed window length of 5, selecting the median of the sorted values ​​from all samples as candidates, and then verifying it using the double majority rule of the most recent 30 days. If the verification fails, the value is shifted up 3 places in the sorted sequence to solidify it as the final threshold. The reason for this is that the threshold must only be triggered when the local rate change has formed a stable accelerating trend and grid refinement and local recalculation are indeed necessary. At the same time, it is necessary to avoid false alarms and frequent refinement caused by short-term noise or occasional measurement errors. Using 90 days as the sample period can cover the fluctuations of temperature, salinity, humidity, and pH within a complete climatic period, so that the threshold has long-term representativeness and does not depend on the anomalies of a few days. The fixed window length of 5 spans both historical and current calculations to ensure the caliber of the trend measurement. Consistency is maintained to avoid systematic biases introduced by changes in caliber. Calculating trend values ​​across all sample locations and dates and using the median as a candidate naturally suppresses the impact of extreme values ​​on the threshold, improving robustness to outliers and occasional spikes. Double majority verification is performed over the most recent 30 days, requiring more than half of the known acceleration locations to be higher than the candidate and more than half of the non-acceleration locations not to be higher than the candidate. This simultaneously controls false positives and false negatives, ensuring the threshold is sufficiently sensitive to areas that truly require refinement while maintaining sufficient specificity for areas that do not require refinement. When a candidate fails verification, shifting it up by 3 positions is a conservative correction with a fixed step size. This increases the trigger threshold without changing the sample set and metric, rapidly reducing the false positive rate and avoiding oscillations caused by the grid switching between thickening and thinning during iteration.

[0032] S5 includes: comparing the updated predicted depth map with historical measured data to obtain the residual deviation value caused by dynamic changes, and determining the environmental factor weights that need further adjustment; processing the residual deviation value using a deviation comparison method, extracting the source of dynamic change from the historical measured data to obtain the factor influence degree, wherein the deviation comparison method calculates the average deviation index by dividing the sum of the absolute values ​​of the differences between each point in the predicted depth map and the corresponding point in the historical measured data by the number of points; if the factor influence degree exceeds a preset threshold, executing a weight adjustment strategy for the source of dynamic change to obtain the adjusted environmental factor weights, wherein the weight adjustment strategy updates the weight value by multiplying the factor influence degree by the current environmental factor weight and normalizing it; generating a data update stream based on the adjusted environmental factor weights, judging the change trend analysis, wherein the data update stream is formed by continuously recording the sequence of adjusted environmental factor weights; extracting a measured verification step from the change trend analysis to complete the adjustment and optimization loop, wherein the measured verification step iterates the residual deviation value by matching the change trend analysis with a new round of historical measured data.

[0033] In one possible implementation, the system, based on the working principle of claim 6, performs all calculations in one complete loop and provides a clear determination method and step-by-step implementation process for each parameter and threshold, and the whole is described in a single segment: The system first reads the updated predicted depth map output by S4 and retrieves historical measured data at the same spatial grid and the same time point, calculates the absolute difference between the two points point by point and records it as the position-level residual deviation value, and sums all position-level residual deviation values ​​and divides them by the total number of positions participating in the comparison to obtain the average deviation index; the system uses the average deviation index as a metric to process the residual deviation using the deviation comparison method and retrieves the data from historical measured data. The system extracts dynamic sources of change from the data, specifically by segmenting the environmental sequences of temperature, salinity, humidity, and pH into time segments. The segmentation rule is that during continuous sampling, a new segment is initiated when the absolute value of the difference between two adjacent records is not less than the minimum step size set before deployment; otherwise, the current segment continues. The minimum step size is fixed at three times the resolution of each sensor before deployment and remains unchanged during operation. For each segment of each environmental factor, the system calculates the average of all location-level residual deviations within that segment, recording it as the residual level of that segment, and calculates the monotonic change direction of the factor's value within the segment. If the factor's value changes monotonically between adjacent records... If a segment shows a continuous unidirectional change and its residual level increases by at least 10% compared to the previous segment, then that segment is marked as a candidate for the dynamic change source of that factor. The system takes the median residual level from all candidates for the same factor as its factor influence. If no candidates are available, the median residual level of the factor across the entire time domain is used as the factor influence. The system reads the preset thresholds fixed before deployment and compares the factor influence of each of the four factors one by one. The preset thresholds are determined by sampling 13 sets of comparison samples from the past 90 days of historical records of the same object at the end of each week. For each set of samples, the factor influence of the four factors is calculated according to the aforementioned process, and the four factors are compared. The largest value is used as the representative value of the sample group. The 13 representative values ​​are sorted from smallest to largest, and the median is used as the candidate threshold. Then, the verification samples of the most recent 30 days are used for verification. The verification rule is that in the samples where significant environmental changes are known, at least half of the representative values ​​are higher than the candidate, and in the samples where the environmental changes are known, at least half of the representative values ​​are not higher than the candidate. If the conditions are met, the candidate is used as the final threshold. If the conditions are not met, the value is shifted up 3 places in the sorting sequence to obtain the final threshold and is fixed in the deployment document. When the influence of any factor is greater than the final threshold, the weight adjustment strategy is entered. Otherwise, the weights of the existing environmental factors remain unchanged and the current cycle ends.The specific steps of the weight adjustment strategy are as follows: Calculate temporary weight values ​​for each of the four factors. Each temporary weight value is equal to the product of the factor's influence and its current environmental factor weight. Then, sum the four temporary weight values ​​to obtain a normalization factor. Next, divide each temporary weight value by the normalization factor to obtain the adjusted environmental factor weight. The sum of the four weights must strictly equal 1, ensuring that each falls within the minimum and maximum values ​​set before deployment and is aligned with the step size. If an out-of-bounds error occurs, the weight is rolled back to the nearest boundary value and normalized again by summing the new four weights. The system writes the quaternary weight vector obtained from this update into the data update stream in chronological order. The data update stream is a weight sequence indexed by timestamps. Execution is performed when the length of the data update stream reaches the fixed window length. Trend analysis is performed with a fixed window length of 5 before deployment and maintained during operation. The trend analysis uses a weighted sum of two metrics as a single trend metric: the number of monotonic segments for each factor's weight within the window and the cumulative sum of weight differences between adjacent time points. The weight coefficients are fixed at equal proportions before deployment. If the trend metric does not exceed the pre-deployment stability threshold in three consecutive evaluations, the weights are considered stable. This stability threshold is determined before deployment based on the median trend metric during historical stable periods, shifted upwards by three ranking positions, and remains unchanged during operation. After stability is determined, the system enters the experimental verification step. This step uses the latest stable weight vector to drive S4 to regenerate and update the predicted depth map, and then it is compared point-by-point with the subsequent new round of historical experimental data. In comparison, the average deviation index and the location-level residual deviation value are recalculated. If the new average deviation index decreases by at least 10% relative to the initial value when entering S5, the weight of this round is accepted and the weight vector is marked as a valid result. If the decrease is less than 10% or an increase occurs, the weight vector is rolled back to the previous set in the data update stream, and the window length of the trend analysis is shortened to half of the original value. Then, the trend judgment and actual measurement verification are repeated. The maximum number of rollbacks is 3. If the condition of a decrease of at least 10% is not met even after exceeding the upper limit, the preset threshold is increased by a fixed step and the weight update of this round ends. The fixed step is set to move the threshold sorting sequence up by 1 position before deployment. All parameters involved in this step and their determination methods are as follows: The location-level residual deviation value is the absolute difference between the predicted and the measured point by point. The average deviation index is obtained directly in the current round by summing the absolute differences and dividing by the number of points. The minimum change step size is fixed before deployment as 3 times the resolution of each sensor. The factor influence degree is obtained directly in the current round as the median of the residual level of the candidate segment of the dynamic change source of each factor or the median of the residual level of the entire domain. The preset threshold is determined by shifting up 3 places when it fails the 90-day historical sample review. The environmental factor weights are quaternary weight vectors, and their respective ranges and step sizes are fixed before deployment. The data update stream is a weighted sequence recorded by time. The window length for trend analysis is 5, and the cumulative weight difference is weighted equally with the number of monotonic segments. The stability threshold is determined by shifting up 3 places from the median of the historical stable period. The upper limit of backtracking is 3 times, and the threshold step size is shifted up 1 place.

[0034] The reason why the threshold in this embodiment adopts the determined path of calculating the representative value of the factor influence based on the historical samples of the past 90 days, taking the median of the ranking as a candidate, and then verifying it using the double majority rule of the most recent 30 days, and then fixing it by moving it up 3 places in the ranking sequence if it fails, is because this path can simultaneously meet the engineering requirements of representativeness, robustness, verifiability, and verifiability: selecting 90 days as the modeling sample period can cover the typical fluctuations of temperature, salinity, humidity, and pH within a complete climatic phase, so that the candidate threshold reflects the long-term background rather than short-term randomness; drawing 13 independent samples at the end of each week can weaken the pull of highly correlated data within the same week on the threshold, ensuring that the samples are independent and the amount of computation is controllable; and using the median of the ranking rather than the extreme value or the average. As a candidate, it can naturally suppress the influence of outliers and spike noise, and improve robustness to anomalies. The "double majority" verification is performed on the most recent 30 days, which requires that at least half of the representative values ​​in the samples with known significant environmental changes are higher than the candidate and at least half of the values ​​in the known stable samples are not higher than the candidate. This can control both false negatives and false positives, so that the threshold is only triggered to adjust the weight when dynamic changes have a real dominant effect on the residual bias, thus avoiding the oscillation of the data update stream caused by frequent ineffective adjustments from the source. When the candidate fails the verification, a fixed step size correction of 3 positions up in the sorting sequence is adopted. This can increase the trigger threshold without changing the sample set and calculation method, quickly reduce false positives, and provide a monotonically convergent correction rule for subsequent loops.

[0035] S6 includes obtaining the weights of environmental factors, employing a genetic algorithm for iterative updates and an adaptive mechanism. The genetic algorithm uses environmental factors as the initial population, iteratively generating new solutions through crossover and mutation operations, determining whether parameter optimization has converged, and obtaining preliminary correction coefficients. For the preliminary correction coefficients, the weights are dynamically adjusted, and the environmental impact assessment results are determined from the mechanism's adaptation to changes. The environmental impact assessment results are obtained by comparing the weight differences before and after iteration. From the environmental impact assessment results, an iterative convergence judgment value is obtained, and the optimization parameters are stabilized to obtain the coefficient correction basis. The coefficient correction basis is formed by weighted averaging of the judgment value. For the coefficient correction basis, the algorithm efficiency is integrated to determine the extended values ​​of the model correction coefficients. The extended values ​​of the model correction coefficients are expanded by multiplying by an efficiency factor. From the extended values ​​of the model correction coefficients, a stable processing output is obtained, resulting in the stabilized model correction coefficients.

[0036] In one possible implementation, the environmental factor weights output by S5 are first read as input. The weight order is fixed as four items: temperature, salinity, humidity, and pH. The sum of these four items equals 1. The upper and lower bounds and step size of each item are set to fixed values ​​before deployment and remain unchanged throughout the process. The system activates the adaptive mechanism of the genetic algorithm, forming the first individual by combining the current four weights in a predetermined order. The remaining individuals take values ​​at equal intervals within their respective upper and lower bounds with step sizes, and are normalized once during generation to make the sum of the four items equal to 1. If any value exceeds the bounds, it directly backtracks to the nearest boundary and then normalizes again. The initial population size is fixed at 50. The system defines fitness as the updated prediction depth obtained by S4 quickly recalculating the current individual weights over the most recent three time slices. The graph shows the percentage decrease in the average deviation index relative to the reference predicted depth map before entering S6; the larger the percentage, the higher the fitness. Each generation iteration sequentially performs selection, crossover, mutation, and fitness evaluation. Selection uses a fixed-size comparison method of 3 individuals, randomly selecting 3 individuals each time and retaining the one with the highest fitness until a new set of 50 individuals is formed. Crossover uses an adjacent pairing method, pairing individuals according to fitness from high to low, generating 2 offspring from each pair. The generation rule is that offspring 1 takes the first two items of the higher fitness and the last two items of the lower fitness, while offspring 2 takes the opposite combination. After generation, normalization is immediately performed, and out-of-bounds rollback is executed. Mutation is performed uniformly after the entire generation crossover is completed. The mutation probability of a single individual is fixed at 5%, the mutation amplitude is one step, and the mutation location is... The four weights rotate in a fixed order, and after mutation, normalization is immediately performed and out-of-bounds rollback is executed. Fitness evaluation is conducted uniformly after each generation, with evaluation criteria and definitions consistent. The system records the current generation's best fitness and the corresponding individual for convergence determination and subsequent calculations. The convergence determination threshold is determined before deployment using the project's historical data from the past 90 days as a sample. Specifically, a representative set of solutions is constructed for each historical day, and the change in the best fitness improvement over five consecutive generations is statistically analyzed. The median is selected as a candidate and then verified using the most recent 30 days of historical data playback to see if it correctly indicates convergence in more than half of the representative solutions. If not, the final convergence threshold is determined by shifting the ranking sequence up three places and documented in the project documentation. The internal system terminates when the change in the optimal fitness improvement over five consecutive generations does not exceed the final convergence threshold or when the total number of generations reaches 100. If either condition is met, the genetic algorithm ends and the system outputs the optimal individual weight and the corresponding preliminary correction coefficient. The preliminary correction coefficient is the median of the global proportional gain of the predicted depth obtained by driving S4 with the optimal weight in the most recent three time slices relative to the reference predicted depth before entering S6. The proportional gain is directly calculated in the current round. The system then enters the dynamic fusion of weights and environmental impact assessment steps. First, the absolute difference of the four weights before and after the iteration is calculated and multiplied by the coverage ratio of the factor at the effective monitoring location in the most recent three time slices to obtain four difference intensities as the environmental impact assessment results.The system calculates a weighted average of the four difference intensities based on their initial weights when entering S6 to obtain a single-iteration convergence judgment value. A smaller value indicates more stable weight changes after adaptive updates. The system then integrates optimization parameter stability information, which is the median of the absolute differences of the four weights of the best individuals over the past five generations. The system averages the iterative convergence judgment value and the stability information with equal weight to obtain the baseline coefficient correction value for a time-slice sample. Finally, it calculates a weighted average of the most recent three time-slice samples (3, 2, 1) to obtain the final baseline coefficient correction. The system then determines the model correction coefficient expansion based on algorithm efficiency. The efficiency factor is determined by dividing the total decrease in the average deviation index of the optimal individual sequence by the total number of algebras in the current round. The upper and lower cutoff boundaries of the efficiency factor are fixed once before deployment using the lower and upper quartiles of the project history. If the boundary is exceeded, it is truncated to the corresponding boundary. The extended value of the model correction coefficient is equal to the product of the coefficient correction basis and the efficiency factor. The system then performs stabilization processing, which includes two steps: the first step is time smoothing, which weights the current extended value with the previous two extended values ​​according to time weights of 3, 2, and 1; the second step is threshold pruning, which restricts the smoothing result to within the allowable range of variation. The allowable range of variation is set once before deployment, using 20% ​​above and below the median of the historical proportional gain obtained from the two rounds of S4 and S5 linkage before entering S6. The final stabilized model correction coefficients are then written into S7. All parameters involved in this step and their determination methods are as follows: the upper and lower bounds and step size of the four environmental factor weights are fixed values ​​obtained before deployment; the population size is 50; the comparison size is 3; crossover is adjacent pairing; the mutation probability is 5%; the mutation amplitude is one step; the number of representative time slices is 3; fitness is the percentage decrease in the average deviation index; and convergence is determined using the optimal fitness of the next 5 generations. The threshold for the magnitude of change in response intensity is set at 100, and the convergence threshold is determined by shifting the threshold three places upwards from the 90-day historical median after a majority of 30-day validation failures. The initial correction coefficient is the median of the proportional gain. The environmental impact assessment results are four vectors of difference intensity. The iterative convergence judgment value is a single value weighted by the initial weights of the difference intensity. The coefficient correction basis is a 3, 2, 1 weighted average of the samples from the last three time slices. The efficiency factor is the ratio of the total decrease to the total number of algebras, truncated according to the upper and lower quartiles of history. Smoothing uses a 3, 2, 1 weighting, and the allowable range of variation is 20% above and below the median of the proportional gain.

[0037] The convergence threshold is determined by using historical data from the past 90 days as a sample, calculating based on the magnitude of change in optimal fitness over five consecutive generations, selecting the median of the ranking as a candidate, and then verifying it using the majority pass rule over the past 30 days. If it fails, it is moved up three positions in the ranking sequence to solidify the threshold. This is because this determined path can accurately define the moment when the genetic algorithm's "effective improvement has tended to stop" in a traceable and noise-resistant manner, thus avoiding both premature termination and invalid iterations. Selecting 90 days can cover a complete climatic stage of temperature, salinity, humidity, and pH changes, ensuring that the candidate threshold reflects long-term statistical background rather than short-term random fluctuations. Using five consecutive generations as a measurement window can smooth out fluctuations in a single generation, filter out occasional local peaks and troughs, and ensure that convergence is triggered only when the magnitude of improvement continuously decreases over at least five generations. This ensures that the algorithm matches the pace of generating new solutions through iteration. By summarizing the improvement magnitude sequence of the 5 generations on a 90-day sample and taking the median, the impact of extreme cases on the threshold can be suppressed, ensuring that the threshold is representative of the overall distribution and robust to outliers. The majority of the most recent 30 days are used for verification, requiring that the candidate threshold can correctly distinguish between the stage of "still having substantial improvement potential" and the stage of "improvement is no longer sufficient to support continued iteration" in more than half of the representative solutions, thus keeping the threshold in sync with the current season and operating conditions. When a candidate fails the verification, a conservative correction is made by shifting up 3 positions, which moderately increases the threshold without changing the sample set and caliber, quickly reducing the risk of false alarm convergence and avoiding premature stopping and repeated restarts due to an excessively low threshold.

[0038] S7 includes obtaining the initial corrosion prediction distribution by multiplying the model correction coefficients by material property parameters after stabilization of the model. Based on the initial distribution, environmental factors are integrated by superimposing them onto the distribution mesh to determine the simulation framework after environmental variable integration, thus obtaining the adjusted distribution generation process. If the generation process meets the preset threshold, simulation data calibration is used. Parameters are adjusted by comparing with historical corrosion data to determine the corrosion rate calculation result. The rate value is obtained by dividing the result by the time interval, thus obtaining the accuracy evaluation index. For the evaluation index, a stable model is applied, and the prediction accuracy level is obtained by updating the coefficient matrix, determining the final value of the accuracy improvement. The corrosion prediction distribution is obtained from the final value and mapped to the environmental mesh to generate the improved prediction accuracy level in the environment.

[0039] In one possible implementation, the stabilized model correction coefficients output from S6 and the material property parameters determined by S2 and S4 are first read. The material property parameters only include the elastic modulus and Poisson's ratio, which are nominal values ​​solidified by batch material inspection records before deployment. At the start of the current loop, the system multiplies the stabilized model correction coefficients by the elastic modulus and then by the Poisson's ratio to obtain the corrected material property parameters. These corrected material property parameters, combined with the mesh and boundary type generated in stage S2, are written into the input channel of the finite element solver. The time step is kept fixed at an initial time step of 1 hour, a maximum time step of 4 hours, and a minimum time step of 15 minutes, and is run to obtain a solution covering all time steps. The system first predicts the initial distribution of corrosion at each time step. Then, it enters an environmental factor integration process. On a three-dimensional distribution grid consistent with the initial distribution, it reads the simultaneous temporal records of four environmental factors—temperature, salinity, humidity, and pH—point-by-point according to time steps. Based on pre-deployed lookup rules, each of the four factors is mapped to an equivalent boundary increment and an equivalent volume increment. Then, temperature, salinity, humidity, and pH are sequentially superimposed onto the corresponding grid positions and written in real-time, forming the integrated simulation framework of environmental variables and initiating the adjusted distribution generation process. At the end of each time step, the system calculates the consistency and stability metrics of the generation process and forms a single process metric. The consistency metric is the sum of the global depth increment at the current time step and the previous time step. The average of the absolute values ​​of the differences in global depth increments at each time step is used as the stability metric. The average of the absolute values ​​of the differences in depth increments between adjacent grids in the current time step's height gradient region is used. The single process metric is a single value obtained by equally weighted averaging of the two metrics. Before deployment, a preset threshold is calculated daily using the historical playback of the same structure over the past 90 days as a sample, forming a sample sequence. This sequence is sorted from smallest to largest, and the median is used as a candidate threshold. The threshold is then verified using records from the most recent 30 days. The verification rule is that at least half of the single process metrics in the known stable samples are not greater than the candidate, and at least half of the single process metrics in the known unstable samples are greater than the candidate. If the verification fails, the threshold is shifted up 3 positions in the sorted sequence and the final threshold value is obtained. The value is fixed in the deployment document; if a single process metric in the current solution is not greater than the final threshold, the system enters the simulation data calibration; if it is greater, the calibration is skipped and the system proceeds to the next time step until the threshold is met or all time steps are completed. After entering the simulation data calibration, the system subtracts the updated distribution from the historical corrosion data point by point in the same grid and at the same time point, takes the absolute value, sums all the points involved in the comparison, and divides the sum by the number of points to obtain the average absolute deviation of the current step as the error of the current step. At the same time, the depth difference of the current step is divided by the length of the time step to obtain the corrosion rate value of the current step and recorded in the rate sequence. The system averages the average absolute deviation of the current step over the entire time series to obtain the accuracy evaluation index. The accuracy evaluation index is a single value and the smaller it is, the better the fit.The system applies a stable model based on accuracy assessment metrics to obtain the prediction accuracy level and determine the final value. The stable model application is accomplished by updating the coefficient matrix and applying it to the depth increment and rate sequence at each time step. The updated coefficient matrix is ​​a scaling matrix of the coefficient set fixed before deployment across four environmental factor channels. The scaling values ​​are determined during the project deployment phase and remain unchanged during operation. The application steps are as follows: at each time step, the updated coefficient matrix is ​​multiplied by the depth increment at that step, and the rate sequence is updated synchronously. Then, the accuracy assessment metrics are recalculated using the same caliber as the initial calculation. If the updated accuracy assessment metric decreases by at least 10 compared to the initial accuracy assessment metric, it is considered a valid improvement. The updated accuracy assessment metric is then divided into 100 accuracy levels in reverse order to obtain the prediction accuracy level. A higher accuracy level indicates better accuracy. The system uses the difference between the prediction accuracy level and the initial accuracy level as a measure of accuracy improvement. The minimum improvement threshold is set as a fixed value before deployment. If the accuracy improvement... If the elevation measure is not less than the minimum elevation threshold, the prediction accuracy level is taken as the final value. If it is less than the minimum elevation threshold, the system applies and evaluates the same value once after scaling down each scale by a fixed scaling factor without changing the proportional direction and weight sum of the update coefficient matrix. The scaling factor is fixed before deployment and the maximum number of repetitions is 3. If the minimum elevation threshold is not reached after exceeding the limit, the accuracy level corresponding to the initial accuracy evaluation index is retained as the final value and the reason for failure is recorded. After determining the final value, the system outputs a set of depth distributions corresponding to the final value as the corrosion prediction distribution and maps the distribution to the environmental grid according to the spatial index. The environmental grid is the grid corresponding to the monitoring layout. The mapping method is nearest neighbor mapping and no other smoothing or interpolation modules are introduced in this step. At the same time as outputting, the system records the stable model correction coefficients, corrected material property parameters, single process metric, final threshold, accuracy evaluation index, prediction accuracy level and final value used in this cycle for auditing purposes.The parameters and determination methods involved in this step are as follows: The model correction coefficient after stabilization is the single proportional value after S6 smoothing and trimming, which is directly read in this round and multiplied by the material property parameters all at once. The material property parameters only include the elastic modulus and Poisson's ratio, and are solidified by material batch inspection before deployment. The time step is set to an initial 1 hour, a maximum of 4 hours, and a minimum of 15 minutes, and is fixed in this step. Environmental factors are temperature, salinity, humidity, and pH, which are recorded simultaneously and mapped to equivalent increments according to the table lookup rules before deployment, and then superimposed in a fixed order. The single process metric is the single value obtained by the equal weighted average of the consistency metric and the stability metric. The definition and calculation caliber of the consistency metric and the stability metric are solidified before deployment. The preset threshold is determined by shifting the median of the 90-day sample up by 3 places if it fails the 30-day review, and it remains unchanged during operation. The accuracy evaluation index is the average of the absolute deviation of the current step in the full time series. The update coefficient matrix is ​​the proportional set solidified before deployment, the accuracy level is 100, the minimum improvement threshold is the single value solidified before deployment, the scaling factor is the single value solidified before deployment, and the maximum repetition is 3 times.

[0040] 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 alterations 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. A finite element method for modeling carbon steel corrosion, characterized in that, include: S1. By collecting real-time data from environmental sensors, an initial set of parameters including temperature, salinity, and material properties is obtained to determine the preliminary corrosion rate distribution; S2. Based on the initial parameter set, use finite element simulation to generate a corrosion depth map of the metal structure and obtain the deviation value between the preliminary simulation results and the measured corrosion data. S3. If the deviation value exceeds the preset threshold, the parameter coupling relationship is processed by the genetic algorithm to obtain an optimized parameter set for correcting the model input. S4. Rerun the finite element simulation using the optimized parameter set to determine the local variation trend of corrosion rate in complex structures and obtain an updated predicted depth map. S5. By comparing the updated predicted depth map with historical measured data, obtain the residual bias caused by dynamic changes and determine the weight of environmental factors. S6. Based on the weights of environmental factors, use a genetic algorithm to iteratively update the adaptive mechanism, determine whether the parameter optimization has converged, and obtain the model correction coefficients after stabilization. S7. By applying the stabilized model correction coefficients to the finite element simulation, the final corrosion prediction distribution is generated, and the level of improved prediction accuracy in the environment is determined.

2. The carbon steel corrosion modeling method based on finite element method according to claim 1, characterized in that: S1 includes: Real-time data is collected by environmental sensors to obtain an initial set of parameters including temperature, salinity, and material properties, and to determine the preliminary corrosion rate distribution. Based on the preliminary corrosion rate distribution, corrosion trend indicators are obtained from historical data. The corrosion trend indicators are obtained by comparing the differences between historical corrosion records and the current distribution, resulting in an extended parameter set. An extended parameter set is adopted, and humidity influencing factors are integrated. Humidity influencing factors are collected in real time from auxiliary sensors to determine the corrosion acceleration threshold. The corrosion acceleration threshold is determined by the product of humidity and salinity in the extended parameter set exceeding a preset threshold, thus obtaining the adjusted distribution. Based on the adjusted distribution, pH monitoring data was integrated, with pH monitoring data expanded from environmental sensors, to determine the final corrosion rate distribution.

3. The carbon steel corrosion modeling method based on finite element method according to claim 1, characterized in that: S2 includes: Based on the initial parameter set, a three-dimensional mesh model of the metal structure is constructed using the finite element method. Node coordinates and element connections are extracted from the mesh model to obtain the basic structure for corrosion simulation. By simulating the corrosion of the basic structure, material properties and boundary conditions are imported, loads are applied to environmental factors, and a finite element solver is used to calculate the corrosion rate at each node with nodal coordinates and loads as inputs. The finite element solver solves the corrosion diffusion equation based on the finite element discretization equation to determine the corrosion depth distribution map. Obtain a corrosion depth distribution map and compare it with the measured corrosion data. If the deviation exceeds the preset threshold, adjust the material property parameters. The adjustment is achieved by changing the elastic modulus and Poisson's ratio parameters. Then, determine the optimized depth map. Based on the optimized depth map, calculate the sum of squares of the differences between the simulated and measured values; By using preliminary deviation quantification indicators, iteratively updating the time step control, and performing local mesh refinement processing, the deviation values ​​between the final simulation results and the measured data are obtained.

4. The carbon steel corrosion modeling method based on finite element method according to claim 1, characterized in that: S3 includes: The deviation value is obtained by calculating the difference between the current model output and the expected value, and it is determined whether the deviation value exceeds the preset threshold. If the deviation value exceeds the preset threshold, the coupling relationship is extracted from the parameter set, and the genetic algorithm is used to select the coupling relationship to obtain the initial optimization parameters. Based on the initial optimization parameters, the fitness function is used to evaluate the degree of parameter fit and determine the iterative direction of the parameter set; By using crossover and mutation operations, the parameter set is updated to obtain the converged optimized parameters; To optimize the parameters, the model input is corrected to obtain the adjusted input set.

5. The carbon steel corrosion modeling method based on finite element method according to claim 1, characterized in that: S4 includes: Obtain the set of optimized parameters from the preset parameter library, select the environmental corrosion factor for the structure, and determine the preliminary simulation input data; The finite element simulation was rerun using the input data to calculate the corrosion rate distribution for complex structures, and the local rate matrix was obtained. Determine the changing trend in the local rate matrix. If the trend exceeds a preset threshold, adjust the structural mesh division and obtain the trend correction value through mesh refinement method. By integrating the material's corrosion resistance properties with trend correction values, a corrosion depth prediction model is generated, and an intermediate depth distribution map is obtained through attribute weighted summation. Based on the intermediate depth distribution map, stress influence factors are integrated, and the predicted depth map is updated by factor overlay calculation.

6. The carbon steel corrosion modeling method based on finite element method according to claim 1, characterized in that: S5 includes: By comparing the updated predicted depth map with historical measured data, the residual deviation value caused by dynamic changes is obtained, and the weights of environmental factors that need further adjustment are determined. The residual deviation value is processed by the deviation comparison method. The dynamic change source is extracted from the historical measured data to obtain the influence degree of the factor. The deviation comparison method obtains the average deviation index by calculating the sum of the absolute values ​​of the differences between each point in the predicted depth map and the corresponding points in the historical measured data, divided by the number of points. If the influence of a factor exceeds a preset threshold, a weight adjustment strategy is executed for the source of dynamic change to obtain the adjusted environmental factor weight. The weight adjustment strategy updates the weight value by multiplying the influence of the factor by the current environmental factor weight and normalizing it. Based on the adjusted environmental factor weights, a data update stream is generated to determine the trend of change. The data update stream is formed by continuously recording the adjusted environmental factor weight sequence. The measured verification step is extracted from the trend analysis to complete the adjustment and optimization loop. The measured verification step is carried out by matching the trend analysis with the new round of historical measured data and iterating the residual deviation value.

7. The carbon steel corrosion modeling method based on finite element method according to claim 1, characterized in that: S6 includes: The weights of environmental factors are obtained, and an adaptive mechanism is adopted by iterative updating using a genetic algorithm. The genetic algorithm uses environmental factors as the initial population, generates new solutions iteratively through crossover and mutation operations, determines whether the parameter optimization has converged, and obtains preliminary correction coefficients. For the initial correction coefficients, the weights are dynamically adjusted to determine the environmental impact assessment results from the mechanism's adaptation to changes. The environmental impact assessment results are obtained by comparing the weight differences before and after the iteration.

8. The carbon steel corrosion modeling method based on finite element method according to claim 7, characterized in that: S6 further includes: The iterative convergence judgment value is obtained from the environmental impact assessment results, the optimization parameters are stabilized, and the coefficient correction basis is obtained. The coefficient correction basis is formed by weighted averaging of the judgment value. Based on the coefficient correction foundation, the efficiency of the fusion algorithm is used to determine the extended value of the model correction coefficients, which are then expanded by multiplying by an efficiency factor. The stable output is obtained from the extended values ​​of the model correction coefficients, resulting in the stable model correction coefficients.

9. The carbon steel corrosion modeling method based on finite element method according to claim 1, characterized in that: S7 includes: By multiplying the model correction coefficients by the material property parameters after stabilization, the finite element simulation input data is obtained, and the initial distribution of corrosion prediction is obtained. Based on the initial distribution, environmental factors are integrated. By superimposing environmental factors onto the distribution grid, the simulation framework after the integration of environmental variables is determined, and the adjusted distribution generation process is obtained. If the generation process meets the preset threshold, then simulated data calibration is used. By comparing with historical corrosion data, the parameters are adjusted, the corrosion rate calculation result is judged, and the rate value is obtained by dividing the result by the time interval, thus obtaining the accuracy evaluation index.

10. A carbon steel corrosion modeling method based on finite element method according to claim 9, characterized in that: The S7 also includes: For the evaluation indicators, a stable model is applied, and the prediction accuracy level is obtained by updating the coefficient matrix to determine the final value of the accuracy improvement. The corrosion prediction distribution is obtained from the final value, and the distribution is mapped to the environmental mesh to generate an improved level of prediction accuracy in the environment.

Images