Method for constructing marine clay-geogrid interface cyclic shear stress prediction model

Abstract

The application provides a marine clay-geogrid interface cyclic shear stress prediction model construction method, and relates to the technical field of marine engineering. The method comprises the following steps: preparing and assembling a marine clay-geogrid interface shear test sample; after the sample is assembled, a multi-working-condition interface cyclic shear test is carried out to obtain effective original test data; based on the effective original test data, an interface cyclic shear test database is constructed; based on the interface cyclic shear test database, a deep learning prediction model is constructed and trained; the accuracy of the prediction model is verified, the optimal prediction model is screened out, and a marine clay-geogrid interface cyclic shear stress prediction formula is built based on the optimal prediction model. The application can obtain reliable test data under multiple working conditions, accurately depict the shear stress evolution law by constructing a CNN-BiLSTM model, build an engineering usable prediction formula, and provide reliable support for the design and safety evaluation of marine engineering reinforced structures.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ocean engineering, and particularly relates to a method for constructing a marine clay-geogrid interface cyclic shear stress prediction model. BACKGROUND

[0002] In ocean engineering such as port wharf, seawall, coastal embankment and offshore wind power foundation, marine clay often becomes an adverse factor affecting the stability of foundation and the safety of structure due to its characteristics of high water content, high compressibility and low shear strength. In order to improve the bearing capacity and deformation control ability of marine clay foundation, geosynthetic materials such as geogrid are generally used for reinforcement in engineering, and the interface shear characteristics between marine clay and geogrid are directly related to the load transfer efficiency, reinforcement effect and long-term service performance of structure. The actual ocean engineering environment has significant complexity, and the geogrid reinforced structure is in a long-term cyclic loading state under the repeated action of waves, tides, traffic loads and superstructure, and its interface stress process presents typical cyclic shear characteristics. At the same time, the temperature of engineering environment presents fluctuation characteristics due to the influence of climate, seasonal change and seawater. Under the combined action of temperature change and cyclic load, the shear stress evolution process of marine clay-geogrid interface is significantly different from that under constant temperature and monotonic loading conditions, and presents complex mechanical characteristics such as accumulation, degradation and hysteresis, which are closely related to normal stress, shear displacement, loading path and temperature, thereby significantly increasing the difficulty of describing and predicting the interface shear behavior. Therefore, it is necessary to carry out special research on marine clay-geogrid interface under the condition of considering the coupling of temperature change and cyclic load, so as to systematically reveal the shear stress evolution law.

[0003] However, the existing method for obtaining and predicting the cyclic shear stress of marine clay-geogrid interface in the prior art has insufficient ability to describe the coupling of temperature effect, cyclic loading and multiple factors under complex working conditions, and the prediction model precision and stability are difficult to meet the actual demand of ocean engineering for the reliability of interface mechanical parameters.

[0004] In terms of physical experiments: Existing experimental studies on the mechanical properties of soil-geosynthetic interfaces are mostly conducted using conventional direct shear equipment under ambient temperature conditions. The test conditions are relatively simple and cannot accurately reflect the interfacial stress behavior under the coupled effects of temperature changes and cyclic loading in marine engineering. While some studies have introduced temperature control conditions, their control range is limited, and the moisture content of the samples is not effectively constrained. Fluctuations in soil moisture content during temperature changes can interfere with the interfacial shear response, affecting the stability and repeatability of the test results. Furthermore, existing interfacial shear testing equipment has low integration, making it difficult to simultaneously achieve wide-range temperature control, interfacial cyclic shear loading, and sample stability monitoring within a single device. This limits the ability to acquire high-quality interfacial cyclic shear test data under complex conditions. For marine clay-geogrid interfaces, research on cyclic shear tests considering temperature effects lacks experimental methods that can maintain stable sample moisture content over a wide temperature range and meet the requirements of interfacial cyclic shear loading. This, to some extent, restricts the engineering applicability of the relevant test data.

[0005] In terms of model prediction: Traditional empirical formulas and theoretical models based on simplified assumptions struggle to simultaneously consider the coupled effects of multiple factors such as temperature, shear displacement, and normal stress characteristics, resulting in limited accuracy in predicting interfacial shear stress under complex marine conditions. In recent years, deep learning methods have been increasingly applied to geotechnical engineering parameter prediction, with some models capable of fitting the nonlinear relationship of interfacial shear stress to a certain extent. However, most existing prediction models still do not fully utilize historical information during cyclic loading in their feature representation or model structure design, making it difficult to comprehensively characterize the cumulative and degenerative evolution of interfacial shear stress with increasing cycle count. Especially after introducing the temperature effect, the evolution path of interfacial shear stress is more significantly affected by the temperature-cycle coupling effect. Existing models still lack the ability to describe the long-term evolution behavior of interfacial shear stress under different temperature conditions, limiting the stability and reliability of prediction results under complex marine engineering conditions.

[0006] Current marine engineering demands increasingly higher accuracy and timeliness in cyclic shear stress data at the marine clay-geogrid interface. However, existing methods for acquiring and predicting cyclic shear stress at the interface remain limited in their applicability under varying temperature conditions, and their ability to characterize the evolution of interfacial shear stress with the number of cycles is insufficient to meet the needs of complex working condition analysis.

[0007] To address the aforementioned issues, it is necessary to construct a method for predicting interfacial shear stress that can fully integrate cyclic shear test data under multi-factor coupling and take into account both temperature effects and cyclic history characteristics, thereby providing a more reliable mechanical basis for the safety assessment of geogrid-reinforced structures in marine engineering. Summary of the Invention

[0008] To address the shortcomings of existing technologies in the study of cyclic shear stress at the marine clay-geogrid interface, such as insufficient coverage of experimental conditions, difficulty in effectively characterizing temperature effects and cyclic history, and limited accuracy and stability of prediction models, this invention proposes a method for constructing a prediction model for cyclic shear stress at the marine clay-geogrid interface. This method constructs a multi-factor interfacial cyclic shear test database and introduces a prediction model capable of simultaneously characterizing the nonlinear features of the interface and cyclic history information. This enables accurate prediction of the evolution of interfacial shear stress under the coupled effects of temperature changes and cyclic loading, thus providing reliable interfacial mechanical parameter support for the design optimization and long-term service safety assessment of geogrid-reinforced structures in marine engineering.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0010] A method for constructing a prediction model for cyclic shear stress at the marine clay-geogrid interface, the method comprising:

[0011] Preparation and assembly of marine clay-geogrid interface shear test specimens;

[0012] After the sample is assembled, a multi-condition interface cyclic shear test is conducted to obtain valid raw test data.

[0013] Based on the aforementioned valid original experimental data, an interface cyclic shear test database is constructed;

[0014] Based on the interface cyclic shearing test database, a deep learning prediction model is constructed and trained.

[0015] The accuracy of the prediction model was verified, the optimal prediction model was selected, and a prediction formula for cyclic shear stress at the marine clay-geogrid interface was built based on the optimal prediction model.

[0016] Compared with the prior art, the beneficial effects of the present invention are:

[0017] A customized, large-scale, temperature and humidity-controlled interfacial shear testing device is employed to achieve stable control over a wide temperature range from 0℃ to 70℃, ensuring uniform temperature distribution within the sample. Effective humidity control within the test chamber ensures stable moisture content of the marine clay sample under different temperature conditions, preventing interference with the interfacial shear mechanical response due to moisture content fluctuations. Furthermore, by incorporating normal stress loading conditions of 50kPa, 100kPa, and 150kPa, typical stress environments from shallow to deep marine foundations are covered. Compared to most existing testing methods that are only applicable to single operating conditions or have limited temperature control ranges and struggle to consider humidity stability, this invention can obtain more realistic and reliable interfacial cyclic shear test data.

[0018] This paper introduces key factors reflecting the characteristics of cyclic loading, such as the number of cyclic shear cycles, into the prediction of interfacial shear stress. A deep learning prediction method combining convolutional feature extraction and bidirectional temporal modeling (CNN-BiLSTM) is employed to characterize the evolution of interfacial shear stress during cyclic shearing. This method can simultaneously learn the nonlinear relationships between multiple input features and the temporal characteristics of shear stress changing with the number of cycles during cyclic shearing. Compared to prediction models using only convolutional neural networks (CNN), while a single CNN can effectively extract the nonlinear relationships between multiple input features, its ability to characterize the temporal dependence of shear stress evolution with the number of cycles during cyclic shearing is limited, making it difficult to fully reflect the accumulation and degradation patterns of interfacial shear stress. Conversely, compared to prediction models using only bidirectional long short-term memory networks (BiLSTM), while a single BiLSTM has strong temporal modeling capabilities, its ability to automatically extract complex features under multi-factor input conditions is relatively insufficient, and it is easily affected by the feature representation method. In contrast, the CNN-BiLSTM fusion prediction method adopted in this invention fully combines the advantages of CNN in multidimensional feature extraction with the temporal modeling capability of BiLSTM, which can more comprehensively characterize the evolution behavior of interface shear stress under the combined action of multi-factor coupling and cyclic shearing, thereby obtaining more stable and reliable prediction results under complex working conditions.

[0019] Based on cyclic shear tests of marine clay-geogrid interfaces under multiple temperature and normal stress conditions, a standardized cyclic shear database of interfaces was constructed by integrating a large amount of effective experimental data. A reasonable data partitioning method ensured the balanced distribution of data under different temperature conditions, normal stress levels, and cyclic stages. The deep learning prediction model built based on this database exhibited good predictive consistency and stability under various working conditions, and could achieve reliable predictions even under complex working conditions without prior training. This method can provide data support for the interfacial mechanical analysis of geogrid-reinforced structures in cold regions, warm and high-temperature marine environments, assisting in interface parameter selection, structural design, and safety assessment. It solves the problems of insufficient generalization ability and difficulty in adapting to complex marine engineering conditions inherent in traditional empirical models.

[0020] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0021] Figure 1 This is a flowchart of a method for constructing a cyclic shear stress prediction model for the marine clay-geogrid interface according to the present invention;

[0022] Figure 2This is a characterization diagram of a marine clay sample according to the present invention, which includes two sub-diagrams, showing the macroscopic appearance of the marine clay sample and its microscopic morphological features under a scanning electron microscope (SEM).

[0023] Figure 3 This is a material characterization diagram of PP bidirectional geogrid according to the present invention;

[0024] Figure 4 This is a schematic diagram of the customized large-size temperature and humidity controllable interface shear test device according to the present invention;

[0025] Figure 5 This is a cyclic shear stress-shear displacement curve of the marine clay-geogrid interface according to the present invention;

[0026] Figure 6 This is a comparison chart of the prediction accuracy of the training sets of various models according to the present invention;

[0027] Figure 7 This is a comparison chart of the prediction accuracy of each model on the validation set according to the present invention. Detailed Implementation

[0028] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings, so as to more clearly understand the purpose, features and advantages of this invention. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of this invention, but are only for illustrating the essential spirit of the technical solutions of this invention. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0029] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0030] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0031] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to mean “and / or” unless otherwise expressly stated herein.

[0032] In the following description, in order to clearly demonstrate the structure and working method of the present invention, a number of directional terms will be used. However, terms such as "front", "back", "left", "right", "outside", "inside", "outward", "inward", "up", and "down" should be understood as convenient terms and not as limiting terms.

[0033] The implementation details of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The following content is only for the convenience of understanding the implementation details and is not necessary for implementing this solution.

[0034] To address the shortcomings of existing technologies in the study of cyclic shear stress at the marine clay-geogrid interface, such as insufficient coverage of experimental conditions, difficulty in effectively characterizing temperature effects and cyclic history, and limited accuracy and stability of prediction models, this invention proposes a method for constructing a cyclic shear stress prediction model for the marine clay-geogrid interface based on CNN-BiLSTM. This method constructs a multi-factor interface cyclic shear test database and introduces a prediction model that can simultaneously characterize the nonlinear features of the interface and cyclic history information. This enables accurate prediction of the evolution law of interface shear stress under the coupled action of temperature changes and cyclic loads, thereby providing reliable interface mechanical parameter support for the design optimization and long-term service safety assessment of geogrid-reinforced structures in marine engineering.

[0035] Example 1

[0036] like Figure 1 As shown, this embodiment provides a method for constructing a cyclic shear stress prediction model for the marine clay-geogrid interface based on CNN-BiLSTM, including the following steps:

[0037] S1: Preparation and assembly of marine clay-geogrid interface shear test specimens;

[0038] Marine clay, commonly used in marine engineering, was selected to prepare marine clay samples, and geogrid was chosen as the reinforcement material. The geogrid samples were directly laid on top of the lower shear box within the test chamber of a customized large-size temperature and humidity-controlled interface shear testing device, and then the upper shear box was placed on top of the lower shear box. The marine clay samples were then layered and compacted into the upper shear box. Sufficient contact between the marine clay and the geogrid was ensured to form a stable soil-reinforcement interface.

[0039] Specifically, such as Figures 2-3As shown, the marine clay samples were taken from representative soil samples obtained during foundation exploration in the project area. Before sample preparation, the undisturbed marine clay was manually sorted using a combination of manual sieving and visual inspection to remove non-soil impurities such as plant roots and shell fragments, ensuring the homogeneity and representativeness of the samples. The target moisture content of the marine clay samples was then controlled at 35%. Moisture content adjustment was performed using a quality control method, calculating the required amount of water to be added or removed based on the initial moisture content of the undisturbed soil sample. When the moisture content was too low, water was gradually and evenly added and thoroughly mixed; when it was too high, natural air drying was used at room temperature. After preparation, the marine clay samples were placed in a sealed container for static curing, allowing moisture to fully migrate within the soil and reach a homogeneous and stable state. The upper shear box is then placed on top of the lower shear box. The clay sample is then filled into the upper shear box in 5 layers using a layered compaction method. Each layer is about 24 mm high, and the total height is about 120 mm. After each layer is filled, compaction and leveling are carried out to meet the requirements of the interface cyclic shear test.

[0040] For the preparation of geogrid samples, polypropylene (PP) biaxial geogrid, commonly used in engineering, was selected as the reinforcement material. According to the dimensional requirements of the interface direct shear test apparatus, the geogrid was cut to a specific size, ensuring the integrity of the geogrid ribs and preventing tension, bending, or damage during the cutting process. Before assembling the geogrid samples, a visual inspection was performed on the cut geogrid samples to ensure their surface was clean, their structure was intact, and their original mechanical properties remained unaffected. The geogrid samples were then laid directly on top of the lower shear box, ensuring their laying direction was consistent with the shear direction, thus realistically simulating the working state of the geogrid under shear conditions in engineering projects.

[0041] Reference Figure 4 The customized large-size temperature and humidity controllable interface shear test device includes: a test chamber, an upper shear box and a lower shear box disposed inside the test chamber, a normal loading system, and a horizontal displacement loading system. The normal loading system is located above the test chamber and includes a normal pressure bar for applying normal stress to the sample inside the upper shear box via a rigid pressure plate. The horizontal displacement loading system includes a horizontal push-pull rod and a horizontal fixing rod; during the test, the lower shear box is connected to the horizontal push-pull rod and is driven by the horizontal push-pull rod to generate horizontal displacement; the upper shear box is connected to the horizontal fixing rod to maintain a fixed horizontal position. The test chamber has temperature and humidity control functions and can accommodate the above-mentioned shear components and maintain the set environmental conditions.

[0042] S2: After the sample assembly is completed, a multi-condition interface cyclic shear test is carried out.

[0043] After the sample is assembled, the lower shear box is connected to the horizontal push-pull rod to induce horizontal displacement during the test; simultaneously, the upper shear box is connected to the horizontal fixing rod to ensure that the upper shear box remains fixed during the test. A rigid bearing plate is placed on top of the upper shear box to achieve uniform application of normal stress.

[0044] Next, using the customized large-size temperature and humidity controllable interface shear test device, the equipment debugging, sample temperature and humidity control, normal stress application and consolidation, and cyclic shear loading and data acquisition operations were completed in sequence to obtain cyclic shear mechanical data of marine clay-geogrid interface under different temperature and normal stress conditions.

[0045] Specifically, after the sample assembly is completed, the test chamber door is closed, and the required test temperatures are set (0℃, 10℃, 20℃, 30℃, 40℃, 50℃, 60℃, 70℃). The humidity inside the test chamber is simultaneously adjusted to ensure the moisture content of the clay sample remains stable during temperature changes. Real-time data feedback is obtained from temperature and humidity sensors embedded inside the marine clay sample. The temperature control system is then activated to perform a 2-hour isothermal pretreatment, ensuring that the temperature difference at each measuring point inside the sample is strictly controlled within ±1℃, thereby achieving overall thermal equilibrium. During this process, the device precisely adjusts the humidity inside the chamber, utilizing the dynamic balance principle of vapor partial pressure and sample moisture migration to stably constrain the moisture content of the marine clay sample to a preset state of 35%. This effectively eliminates the interference of moisture fluctuations caused by sample dehydration or condensation under wide temperature range conditions on the interfacial cyclic shear mechanical response. After the test program is started, a normal stress of 50 kPa, 100 kPa, or 150 kPa is applied to the normal loading system by a servo motor, and consolidation is carried out for 2 hours under this stress condition. After consolidation, the servo motor drives the horizontal push-pull rod to push the lower shear box 6 mm to the right at a shear rate of 1 mm / min, and then pulls it 6 mm to the left. The rightward and leftward displacements constitute a complete cycle, and a total of 10 cycles are performed. The test is terminated after the cycle is completed. During the test, the customized large-size temperature and humidity controllable interface shear test device synchronously and automatically collects real-time data such as cyclic shear stress, shear displacement, interface shear stress, normal displacement, and normal stress. The test is repeated 3 times under each working condition. After removing abnormal deviation data, the average value is taken as the valid original test data for that working condition. The role of the effective original experimental data is twofold: firstly, to eliminate random errors and instrument disturbances in a single measurement through multiple sets of parallel experiments, ensuring the repeatability of the experimental data; secondly, to extract feature data that can truly represent the evolution law of interface mechanical response under the working condition through mean value processing, thereby providing a high signal-to-noise ratio training source for the subsequent construction of deep learning models.

[0046] S3: Construct an interface cyclic shear test database based on valid original experimental data;

[0047] Based on the interface cyclic shear mechanics data, the collected valid raw test data are screened, organized and calibrated, the training set and test set are divided according to a set ratio, the model input parameters and output parameters are determined, and each parameter is normalized to eliminate the influence of differences in different dimensions and value ranges, thereby constructing a standardized marine clay-geogrid interface cyclic shear test database.

[0048] Specifically, after completing the multi-condition marine clay-geogrid interface cyclic shear test, the effective raw experimental data collected during the test were systematically organized and processed to construct an interface cyclic shear test database. During database construction, the effective raw experimental data were screened to remove contact nonlinear interference in the initial loading stage and mechanical hysteresis effects during cyclic switching transients. The cyclic shear process data were also reconstructed in segments to ensure the physical rationality and continuity of the data. Based on this, and considering the engineering characteristics of the interface cyclic shear behavior, representative data were extracted and reorganized to form multiple sets of effective sample data for subsequent prediction model construction and training. Based on the above effective sample data, a multi-dimensional feature database was constructed with temperature, normal stress, shear displacement, and number of cycles as input parameters and interface cyclic shear stress as the output parameter. The input and output data were uniformly normalized to reduce the impact of different dimensions on model training. Subsequently, the database was randomly divided into training and test sets according to a certain proportion, with the training set used for model training and the test set used for model generalization performance verification. The interface cyclic shear test database constructed through the above process has a good foundation in terms of sample size, representativeness, and structural integrity, and can provide reliable data support for deep learning models to characterize the evolution of interface shear stress in the cyclic shear process.

[0049] S4: Based on the interface cyclic shearing test database, construct and train a deep learning prediction model;

[0050] Based on the interface cyclic shear database, deep neural network (DNN) models, convolutional neural network (CNN) models, bidirectional long short-term memory network (BiLSTM) models, and a fusion model of convolutional feature extraction and bidirectional temporal modeling (CNN-BiLSTM) were constructed respectively. The training set was used to train each model to predict the cyclic shear stress at the marine clay-geogrid interface.

[0051] Specifically, the DNN model adopts a fully connected network structure of "5–128–64–32–1". The input parameters are temperature, normal stress, shear displacement, normal displacement and number of iterations. The output parameter is the interface cyclic shear stress. The training uses the Adam optimization algorithm with a learning rate of 0.001.

[0052] The CNN model adopts a one-dimensional convolutional network structure of "5–64–32–64–1", with a kernel size of 3 and an activation function of ReLU. Training uses the AdamW optimizer combined with cosine annealing learning rate scheduling.

[0053] The BiLSTM model adopts a temporal network structure of "(32×5)–128–64–1", takes a five-dimensional time series of length 32 as input, and uses the Adam optimization algorithm for training with a learning rate of 0.001.

[0054] The CNN–BiLSTM fusion model adopts a network structure of “(32×5)–64–32–BiLSTM(128)–64–1”, combining convolutional feature extraction and bidirectional temporal modeling. During training, a random dropout layer with a dropout rate of 0.30 is set after the BiLSTM layer to randomly drop some neuron connections. This CNN–BiLSTM fusion model is a hybrid model that combines the local spatial feature extraction capability of CNN with the global temporal dependency capture capability of BiLSTM.

[0055] The CNN–BiLSTM fusion model employs a sliding window technique, using a five-dimensional feature vector with 10 time steps, containing temperature, normal stress, shear displacement, normal displacement, and cycle count as temporal input. Internally, the model integrates a one-dimensional convolutional layer (1D-CNN) with a kernel size of 3 and 64 output channels to initially extract spatial correlations and local nonlinear features among the multi-dimensional input features. Subsequently, a three-layer, 128-hidden-unit bidirectional long short-term memory (BiLSTM) network is used to deeply capture the long-term memory and complex dependencies of interface shear stress during cyclic loading from both positive and negative temporal dimensions. For training, the CNN–BiLSTM model uses the Adam optimization algorithm combined with a 1×10⁻⁶ kcal / m² algorithm. -5 The weights are decayed, and a dropout rate of 0.3 is set to prevent overfitting. An adaptive learning rate scheduler is introduced to ensure that the model has excellent stability when it reaches the optimal convergence state.

[0056] This CNN–BiLSTM fusion model achieves deep decoupling between multi-source environmental factors and cyclical historical features. Firstly, leveraging the local perceptual capabilities of 1D-CNN, and referencing… Figure 5 The model can automatically learn the nonlinear intervention law of interface shear strength when the temperature fluctuates between 0℃ and 70℃, overcoming the limitation of traditional empirical formulas in accurately representing the coupling effect of multiple factors. Secondly, the BiLSTM structure significantly enhances the model's ability to capture the "historical information" of the interface, and can accurately characterize the mechanical behavior such as the accumulation, degradation and hysteresis of shear stress as the number of cycles increases.

[0057] S5: Verify the accuracy of the prediction model, select the optimal prediction model, and build a prediction formula for cyclic shear stress at the marine clay-geogrid interface based on the optimal prediction model.

[0058] By comparing the prediction results of different models on the test set with the measured data, the prediction performance of the four types of models—DNN, CNN, BiLSTM, and CNN-BiLSTM—was evaluated using the correlation coefficient (R²). 2 The root mean square error (RMSE) and mean absolute percentage error (MAPE) were evaluated to screen and determine the model with the best prediction accuracy and stability as the final prediction model for cyclic shear stress at the marine clay-geogrid interface. The optimal prediction model was the CNN-BiLSTM model. Based on the optimal prediction model, a prediction formula for cyclic shear stress at the marine clay-geogrid interface was constructed.

[0059] Specifically, the generalization ability of the model is verified using a test set. The R-value of the CNN–BiLSTM model test set is [missing information]. 2 The deviations between the predicted and measured values ​​of RMSE and MAPE were the smallest among the four models. When extreme working conditions were selected for verification, the deviations between the predicted and measured values ​​of the cyclic shear stress at the lower interface were the smallest among the four models, meeting the engineering accuracy requirements. Therefore, the CNN-BiLSTM model with the best prediction accuracy and stability was finally selected and determined as the prediction model for the cyclic shear stress at the marine clay-geogrid interface.

[0060] Next, based on the coupling law of "temperature-cycle shear number-normal stress-normal displacement-shear displacement-shear stress" of the optimal prediction model CNN-BiLSTM, a prediction formula for cyclic shear stress at the marine clay-geogrid interface is constructed. The prediction formula is as follows:

[0061]

[0062] In the formula, Corresponding to cyclic shear stress, Corresponding interface normal stress (characterizing the level of interface normal constraint). Corresponding interface shear displacement (reflecting the degree of interface shear deformation). The normal displacement of the interface during shearing (characterizing the interface structure adjustment and deformation characteristics), T corresponds to the temperature of the interface, N corresponds to the number of cyclic shearing cycles experienced by the interface (used to characterize the cyclic loading history), and a i The corresponding interface shear response weighting coefficients are as follows: b0 corresponds to the interface baseline cyclic degradation coefficient, b1 corresponds to the interface temperature-related degradation coefficient, and c0 corresponds to the residual shear stress term. Specific values ​​are as follows:

[0063] a0=1.204; a1=0.237; a2=-1.009; a3=7.008; a4=0.0278; a5=0.00777; b0=0.182; b1=0.00443; c0=0.96.

[0064] This prediction formula comprehensively considers normal constraints, interface deformation characteristics, and temperature-cycle degradation effects. It can be directly used for engineering calculations and rapid evaluation of cyclic shear stress at the marine clay-geogrid interface without the need for a deep learning model. Further sensitivity analysis using a CNN-BiLSTM model shows that temperature and cycle number account for over 50% of the shear stress prediction, making them the main controlling factors for the interface's cyclic shear performance. Based on this principle, targeted parameter control suggestions can be provided for engineering design: In cold or high-temperature service environments, priority should be given to reducing temperature fluctuations by controlling the construction season, setting up insulation or heat insulation layers, and optimizing the foundation cover structure, thereby reducing the adverse effects of temperature changes on interface shear performance; Under engineering conditions with significant long-term cyclic loading, measures such as reasonably controlling traffic load levels, optimizing the arrangement of reinforcement layers, and increasing the number of reinforcement layers should be taken to reduce the effective cyclic shear amplitude per unit interface, thereby slowing down the cumulative degradation process of interface shear stress; At the same time, in the design stage, the prediction formula can be combined to conduct comparative analysis of different temperatures and cyclic conditions, and the combination of structural parameters with higher stability of interface shear performance can be selected to achieve synergistic optimization of interface cyclic shear performance and engineering durability.

[0065] This embodiment provides a method for constructing a prediction model for cyclic shear stress at the marine clay-geogrid interface. The core steps include: preparing marine clay and PP bidirectional geogrid samples and assembling them in a customized large-size temperature and humidity-controlled interface shear test device; conducting cyclic shear tests in a wide temperature range of 0℃-70℃ and under multiple normal stresses of 50kPa-150kPa; collecting effective data to construct an interface cyclic shear test database containing parameters such as temperature, normal stress, and shear displacement; constructing and training four types of deep learning models: DNN, CNN, BiLSTM, and CNN-BiLSTM; finally selecting the optimal CNN-BiLSTM fusion model; and building a prediction formula that can be directly applied in engineering based on this model. It achieves stable control over a wide temperature range and constant sample moisture content, and the obtained experimental data is authentic and reliable. The CNN-BiLSTM fusion model fully integrates the advantages of CNN in multidimensional feature extraction and the temporal modeling capabilities of BiLSTM, which can accurately characterize the accumulation and degradation law of interfacial shear stress under temperature-cyclic loading coupling. Its prediction accuracy and stability far exceed those of other models. The constructed prediction formula can be quickly calculated without the need for a deep learning model, providing reliable interfacial mechanical parameters for the design optimization and long-term service safety assessment of geogrid reinforced structures in marine engineering.

[0066] Example 2

[0067] This embodiment takes the soft soil foundation reinforcement project of the rear storage yard of a large port project in southeastern coastal my country as the research object. It addresses the prediction problem of interfacial shear mechanical parameters under the condition of geogrid reinforcement in marine clay foundations, and verifies the applicability and effectiveness of the method for constructing a prediction model of cyclic shear stress at the marine clay-geogrid interface considering the influence of temperature. The specific steps are as follows:

[0068] Step 1: Determination of experimental parameters and sample preparation

[0069] Experimental parameters were determined as follows: Geological exploration and physical property investigation were conducted on the foundation of the project area. In-situ exploration and indoor geotechnical tests confirmed that the foundation soil in this area is mainly marine clay with a particle size of less than 0.005 mm. Considering the climate conditions and service environment of the foundation in the project area, it was determined that there are significant environmental temperature variations during actual service. The temperature range of 0℃ to 70℃ was selected as the temperature range for subsequent tests. Simultaneously, the moisture content of the collected marine clay samples was tested. The results showed that the actual moisture content of the foundation soil was 35%, and this measured moisture content was used as the control basis for sample preparation to ensure that the prepared samples accurately reflect the physical state of the foundation soil. Furthermore, considering the overburden and surcharge weight borne by the geogrid reinforcement layer in the port ballast yard foundation, 50 kPa, 100 kPa, and 150 kPa were selected as the normal stresses for the interface cyclic shear test, respectively, to characterize the typical stress states under shallow, medium-depth, and deep foundation conditions.

[0070] Sample Preparation: PP biaxial geogrid, commonly used in engineering, was selected as the reinforcement material. According to the size requirements of the interface direct shear test apparatus, the geogrid was cut to a size of 560mm × 280mm. During the cutting process, the geogrid ribs were ensured to be intact, without tension, bending, or damage. Before sample assembly, the cut geogrid was visually inspected to ensure its surface was clean, its structure was intact, and its original mechanical properties were not affected. The geogrid sample was then laid directly on top of the lower shear box, ensuring its laying direction was consistent with the shear direction, so as to realistically simulate the working state of the geogrid under shear conditions in engineering.

[0071] Marine clay samples were taken from representative soil samples obtained during foundation exploration in the project area. Before sample preparation, the undisturbed marine clay was manually sorted using a combination of manual sieving and visual inspection to remove non-soil impurities such as plant roots and shell fragments, ensuring the homogeneity and representativeness of the samples. The target moisture content of the marine clay samples was then controlled at 35%. Moisture content adjustment was performed using a quality control method, calculating the required amount of water to be added or removed based on the initial moisture content of the undisturbed soil sample. When the moisture content was too low, water was gradually and evenly added and thoroughly mixed; when it was too high, natural air drying was used at room temperature. After preparation, the clay was placed in a sealed container for static curing, allowing moisture to fully migrate within the soil and reach a homogeneous and stable state. The upper shear box is then placed on top of the lower shear box. The clay sample is then filled into the upper shear box in 5 layers using a layered compaction method. Each layer is about 24 mm high, and the total height is about 120 mm. After each layer is filled, compaction and leveling are carried out to meet the requirements of the interface cyclic shear test.

[0072] Step 2: Multi-condition interface cyclic shear test

[0073] After the sample is assembled, connect the lower shear box to the horizontal push-pull rod to allow it to move horizontally during the test. Simultaneously, connect the upper shear box to the horizontal fixing rod to ensure it remains fixed during the test. Then, place a rigid bearing plate on top of the upper shear box to ensure that the normal stress is applied evenly to the clay sample. Close the test chamber door and set the required temperature conditions (0℃, 10℃, 20℃, 30℃, 40℃, 50℃, 60℃, 70℃), and simultaneously adjust the humidity inside the test chamber to ensure that the moisture content of the clay sample remains stable during temperature changes. After the test program starts, the servo motor drives the normal loading system to apply a normal stress of 50kPa, 100kPa, or 150kPa, and perform consolidation for 2 hours under this stress condition. After consolidation, the servo motor drives the horizontal push-pull rod to push the lower shear box 6mm to the right at a shear rate of 1mm / min, then pull it 6mm to the left. This rightward and leftward displacement constitutes a complete cycle, which is performed 10 times. The test terminates after each cycle. During the test, the equipment automatically and synchronously collected real-time data such as cyclic shear stress, shear displacement, interfacial shear stress, normal displacement, and normal stress. The test was repeated three times under each working condition, and the average value after removing abnormal deviation data was taken as the valid test result for that working condition.

[0074] Step 3: Construction of the Interface Cyclic Shear Test Database

[0075] After completing multi-condition marine clay-geogrid interface cyclic shear tests, the raw data collected during the tests were systematically organized and processed to construct an interface cyclic shear test database. During database construction, the raw data was first screened and cleaned, removing abnormal data from the initial loading stage, cycle switching stage, and occasional disturbances. The cyclic shear process data was then reconstructed in segments to ensure the physical rationality and continuity of the data. Based on this, and considering the engineering characteristics of interface cyclic shear behavior, representative high-frequency data were extracted and reorganized, ultimately forming 2800 sets of valid sample data for subsequent prediction model construction and training. Based on these valid sample data, a multi-dimensional feature database was constructed with temperature, normal stress, shear displacement, and cycle number as input parameters and interface cyclic shear stress as the output parameter. The input and output data were uniformly normalized to reduce the impact of different dimensions on model training. Subsequently, the database was randomly divided into a training set (2300 sets) and a test set (500 sets) at an 8:2 ratio. The training set was used for model training, and the test set was used for model generalization performance verification. The interface cyclic shear test database constructed through the above process has a good foundation in terms of sample size, representativeness, and structural integrity, and can provide reliable data support for deep learning models to characterize the evolution of interface shear stress in the cyclic shear process.

[0076] Step 4: Building and Training Deep Learning Prediction Model

[0077] Based on the interface cyclic shear test database constructed in step 3, four models, namely DNN, CNN, BiLSTM and CNN–BiLSTM, are used to predict the cyclic shear stress at the marine clay-geogrid interface.

[0078] The DNN model adopts a fully connected network structure of "5–128–64–32–1". The input parameters are temperature, normal stress, shear displacement, normal displacement and number of iterations. The output is the interface cyclic shear stress. The training uses the Adam optimization algorithm with a learning rate of 0.001.

[0079] The CNN model adopts a one-dimensional convolutional network structure of "5–64–32–64–1", with a kernel size of 3 and an activation function of ReLU. Training uses the AdamW optimizer combined with cosine annealing learning rate scheduling.

[0080] The BiLSTM model adopts a temporal network structure of "(32×5)–128–64–1", takes a five-dimensional time series of length 32 as input, and uses the Adam optimization algorithm for training with a learning rate of 0.001.

[0081] The CNN–BiLSTM fusion model adopts a network structure of “(32×5)–64–32–BiLSTM(128)–64–1”, which combines convolutional feature extraction and bidirectional temporal modeling. During training, a random dropout layer with a dropout rate of 0.30 is set after the BiLSTM layer to randomly drop some neuron connections.

[0082] Training results show that the CNN–BiLSTM fusion model training set has RMSE=2.4241, MAPE=13.35%, and R0. 2 =0.983, the best fit (see Figure 6 This is far superior to the unoptimized CNN (training set RMSE=6.9895, MAPE=24.09%).

[0083] Table 1. Index Results of the Four Models

[0084]

[0085] Step 5: Validation of Prediction Model Accuracy

[0086] The generalization ability of the model was verified using a test set (500 sets). The CNN–BiLSTM model test set showed RMSE=4.3, MAPE=13%, and R0. 2 =0.97, the predicted value deviates from the measured value by ≤13% (see Figure 7 Extreme working conditions were selected for verification: at 70℃ and 150kPa, the predicted value of the interfacial cyclic shear stress was 67.1kPa, and the measured value was 64.4kPa, with a deviation of 4.2%; at –0℃ and 150kPa, the predicted value of the interfacial cyclic shear stress was 69.5kPa, and the measured value was 65.5kPa, with a deviation of 5.8%, which met the engineering accuracy requirements.

[0087] This embodiment takes a soft soil foundation reinforcement project in the rear storage yard of a large port project in southeastern coastal my country as the research object to verify the applicability and effectiveness of the proposed prediction model construction method. The core steps include: determining experimental parameters that fit the actual working conditions, such as a temperature range of 0℃-70℃, a moisture content of 35%, and a normal stress of 50kPa-150kPa, based on on-site exploration; preparing and assembling PP biaxial geogrid and marine clay samples; conducting multi-condition cyclic shear tests using a customized temperature and humidity control device; and constructing the model after collecting effective data. A database of cyclic shear tests on interfaces containing 2800 samples (divided into training and test sets in an 8:2 ratio) was established. Four types of models—DNN, CNN, BiLSTM, and CNN-BiLSTM—were constructed and trained. The CNN-BiLSTM fusion model performed best, with RMSE=2.4241, MAPE=13.35%, and R²=0.983 on the training set, and RMSE=4.3, MAPE=13%, and R²=0.97 on the test set. The prediction bias was only 4.2%-5.8% under extreme conditions. By replicating actual engineering conditions, the engineering applicability of the prediction model was fully verified. The CNN-BiLSTM fusion model maintained high prediction accuracy and stability even with the support of real engineering data, meeting the reliability requirements of marine engineering for interface mechanical parameters. This provides a directly applicable practical case for predicting interface shear stress in similar marine clay-geogrid reinforced structures such as ports and storage yards, contributing to engineering design optimization and safety assessment.

[0088] This invention discloses a method for constructing a cyclic shear stress prediction model for the marine clay-geogrid interface, relating to the field of marine engineering technology. The core of this method is to solve the problem of predicting interface shear stress under complex working conditions through a "customized experiment + deep learning fusion model." The specific process is as follows: Marine clay and PP biaxial geogrid samples are prepared. A customized interface shear test device with a wide temperature range of 0℃-70℃ and controllable humidity is used to conduct cyclic shear tests under multiple normal stresses of 50kPa-150kPa. Effective data is collected to construct a standardized database containing parameters such as temperature, normal stress, and shear displacement. Subsequently, four types of models—DNN, CNN, BiLSTM, and CNN-BiLSTM—are constructed and trained. The optimal model, which integrates the advantages of CNN's multi-dimensional feature extraction and BiLSTM's temporal modeling capabilities, is selected. Finally, a prediction formula that can be directly applied in engineering is built based on this model. It achieves stable control over a wide temperature range and constant sample moisture content, and the acquired data is authentic and reliable. The CNN-BiLSTM fusion model can accurately characterize the accumulation and degradation law of interfacial shear stress under temperature-cyclic loading coupling, and the prediction accuracy (R²=0.983 for training set and R²=0.97 for test set) and stability far exceed those of traditional models. The constructed prediction formula can be calculated quickly without the need for a deep learning model. At the same time, the applicability of the method is verified through engineering examples. It provides reliable interfacial mechanical parameter support for the design optimization, safety assessment and long-term service performance guarantee of geogrid reinforced structures in marine engineering such as ports and seawalls, and solves the problems of insufficient coverage of working conditions, weak ability to characterize multiple factors coupling and limited prediction accuracy of existing technologies.

[0089] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the invention by those skilled in the art without departing from the spirit and essence of the invention. Such modifications or substitutions should all fall within the scope of the invention, or any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the invention should be covered within the protection scope of the invention. Therefore, the protection scope of the invention should be determined by the scope of the claims.

Claims

1. A method for constructing a prediction model for cyclic shear stress at the marine clay-geogrid interface, characterized in that, The method includes: Preparation and assembly of marine clay-geogrid interface shear test specimens; After the sample is assembled, a multi-condition interface cyclic shear test is conducted to obtain valid raw test data. Based on the aforementioned valid original experimental data, an interface cyclic shear test database is constructed; Based on the interface cyclic shearing test database, a deep learning prediction model is constructed and trained. The accuracy of the prediction model was verified, the optimal prediction model was selected, and a prediction formula for cyclic shear stress at the marine clay-geogrid interface was constructed based on the optimal prediction model; wherein, After the sample is assembled, a multi-condition interface cyclic shear test is conducted to obtain valid raw test data, including: After the sample assembly is completed, the test chamber door is closed, and different temperature conditions required for the test are set. The humidity inside the test chamber is simultaneously adjusted to ensure that the moisture content of the clay sample remains stable during temperature changes. Real-time data feedback is obtained through temperature and humidity sensors embedded inside the marine clay sample. The temperature control system is then activated to perform a period of constant temperature pretreatment, ensuring that the temperature difference at each measuring point inside the sample is controlled within ±1℃. During this process, the moisture content of the marine clay sample is stably constrained to a preset state of 35%. After the test program is started, a servo motor drives the normal loading system to apply normal stress, and the test is conducted under this stress condition for a period of time. Consolidation is performed between the layers; after consolidation, the servo motor drives the horizontal push-pull rod to push the lower shear box to the right at a fixed shear rate, and then pulls it to the left by a displacement. The rightward and leftward displacements constitute a complete cycle, and multiple cycles are performed. The test ends after the cycle ends. During the test, the customized large-size temperature and humidity controllable interface shear test device synchronously and automatically collects real-time data including cyclic shear stress, shear displacement, interface shear stress, normal displacement, and normal stress. The test is repeated multiple times under each working condition. After removing abnormal deviation data in the real-time data, the average value is taken as the valid original test data for that working condition. Furthermore, the formula for predicting cyclic shear stress at the marine clay-geogrid interface based on the optimal prediction model is as follows: In the formula, Corresponding cyclic shear stress; The corresponding interface normal stress characterizes the level of interface normal constraint. The corresponding interface shear displacement reflects the degree of interface shear deformation; The normal displacement generated at the interface during shearing represents the interface structure adjustment and deformation characteristics; T corresponds to the temperature of the interface; N corresponds to the number of cyclic shearing cycles experienced by the interface, used to represent the cyclic loading history; a i The corresponding interface shear response weighting coefficient; b0 corresponds to the interface baseline cyclic degradation coefficient; b1 corresponds to the interface temperature-related degradation coefficient; c0 corresponds to the residual shear stress term; the specific values ​​are as follows: a0=1.204; a1=0.237; a2=-1.009; a3=7.008; a4=0.0278; a5=0.00777; b0=0.182; b1=0.00443; c0=0.96; This prediction formula comprehensively considers normal constraints, interface deformation characteristics, and temperature-cyclic degradation effects, and can be directly used for engineering calculation and rapid evaluation of cyclic shear stress at the marine clay-geogrid interface.

2. The method according to claim 1, characterized in that, The preparation and assembly of the marine clay-geogrid interface shear test specimen includes: First, the undisturbed marine clay was subjected to a combination of manual sieving and visual inspection to remove non-soil impurities. Then, the target moisture content of the marine clay sample was controlled to 35%. Next, the marine clay sample was placed in a sealed container for static curing. After that, the upper shear box in the customized large-size temperature and humidity controllable interface shear test chamber was placed on top of the lower shear box. The marine clay sample was filled in layers in the upper shear box using a layered compaction method. After each layer was filled, it was compacted and leveled. For the preparation of geogrid samples, polypropylene (PP) biaxial geogrid was selected as the reinforcement material. According to the size requirements of the interface direct shear test device, the geogrid was cut into a certain size. Before assembling the geogrid samples, the cut geogrid samples were visually inspected. Then, the clean and structurally intact geogrid samples were directly laid on the top of the lower shear box, and its laying direction was kept consistent with the shear direction.

3. The method according to claim 2, characterized in that, The customized large-size temperature and humidity controllable interface shear test device includes: a test chamber, an upper shear box and a lower shear box set inside the test chamber, a normal loading system and a horizontal displacement loading system; The normal loading system is located above the test chamber and includes a normal pressure bar for applying normal stress to the specimen inside the shear chamber through a rigid pressure plate. The horizontal displacement loading system includes a horizontal push-pull rod and a horizontal fixed rod; during the test, the lower shear box is connected to the horizontal push-pull rod and is driven by the horizontal push-pull rod to generate horizontal displacement; the upper shear box is connected to the horizontal fixed rod to keep the horizontal position fixed. The test chamber has temperature and humidity control functions, and houses a shearing component to maintain the set environmental conditions.

4. The method according to claim 3, characterized in that, The construction of the interface cyclic shear test database based on the valid original test data includes: After completing the multi-condition marine clay-geogrid interface cyclic shear test, the valid raw experimental data collected during the test were processed to construct an interface cyclic shear test database. During database construction, the valid raw experimental data were screened to remove contact nonlinear interference in the initial loading stage and mechanical hysteresis effects during cyclic switching transients. The cyclic shear process data were then reconstructed in segments. Based on this, representative data were extracted and reorganized to form multiple sets of valid sample data. Based on these valid sample data, a multi-dimensional feature database was constructed with temperature, normal stress, shear displacement, and number of cycles as input parameters and interface cyclic shear stress as the output parameter. The input and output data were then uniformly normalized. Subsequently, the database was randomly divided into training and test sets according to a certain proportion using random sampling.

5. The method according to claim 4, characterized in that, The construction and training of a deep learning prediction model based on the interface cyclic shearing test database includes: Based on the interface cyclic shear database, a deep neural network (DNN) model, a convolutional neural network (CNN) model, a bidirectional long short-term memory (BiLSTM) network model, and a CNN-BiLSTM fusion model for convolutional feature extraction and bidirectional temporal modeling were constructed. The training set was used to train each model to predict the cyclic shear stress at the marine clay-geogrid interface.

6. The method according to claim 5, characterized in that, The DNN model adopts a fully connected network structure of "5–128–64–32–1". The input parameters are temperature, normal stress, shear displacement, normal displacement and number of iterations. The output parameter is the interface cyclic shear stress. The training adopts the Adam optimization algorithm with a learning rate of 0.

001. The CNN model adopts a one-dimensional convolutional network structure of "5–64–32–64–1", with a kernel size of 3 and an activation function of ReLU. The training uses the AdamW optimizer combined with cosine annealing learning rate scheduling. The BiLSTM model adopts a "32×5–128–64–1" temporal network structure, takes a five-dimensional time series of length 32 as input, and uses the Adam optimization algorithm for training with a learning rate of 0.

001. The CNN–BiLSTM fusion model adopts a network structure of "32×5–64–32–BiLSTM128–64–1", combining convolutional feature extraction and bidirectional temporal modeling. During training, a random deactivation layer with a dropout rate of 0.30 is set after the BiLSTM layer to randomly drop some neuron connections.

7. The method according to claim 6, characterized in that, The CNN–BiLSTM fusion model employs a sliding window technique, using a five-dimensional feature vector with a length of 10 time steps, containing temperature, normal stress, shear displacement, normal displacement, and number of iterations as temporal input. The model integrates a one-dimensional convolutional layer with a kernel size of 3 and 64 output channels to initially extract spatial correlations and local nonlinear features among the multi-dimensional input features. Subsequently, a three-layer, 128-hidden-unit-per-layer bidirectional long short-term memory (BiLSTM) network is used to deeply capture the long-term memory and complex dependencies of the interface shear stress during cyclic loading from both positive and negative temporal dimensions. In terms of training strategy, the CNN–BiLSTM model uses the Adam optimization algorithm combined with 1×10 -5 The weights are decayed, and a dropout rate of 0.3 is set to prevent overfitting. An adaptive learning rate scheduler is introduced to ensure that the model has excellent stability when it reaches the optimal convergence state.

8. The method according to claim 7, characterized in that, The accuracy of the verification prediction model is determined, the optimal prediction model is selected, and a prediction formula for cyclic shear stress at the marine clay-geogrid interface is constructed based on the optimal prediction model, including: The prediction results of the four models on the test set were compared with the measured data, using the correlation coefficient R. 2 The CNN–BiLSTM fusion model was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). The R-value of the CNN–BiLSTM fusion model on the test set was [not specified]. 2 The deviations between the predicted and measured values ​​of RMSE and MAPE are the smallest among the four types of models, meeting the engineering accuracy requirements. Therefore, the optimal prediction model is the CNN-BiLSTM model. Based on the optimal prediction model, a prediction formula for cyclic shear stress at the marine clay-geogrid interface is established.

Images