A method for determining a failure simulation mode of a reinforced soil retaining wall based on numerical simulation

Abstract

This invention discloses a method for identifying failure modes of reinforced soil retaining walls based on numerical simulation, belonging to the field of civil structure safety prediction technology. The method includes: acquiring three-dimensional numerical simulation data, simulating the interaction between soil and reinforcement materials, setting boundary conditions and loading conditions, and generating a standard simulation dataset; identifying the boundary conditions and loading methods from the standard simulation dataset using an adaptive density clustering algorithm to obtain an enhanced simulation dataset; fusing the enhanced simulation dataset with the standard simulation dataset to generate an extended simulation dataset; dynamically optimizing and adjusting the failure mode identification criteria based on the failure development trend and real-time environmental changes; classifying failure modes using the new criteria; and generating a failure mode identification report. This invention improves the accuracy and reliability of failure mode prediction, providing effective technical support for failure detection and early warning in complex soil structures.

Description

Technical Field

[0001] This invention relates to the field of civil structure safety prediction technology, and in particular to a method for determining the failure simulation mode of reinforced soil retaining walls based on numerical simulation. Background Technology

[0002] In recent years, reinforced soil retaining walls, as a novel type of civil engineering structure, have received widespread attention in civil engineering applications. By embedding reinforcing materials into the soil, the stability and deformation resistance of the soil can be effectively enhanced, especially under large loads, exhibiting good compressive strength and durability. In existing technologies, numerical simulation methods and computer-aided engineering have become the mainstream research approach for predicting the failure modes of reinforced soil retaining walls. Using the finite element method, a three-dimensional numerical model of the reinforced soil retaining wall can be established to simulate the interaction between the soil and the reinforcing material. By setting boundary conditions and loading conditions, the failure characteristics of the structure can be analyzed.

[0003] However, traditional numerical simulation methods often rely on empirically set loading conditions and boundary conditions, which cannot accurately reflect the complex changes in the actual environment, leading to deviations in simulation results. Although methods such as finite element analysis can generate certain failure mode data, existing analysis methods lack in-depth identification and analysis of the dynamic evolution of failure modes under different working conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a numerical simulation-based method for determining the failure mode of reinforced soil retaining walls, which solves the problem of deviation in loading conditions and boundary conditions.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides a method for identifying the failure mode of reinforced soil retaining walls based on numerical simulation, which includes: collecting three-dimensional numerical simulation data, simulating the interaction between soil and reinforcement materials, setting boundary conditions and loading conditions, and generating a standard simulation dataset.

[0008] The adaptive density clustering algorithm is used to identify the boundary conditions and loading methods of the working conditions from the standard simulation dataset to obtain the enhanced simulation dataset. The enhanced simulation dataset is then fused with the standard simulation dataset to generate an extended simulation dataset.

[0009] Based on the extended simulation dataset, feature extraction is performed using a convolutional neural network to obtain damage mode features, and cluster analysis is used to identify different damage mode types according to the damage mode discrimination criteria.

[0010] The evolution of damage mode characteristics and damage mode types at different time points is analyzed, and time series analysis methods are used to monitor the trend of characteristic changes and the transformation of mode types, so as to identify the turning point of damage and the trend of damage development.

[0011] Based on the development trend of damage and changes in real-time environmental conditions, the criteria for identifying damage modes are dynamically optimized and adjusted, and damage modes are classified using the new criteria to generate a damage mode identification report.

[0012] As a preferred embodiment of the numerical simulation-based method for determining the failure mode of reinforced soil retaining walls described in this invention, the specific steps for acquiring three-dimensional numerical simulation data, simulating the interaction between the soil and the reinforcing material, setting boundary conditions and loading conditions, and generating a standard simulation dataset are as follows.

[0013] Collect three-dimensional numerical simulation data, construct a three-dimensional structure of soil and reinforcement material by selecting the mechanical properties of soil and reinforcement material, and generate a structural contact dataset;

[0014] Based on the structural contact dataset, the boundary conditions and loading conditions between the soil and the reinforcement material are set, and the loading condition dataset is generated.

[0015] The loading condition dataset is calculated using the finite element method to obtain a calculation data package. Three-dimensional displacement field, stress field, and strain field data are extracted from the calculation data package. Through data cleaning, denoising, and standardization, a standard simulation dataset is generated.

[0016] As a preferred embodiment of the numerical simulation-based method for determining the failure mode of reinforced soil retaining walls described in this invention, the step of identifying the boundary conditions and loading methods of the working conditions from the standard simulation dataset using an adaptive density clustering algorithm to obtain the enhanced simulation dataset comprises the following specific steps:

[0017] The standard simulation dataset is clustered using an adaptive density clustering algorithm, and boundary conditions and loading methods with similar characteristics are identified to generate a clustered dataset.

[0018] Enhanced datasets are extracted from the clustered datasets, and random parameter adjustments and operating condition combinations are performed on each enhanced dataset to obtain an enhanced simulation dataset.

[0019] As a preferred embodiment of the numerical simulation-based method for determining the failure mode of reinforced soil retaining walls according to the present invention, the specific steps for fusing the enhanced simulation dataset with the standard simulation dataset to generate an extended simulation dataset are as follows.

[0020] The standard simulation dataset and the enhanced simulation dataset are merged, and a comprehensive simulation dataset is generated by matching similar working conditions and performing cross-validation.

[0021] Physical consistency verification is performed on the integrated simulation set to obtain the verified integrated simulation set. Then, the verified integrated simulation set is denoised and data augmented to generate an extended simulation dataset.

[0022] As a preferred embodiment of the numerical simulation-based method for determining the failure mode of reinforced soil retaining walls according to the present invention, the step of extracting features from the extended simulation dataset using a convolutional neural network to obtain failure mode features is as follows:

[0023] The extended simulation dataset is input into a convolutional neural network for feature extraction, generating a high-dimensional feature dataset.

[0024] Principal component analysis is used to reduce the dimensionality of high-dimensional feature datasets, and principal components are selected to further reduce data dimensionality, thereby generating destruction mode features.

[0025] As a preferred embodiment of the numerical simulation-based method for identifying the failure mode of reinforced soil retaining walls according to the present invention, the step of using cluster analysis to identify different failure mode types based on failure mode identification criteria is as follows:

[0026] The damage pattern features are clustered using the K-means clustering algorithm to obtain a damage pattern cluster dataset. The damage pattern features of each cluster in the damage pattern cluster dataset are then analyzed to identify the damage pattern category trend corresponding to the cluster.

[0027] The trends of destruction patterns are compared one by one with the criteria for determining destruction patterns. The characteristics of destruction patterns that meet the criteria are classified into different destruction pattern types based on feature similarity and matching degree.

[0028] As a preferred embodiment of the numerical simulation-based method for determining the failure mode of reinforced soil retaining walls according to the present invention, the following steps are taken: The evolution of failure mode characteristics and failure mode types at different time points, combined with time series analysis methods, is monitored to track the trend of characteristic changes and mode type transitions.

[0029] The destruction mode features and destruction mode types are divided into multiple time windows according to the time series, the time window dataset is obtained, the destruction mode features in the time window dataset are aggregated, the statistics of each feature are calculated, and the time series dataset is generated.

[0030] Based on time series datasets, regression analysis is used to analyze the changing trends of damage mode characteristics over time, predict the future evolution of damage modes, obtain damage mode prediction data, and verify the damage mode prediction data through residual analysis to evaluate the difference between the prediction results and the actual situation, and generate verified prediction data.

[0031] As a preferred embodiment of the numerical simulation-based method for identifying the failure mode of reinforced soil retaining walls according to the present invention, the specific steps for identifying the failure inflection point and failure development trend are as follows.

[0032] Inflection point detection is performed on the verified prediction data to identify disruptive inflection points;

[0033] By combining the verified prediction data with the failure inflection point, the rate and direction of change of the failure mode characteristics are analyzed.

[0034] Based on the rate and direction of change, analyze the evolution of damage mode characteristics, identify points of change in damage progression, and obtain the trend of damage development.

[0035] As a preferred embodiment of the numerical simulation-based method for determining the failure mode of reinforced soil retaining walls according to the present invention, the specific steps of dynamically optimizing and adjusting the failure mode determination criteria based on the failure development trend and real-time environmental condition changes are as follows.

[0036] Collect environmental condition data, perform data fusion and cleaning, analyze real-time environmental condition changes, combine real-time environmental condition changes with the trend of damage development, and generate a real-time environmental dataset.

[0037] The criteria for determining damage modes are dynamically adjusted based on real-time environmental data, and new criteria for determining damage modes are generated.

[0038] As a preferred embodiment of the numerical simulation-based method for determining the failure modes of reinforced soil retaining walls according to the present invention, the specific steps for classifying failure modes using a new failure mode discrimination standard and generating a failure mode discrimination report are as follows:

[0039] Based on the new damage mode discrimination criteria, the damage mode features within each time window are re-matched and classified. By comparing the similarity between the damage mode features and the new discrimination criteria, damage mode labels are generated.

[0040] The destruction mode labels are used to classify different destruction mode types, and cross-validation and parameter adjustment are performed based on the destruction mode labels to generate a destruction mode discrimination report.

[0041] The beneficial effects of this invention are as follows: by using adaptive density clustering algorithm, convolutional neural network, clustering analysis technology and data fusion, feature extraction and pattern classification, the accuracy and diversity of simulation datasets are enhanced; through the intelligent steps of computer-aided engineering, different failure modes can be automatically identified and classified, and by combining time series analysis and real-time environmental conditions to dynamically optimize the discrimination criteria, the accuracy and reliability of failure mode prediction are improved, providing effective technical support for failure detection and early warning in complex soil structures. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of a simulation mode discrimination method for reinforced soil retaining walls based on numerical simulation.

[0044] Figure 2 A flowchart generated for a standard simulation dataset.

[0045] Figure 3 This is a flowchart for feature extraction of destruction modes.

[0046] Figure 4 A flowchart for dynamically optimizing the discrimination criteria. Detailed Implementation

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0048] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0050] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for determining the failure simulation mode of reinforced soil retaining walls based on numerical simulation, including the following steps:

[0051] S1. Collect three-dimensional numerical simulation data, simulate the interaction between soil and reinforced materials, set boundary conditions and loading conditions, and generate a standard simulation dataset.

[0052] S1.1 Collect three-dimensional numerical simulation data, construct a three-dimensional structure of soil and reinforcement material by selecting the mechanical properties of soil and reinforcement material, and generate a structural contact dataset.

[0053] It should be noted that three-dimensional numerical simulation data is collected to obtain the spatial geometry and mechanical properties of the soil and reinforcing materials. These mechanical properties include, but are not limited to, parameters such as the elastic modulus, Poisson's ratio, and friction angle of the soil, and physical properties such as the tensile strength and elastic modulus of the reinforcing materials. Using the mechanical properties of the soil and reinforcing materials, including the elastic modulus, Poisson's ratio, and friction angle of the soil, and the tensile strength and elastic modulus of the reinforcing materials, the interaction between the soil and reinforcing materials is calculated based on these mechanical parameters and their geometric shapes. The stiffness and friction coefficient of the contact surface between the soil and reinforcing materials are determined; these contact characteristics describe the interaction between them. Based on this, mechanical data at the contact points between the soil and reinforcing materials, including contact forces, displacements, and frictional forces, are obtained through numerical calculation methods, thereby generating a structural contact dataset.

[0054] S1.2. Based on the structural contact dataset, set the boundary conditions and loading conditions between the soil and the reinforcement material, and generate the loading condition dataset.

[0055] It should be noted that boundary conditions between the soil and reinforced material are set based on information from the structural contact dataset. These boundary conditions include the stiffness, friction coefficient, and other physical properties of the contact surface between the soil and reinforced material, ensuring accurate reflection of the interaction between them in the numerical simulation. Loading cases are set according to the mechanical properties and operating conditions of the soil and reinforced material, including the type, magnitude, direction, and location of the applied external loads. These loading cases cover the load variations that the soil and reinforced material may experience under different conditions, simulating situations they might encounter in actual use. A loading case dataset is generated by combining the boundary conditions and loading cases.

[0056] S1.3 Calculate the loading condition dataset using the finite element analysis method, obtain the calculation data package, and extract the three-dimensional displacement field, stress field, and strain field data from the calculation data package. Generate a standard simulation dataset through data cleaning, denoising, and standardization.

[0057] It should be noted that the loading condition dataset is calculated using the finite element method. Based on the set boundary conditions and loading conditions, the mechanical response between the soil and the reinforced material is solved numerically to generate a calculation data package. The calculation data package includes response data for each part of the structure under different loading conditions, covering displacement, stress, and strain indices. Specifically, it includes the displacement, stress, and strain values ​​of the soil and reinforced material at each node. The response data is calculated using finite element analysis based on three-dimensional spatial coordinates. The displacement value at each node describes the displacement of the structure in each direction, the stress value reflects the internal mechanical state of the material, and the strain value reflects the degree of deformation of the material under external forces. The deformation and stress conditions of the interaction body between the soil and reinforced material under loading conditions are also shown. Data cleaning, denoising, and standardization are performed to remove unreasonable noise and outliers, ensuring the accuracy and usability of the data and generating a standard simulation dataset.

[0058] It should also be noted that the finite element method is a numerical computation method used to solve partial differential equations in complex physical problems, and is particularly suitable for fields such as structural mechanics, heat conduction, and fluid mechanics. The finite element method approximates the entire behavior by dividing the entire problem domain into several small, simple sub-domains and establishing a local mathematical model in each sub-domain.

[0059] S2. Identify the working condition boundary conditions and working condition loading methods from the standard simulation dataset using an adaptive density clustering algorithm to obtain an enhanced simulation dataset. Then, fuse the enhanced simulation dataset with the standard simulation dataset to generate an extended simulation dataset.

[0060] S2.1. Cluster the standard simulation dataset using an adaptive density clustering algorithm, identify the working conditions, boundary conditions and loading methods with similar characteristics, and generate a clustered dataset.

[0061] It should be noted that the boundary conditions and loading methods of each data point in the standard simulation dataset are extracted to form a feature vector. An adaptive density clustering algorithm identifies high-density regions as cluster centers based on the local density of each data point and the distance relationships between data points. Core points and their adjacent high-density data points are grouped into the same cluster, while low-density regions are considered noise. In this way, the adaptive density clustering algorithm can automatically determine the number of clusters without pre-setting. The adaptive density clustering algorithm identifies high-density regions as cluster centers and low-density regions as noise based on the density distribution of data points. After clustering, the data points in the standard simulation dataset are divided into multiple clusters, each cluster representing a group of data points with similar boundary conditions and loading methods, thus generating a clustered dataset.

[0062] It should also be noted that adaptive density clustering is a clustering method based on data point density, primarily used to discover data clusters with different density distributions. Unlike traditional clustering algorithms, adaptive density clustering does not require a pre-defined number of clusters. Instead, it automatically determines the number and shape of clusters by evaluating the local density of data points and the distance relationships between them. Specifically, adaptive density clustering identifies high-density regions based on the local density of data points and groups similar data points within these regions into clusters, while low-density regions are treated as noise. Data points can be adaptively grouped into different clusters based on their own density characteristics.

[0063] S2.2 Extract augmented datasets from the clustered datasets, and perform random parameter adjustments and operating condition combinations on each augmented dataset to obtain augmented simulation datasets.

[0064] It should be noted that, based on the characteristics of each cluster in the clustered dataset, representative boundary conditions and loading methods are selected. Random parameter adjustments are performed on each augmented dataset; this process includes randomly varying the parameters in the boundary conditions to simulate different combinations of operating conditions. In this way, multiple new augmented datasets can be generated, further enriching the diversity and complexity of the simulation data. During the generation of augmented datasets, different operating conditions will also be combined, and by adjusting the combination methods, it is ensured that the generated augmented simulation datasets can cover more possible actual operating conditions, thus obtaining augmented simulation datasets.

[0065] S2.3 Merge the standard simulation dataset and the enhanced simulation dataset, and generate a comprehensive simulation dataset by matching similar working conditions and performing cross-validation.

[0066] It should be noted that by matching similar boundary conditions and loading methods in the standard simulation dataset and the enhanced simulation dataset, it is ensured that the merged dataset covers the same or similar simulation scenarios and performs cross-validation. The consistency and effectiveness are verified by comparing similar working conditions in the standard and enhanced simulation datasets. Feature vectors for each working condition are extracted from both datasets, and the distance between them is calculated to measure the similarity of the working condition features. The matching degree is derived by comparing the similarity of the feature vectors, for example, using cosine similarity. These calculations ensure the accuracy and reliability of the merged data, thereby generating a comprehensive simulation set.

[0067] It should be noted that the expression for calculating the error between different operating conditions is as follows:

[0068] ;

[0069] in: This refers to the error between different operating conditions; This represents the total number of operating conditions. An index for the current operating condition; For the first in the standard simulation dataset Feature vectors of each working condition; To enhance the first in the simulation dataset Feature vectors of each working condition; For calculating operating conditions and Distance between them, working conditions For a specific operating condition in the standard simulation dataset, the operating condition is... To enhance a specific operating condition in the simulation dataset.

[0070] S2.4 Perform physical consistency verification on the integrated simulation set, obtain the verified integrated simulation set, and perform denoising and data augmentation on the verified integrated simulation set to generate an extended simulation dataset.

[0071] It should be noted that the data in the comprehensive simulation set is compared with known physical laws to check whether each simulation data conforms to physical laws, such as mechanical equilibrium and energy conservation. If any inconsistencies or unreasonable parts are found, corresponding adjustments are made to ensure the physical authenticity of the simulation data, resulting in a verified comprehensive simulation set, which has excluded parts that do not conform to physical laws. The verified comprehensive simulation set is then further improved in quality by using appropriate denoising algorithms, such as wavelet denoising, to remove noise and outliers. Data augmentation is also performed, and an extended simulation dataset is generated by randomly transforming and expanding the data in the verified comprehensive simulation set.

[0072] S3. Based on the extended simulation dataset, feature extraction is performed using a convolutional neural network to obtain damage mode features, and cluster analysis is used to identify different damage mode types according to the damage mode discrimination criteria.

[0073] S3.1 Input the extended simulation dataset into the convolutional neural network for feature extraction to generate a high-dimensional feature dataset.

[0074] It should be noted that each data item in the extended simulation dataset is used as input and fed into a convolutional neural network for processing. The convolutional neural network extracts features from the input data through multiple convolutional and pooling layers, extracting spatial features and local patterns from the data layer by layer. In the convolutional layers, the data is filtered by convolutional kernels to extract feature information at different scales and orientations; the pooling layers are used to reduce the dimensionality of the data while retaining important feature information, generating a high-dimensional feature dataset.

[0075] It should also be noted that the training process of a convolutional neural network (CNN) involves dividing the input dataset into training and validation sets, and standardizing or normalizing the data to improve the stability of the training. During forward propagation, the input data passes through convolutional layers to extract local features, then through pooling layers for downsampling, and finally through fully connected layers to output the prediction result. A loss function is calculated to measure the difference between the prediction result and the true label. Next, the gradient of the loss function with respect to each parameter is calculated through backpropagation, and the parameters are updated using optimization algorithms (such as gradient descent). This process iterates until the CNN converges and the parameters are adjusted to minimize the loss.

[0076] S3.2. Principal component analysis is used to reduce the dimensionality of the high-dimensional feature dataset, and principal components are selected to reduce the data dimensionality and generate destruction mode features.

[0077] It should be noted that the covariance matrix reflects the correlation between various features in a dataset. By calculating the covariance between each pair of features, the linear relationship between features can be obtained. The eigenvalues ​​and eigenvectors of the covariance matrix are calculated, where the eigenvalues ​​represent the importance of each principal component in the data, and the eigenvectors represent the main direction of data variation. By sorting the eigenvalues, the eigenvectors with the largest eigenvalues ​​are selected as principal components, which represent the most informative parts of the data. The high-dimensional feature dataset is projected onto the principal components, thereby reducing the dimensionality of the data and preserving most of the variability and key information. This dimensionality reduction process generates the destruction mode features.

[0078] S3.3. Perform cluster analysis on the damage mode features using the K-means clustering algorithm to obtain a damage mode cluster dataset. Then, perform damage mode feature analysis on each cluster in the damage mode cluster dataset to identify the damage mode category trend corresponding to the cluster.

[0079] It should be noted that the damage mode feature data is input into the K-means algorithm, which divides the data into clusters according to a certain number by calculating the distance between data points. The K-means algorithm iteratively optimizes the centroid of each cluster, minimizing the differences between data points within each cluster and maximizing the differences between different clusters, thus obtaining a damage mode cluster dataset, where each cluster represents a damage mode with similar characteristics. Data features such as stress, displacement, and strain are extracted from each damage mode cluster, reflecting the characteristics of the damage modes under different working conditions. Then, the K-means clustering algorithm is used to further divide the feature data within the damage mode clusters. By calculating the similarity or distance (e.g., Euclidean distance) between data points, different subgroups within the damage mode clusters are identified. For example, by calculating the Euclidean distance between data points, if the distance between two data points is small, they can be considered to belong to the same subgroup. By observing the changing trends of the data within each damage mode cluster, statistical methods (such as calculating the mean and standard deviation) are used to observe the changing patterns of the damage mode features, identifying the damage mode category trend corresponding to the damage mode cluster. For example, by observing the changing trends of damage mode characteristics, we can discover that certain characteristics gradually increase over time, thereby identifying the corresponding damage mode category trends. In this way, we can accurately identify the damage mode category trends corresponding to each damage mode cluster. The damage mode category represented by the cluster is determined by the data characteristics within each cluster, and the evolution trend of the damage mode category under different times or conditions generates the damage mode category trend.

[0080] S3.4 Compare the damage mode category trends with the damage mode discrimination criteria one by one, and classify the damage mode features that meet the damage mode discrimination criteria into different damage mode types according to feature similarity and matching degree.

[0081] It should be noted that the process involves comparing each feature in the damage pattern category trend with the corresponding feature in the damage pattern discrimination criteria, calculating the differences between them, and assessing the similarity. If a feature in the damage pattern category trend matches the conditions in the discrimination criteria, the damage pattern is considered to meet the criteria. For damage pattern features that meet the conditions, they are classified into specific damage pattern types, thus forming a clear damage pattern classification.

[0082] S4. Evolve the characteristics and types of damage modes at different time points, and combine time series analysis methods to monitor the trend of characteristic changes and the transformation of mode types, and identify the turning points and development trends of damage.

[0083] S4.1 Divide the destruction mode features and destruction mode types into multiple time windows according to the time series, obtain the time window dataset, aggregate the destruction mode features in the time window dataset, calculate the statistics of each feature, and generate a time series dataset.

[0084] It should be noted that, based on the time intervals of the time series data, the time range of each time window is determined, and the damage mode features and damage mode types are assigned to the corresponding time windows in chronological order, obtaining a time window dataset. This dataset contains the damage mode features and damage mode types within each time window. For the damage mode features in each time window dataset, aggregation processing is performed, and statistics for each feature are calculated, including mean, standard deviation, maximum, and minimum values. These statistics reflect the distribution and changing trends of the damage mode features within the time window. By aggregating the damage mode features from all time window datasets, a time series dataset is generated.

[0085] S4.2 Based on the time series dataset, the characteristics of the damage mode change over time are analyzed using regression analysis to predict the future evolution of the damage mode, obtain damage mode prediction data, and verify the damage mode prediction data using residual analysis to evaluate the difference between the prediction results and the actual situation, and generate verified prediction data.

[0086] It should be noted that the regular changes in the characteristics of the damage mode over time are identified by analyzing their changing trends. This process involves extracting long-term trends by calculating the rate of change at each time point or using smoothing methods (such as moving averages), thereby identifying trends such as growth, decline, or fluctuation. Based on the identified trends, regression analysis is used to predict the future evolution of the damage mode, generating damage mode prediction data. The difference between the predicted and actual values ​​at each time point is calculated based on the damage mode prediction data, and the distribution of the residuals is analyzed. If the residuals exhibit a random distribution, the prediction is accurate; if the residuals show an inherent regularity, the prediction fails to fully reflect the actual situation. Based on this, parameters are adjusted according to the magnitude of the residuals to improve prediction accuracy, generating validated prediction data.

[0087] It should also be noted that residual analysis refers to the difference between predicted values ​​and actual observed values. Residual analysis checks the deviation between the predicted results and the actual data by calculating the residuals of each data point.

[0088] Random distribution refers to a distribution where, within a certain range, the occurrence of data points does not follow a fixed pattern; the probability of each data point appearing is independent, and all possible values ​​have the same probability of occurrence. For example, when rolling a fair die, the probability of each number appearing is equal and is unaffected by the result of the previous roll.

[0089] Built-in patterns refer to hidden, inherent patterns or relationships within data or phenomena. These patterns are typically not observable directly but require in-depth analysis to reveal. Unlike random distributions, built-in patterns indicate that data changes possess a degree of predictability and structure. For example, the long-term trend of the stock market may be influenced by economic factors, and this trend may exhibit regular changes over a certain period, although short-term fluctuations may appear random.

[0090] S4.3 Detect inflection points in the verified prediction data and identify disruptive inflection points.

[0091] It should be noted that by analyzing the time series of the validated predicted data, the changing trends and key points in the data are identified. In time series analysis, inflection points refer to moments when the data trend undergoes a significant change, potentially indicating a shift in the disruption mode from one stage to another. By calculating the derivatives or slope changes of the data, abrupt changes in numerical values ​​are detected and marked as potential inflection points. Combined with the context and trend analysis of the data, the accuracy of these inflection points in reflecting changes in the disruption mode is further verified. Through this process, disruption inflection points are identified.

[0092] S4.4 Combine the verified prediction data with the failure inflection point to analyze the rate and direction of change of failure mode characteristics.

[0093] It should be noted that inflection points are marked based on time series data; an inflection point is the moment when the characteristics of the damage mode change. By validating and predicting the rate of change of data before and after the inflection point, the speed and direction of change of the damage mode characteristics can be analyzed. Specifically, the rate of change can be obtained by calculating the difference between consecutive data points and dividing by the time interval. The direction of change is determined by comparing the trend of data point values ​​before and after the inflection point. If the data value increases continuously before the inflection point and decreases sharply after the inflection point, the rate of change is negative, indicating a sharp slowdown in the damage mode. Conversely, if the data value decreases before the inflection point and increases sharply after the inflection point, the rate of change is positive, indicating an acceleration in the damage mode. Through these analyses, the rate and direction of change of the damage mode characteristics can be accurately identified.

[0094] S4.5. Based on the rate and direction of change, analyze the evolution of the damage mode characteristics, identify the points of change in damage progress, and obtain the damage development trend.

[0095] It should be noted that the rate of change at each time point is calculated by analyzing the changes in the characteristics of the damage mode over time. The rate of change is obtained by comparing the differences between adjacent data points and dividing by the time interval, while the direction of change is determined by judging the increase or decrease in the data trend. If the rate of change of the damage mode characteristics increases significantly, it indicates that the damage is accelerating, and vice versa. Based on the analysis of the rate and direction of change, the time points where the damage mode characteristics change drastically are identified; these points are the points of change in damage progress. By evaluating the data trends before and after the points of change, the evolution pattern of the damage mode can be obtained, thereby predicting the future development direction and extent of the damage mode. By comparing the rate of change within different time periods, the acceleration or deceleration trend of damage progress can be revealed, and the development trend of damage can be obtained.

[0096] S5. Based on the trend of damage development and real-time changes in environmental conditions, dynamically optimize and adjust the damage mode identification criteria, classify damage modes using the new damage mode identification criteria, and generate a damage mode identification report.

[0097] S5.1 Collect environmental condition data, perform data fusion and cleaning, analyze real-time environmental condition changes, combine real-time environmental condition changes with the trend of destructive development, and generate a real-time environmental dataset.

[0098] It should be noted that environmental condition data is collected, including factors such as temperature, humidity, air pressure, and wind speed. This data may influence the development of failure modes. Through data fusion and cleaning, environmental condition data from different sensors and data sources are integrated into a unified data format, removing noise, outliers, and missing data to ensure accuracy and consistency. By monitoring changes in environmental conditions (such as temperature, humidity, and air pressure) in real time, trend analysis methods (such as moving averages or smoothing techniques) can be used to observe fluctuations in environmental conditions over time. In the analysis, points of change refer to moments when environmental conditions undergo significant changes; these points may represent the significant impact of environmental factors on the development of failure modes. For example, sudden temperature changes may cause thermal expansion or contraction of materials, thus affecting structural stability and leading to changes in failure modes. By comparing changes in environmental conditions with the evolution of failure modes, the potential impact of environmental factors on failure modes can be identified. Based on this, the trends in environmental condition changes are correlated with the development trends of failure modes to assess the actual impact of environmental factors on failure modes and generate a real-time environmental dataset.

[0099] S5.2 Dynamically adjust the damage mode discrimination criteria based on real-time environmental data and generate new damage mode discrimination criteria.

[0100] It should be noted that trend analysis of real-time environmental data is performed by examining the correlation between environmental conditions and damage modes to identify key changes. These changes are then compared with the development trends of damage modes to calculate their correlation and assess the intensity of the impact of environmental changes on damage modes. A high correlation indicates that environmental conditions have a significant impact on damage modes. If the impact of certain environmental conditions on damage modes is found to be intensifying, the weights of various features in the discrimination criteria are adjusted according to the changes, so that the discrimination criteria can more accurately reflect the damage modes under current environmental conditions, thus generating new discrimination criteria for damage modes.

[0101] S5.2. Based on the new damage mode discrimination criteria, the damage mode features within each time window are re-matched and classified. By comparing the similarity between the damage mode features and the new discrimination criteria, damage mode labels are generated.

[0102] It should be noted that the damage mode features within each time window are compared one by one with the updated discrimination criteria, calculating the similarity between each damage mode feature and each category of features in the discrimination criteria. Damage mode features with high similarity will be classified into the corresponding damage mode type, while features with low similarity may be re-evaluated or labeled as types that do not meet the current discrimination criteria. In this way, the damage mode features within each time window will be reclassified according to the updated criteria, and damage mode labels will be generated based on the matching and classification results.

[0103] S5.3 Divide the destruction mode labels into different destruction mode types, and perform cross-validation and parameter adjustment based on the destruction mode labels to generate a destruction mode discrimination report.

[0104] It should be noted that, based on the characteristics of each category in the damage mode labels, the labels are divided into different damage mode types according to the criteria. Each type represents a specific damage mode. Cross-validation and parameter tuning are performed based on the damage mode labels. Cross-validation ensures generalization ability and evaluates the accuracy of each damage mode type by dividing the dataset into multiple subsets and selecting one subset for validation at a time. Based on the cross-validation results, relevant parameters are adjusted and the classification criteria are optimized to improve classification accuracy and generate a damage mode discrimination report.

[0105] In summary, this invention enhances the accuracy and diversity of simulation datasets through adaptive density clustering algorithms, convolutional neural networks, clustering analysis techniques, data fusion, feature extraction, and pattern classification. By employing intelligent steps in computer-aided engineering, it can automatically identify and classify different failure modes. Combined with time series analysis and dynamic optimization of discrimination criteria based on real-time environmental conditions, it improves the accuracy and reliability of failure mode prediction, providing effective technical support for failure detection and early warning in complex soil structures.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for determining the failure mode of reinforced soil retaining walls based on numerical simulation, characterized in that: include, Collect three-dimensional numerical simulation data to simulate the interaction between soil and reinforced materials, set boundary conditions and loading conditions, and generate a standard simulation dataset. The adaptive density clustering algorithm is used to identify the boundary conditions and loading methods of the working conditions from the standard simulation dataset to obtain the enhanced simulation dataset. The enhanced simulation dataset is then fused with the standard simulation dataset to generate an extended simulation dataset. Based on the extended simulation dataset, feature extraction is performed using a convolutional neural network to obtain damage mode features, and cluster analysis is used to identify different damage mode types according to the damage mode discrimination criteria. The evolution of damage mode characteristics and damage mode types at different time points is analyzed, and time series analysis methods are used to monitor the trend of characteristic changes and the transformation of mode types, so as to identify the turning point of damage and the trend of damage development. Based on the development trend of damage and changes in real-time environmental conditions, the criteria for identifying damage modes are dynamically optimized and adjusted, and damage modes are classified using the new criteria to generate a damage mode identification report.

2. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 1, characterized in that: The process involves collecting three-dimensional numerical simulation data to simulate the interaction between soil and reinforced materials, setting boundary conditions and loading scenarios, and generating a standard simulation dataset. The specific steps are as follows: Collect three-dimensional numerical simulation data, construct a three-dimensional structure of soil and reinforcement material by selecting the mechanical properties of soil and reinforcement material, and generate a structural contact dataset; Based on the structural contact dataset, the boundary conditions and loading conditions between the soil and the reinforcement material are set, and the loading condition dataset is generated. The loading condition dataset is calculated using the finite element method to obtain a calculation data package. Three-dimensional displacement field, stress field, and strain field data are extracted from the calculation data package. Through data cleaning, denoising, and standardization, a standard simulation dataset is generated.

3. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 2, characterized in that: The process of identifying the boundary conditions and loading methods of the operating conditions from the standard simulation dataset using an adaptive density clustering algorithm to obtain the enhanced simulation dataset involves the following specific steps. The standard simulation dataset is clustered using an adaptive density clustering algorithm, and boundary conditions and loading methods with similar characteristics are identified to generate a clustered dataset. Enhanced datasets are extracted from the clustered datasets, and random parameter adjustments and operating condition combinations are performed on each enhanced dataset to obtain an enhanced simulation dataset.

4. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 3, characterized in that: The specific steps for fusing the enhanced simulation dataset with the standard simulation dataset to generate an extended simulation dataset are as follows. The standard simulation dataset and the enhanced simulation dataset are merged, and a comprehensive simulation dataset is generated by matching similar working conditions and performing cross-validation. Physical consistency verification is performed on the integrated simulation set to obtain the verified integrated simulation set. Then, the verified integrated simulation set is denoised and data augmented to generate an extended simulation dataset.

5. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 4, characterized in that: The feature extraction based on the extended simulation dataset is performed using a convolutional neural network to obtain damage mode features, and the specific steps are as follows: The extended simulation dataset is input into a convolutional neural network for feature extraction, generating a high-dimensional feature dataset. Principal component analysis is used to reduce the dimensionality of high-dimensional feature datasets, and principal components are selected to further reduce data dimensionality, thereby generating destruction mode features.

6. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 1, characterized in that: The cluster analysis method is used to identify different damage mode types based on the damage mode discrimination criteria. The specific steps are as follows. The damage pattern features are clustered using the K-means clustering algorithm to obtain a damage pattern cluster dataset. The damage pattern features of each cluster in the damage pattern cluster dataset are then analyzed to identify the damage pattern category trend corresponding to the cluster. The trends of destruction patterns are compared one by one with the criteria for determining destruction patterns. The characteristics of destruction patterns that meet the criteria are classified into different destruction pattern types based on feature similarity and matching degree.

7. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 6, characterized in that: The evolution of damage mode characteristics and damage mode types at different time points, combined with time series analysis methods, is monitored to track the trends in characteristic changes and shifts in mode types. The specific steps are as follows: The destruction mode features and destruction mode types are divided into multiple time windows according to the time series, the time window dataset is obtained, the destruction mode features in the time window dataset are aggregated, the statistics of each feature are calculated, and the time series dataset is generated. Based on time series datasets, regression analysis is used to analyze the changing trends of damage mode characteristics over time, predict the future evolution of damage modes, obtain damage mode prediction data, and verify the damage mode prediction data through residual analysis to evaluate the difference between the prediction results and the actual situation, and generate verified prediction data.

8. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 1, characterized in that: The specific steps for identifying the turning point and development trend of destruction are as follows. Inflection point detection is performed on the verified prediction data to identify disruptive inflection points; By combining the verified prediction data with the failure inflection point, the rate and direction of change of the failure mode characteristics are analyzed. Based on the rate and direction of change, analyze the evolution of damage mode characteristics, identify points of change in damage progression, and obtain the trend of damage development.

9. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 8, characterized in that: The criteria for identifying damage modes are dynamically optimized and adjusted based on the trend of damage development and real-time changes in environmental conditions. The specific steps are as follows: Collect environmental condition data, perform data fusion and cleaning, analyze real-time environmental condition changes, combine real-time environmental condition changes with the trend of damage development, and generate a real-time environmental dataset. The criteria for determining damage modes are dynamically adjusted based on real-time environmental data, and new criteria for determining damage modes are generated.

10. The method for determining the failure mode of reinforced soil retaining walls based on numerical simulation as described in claim 9, characterized in that: The specific steps for classifying damage modes using the new damage mode discrimination criteria and generating a damage mode discrimination report are as follows. Based on the new damage mode discrimination criteria, the damage mode features within each time window are re-matched and classified. Damage mode labels are generated by comparing the similarity between the damage mode features and the new discrimination criteria. The destruction mode labels are used to classify different destruction mode types, and cross-validation and parameter adjustment are performed based on the destruction mode labels to generate a destruction mode discrimination report.

Images