A method and system for predicting and optimizing dynamic load characteristics of rubber sand

Abstract

This invention relates to the field of intelligent geotechnical engineering technology, and in particular to a method and system for predicting and optimizing the dynamic load characteristics of rubber sand. The method includes the following steps: obtaining core control variables; setting the core control variables based on the material property boundaries of the rubber sand to construct a multi-condition consolidated drained cyclic dynamic triaxial test; obtaining rubber particle samples through waste tires to obtain standard dynamic triaxial samples of the rubber sand; collecting the original test dataset; preprocessing the original test dataset to obtain a standard test dataset, dividing it into a training set and a test set; constructing a two-branch time-series fusion machine learning model to obtain a dynamic load characteristic prediction model for the rubber sand; obtaining the performance prediction results; and constructing a multi-objective engineering optimization model to obtain the optimal engineering solution. This invention provides accurate, comprehensive, and reliable technical support for the long-term stability design of rubber sand under dynamic load conditions such as traffic roadbeds and the engineering application of waste tire solid waste resource utilization.

Description

Technical Field

[0001] This invention relates to the field of intelligent geotechnical engineering technology, and in particular to a method and system for predicting and optimizing the dynamic load characteristics of rubber sand. Background Technology

[0002] Crushing waste tires into rubber granules and strips, and mixing them with sand to prepare rubber sand as roadbed filler, is currently recognized as the most promising waste tire resource utilization method for engineering applications. It can simultaneously solve the problem of solid waste disposal and the shortage of natural fillers in geotechnical engineering. However, existing rubber sand technology still has many shortcomings that are difficult to overcome, which seriously restrict its large-scale engineering application: First, the research dimensions are seriously disconnected from the engineering scenarios. Most existing studies focus on the static mechanical properties of rubber sand, such as static shear strength and compression modulus, completely ignoring the long-term cyclic dynamic loads experienced by core application scenarios such as traffic roadbeds, railway roadbeds, and airport runways. However, stiffness attenuation, damping evolution, and cumulative plastic deformation under cyclic dynamic loads are the core factors determining the long-term service safety of roadbeds. Second, existing studies generally adopt the simplified assumption of "equivalent spherical particles," completely ignoring the fact that mechanical crushing of waste tires will simultaneously produce two typical types of rubber particles: spherical rubber particles and long strip rubber strips. The objective facts of the morphological products have not been quantitatively characterized, nor have the differences in mechanical behavior of rubber components with different morphologies been quantitatively characterized, nor have the influence mechanisms of the true particle morphology on the rubber-soil interface interlocking, force chain transmission law and macroscopic dynamic characteristics been studied. This results in large dispersion and poor universality of the experimental results, making it difficult to directly guide engineering design. Finally, the engineering design methods are outdated and inefficient. Currently, the design of rubber sand still relies on a large number of indoor tests. For rubber sand with different engineering conditions, different rubber content, and different particle size and morphology ratios, tests need to be carried out again. The test cycle is long and the cost is high. Moreover, it is impossible to quickly optimize a multi-objective optimal design scheme that takes into account vibration isolation and damping effect, engineering bearing capacity, construction cost and solid waste disposal.

[0003] In recent years, machine learning technology has been widely used in the field of predicting the mechanical properties of geotechnical engineering due to its powerful nonlinear fitting capabilities. However, the application of existing machine learning in the field of rubber sand still has obvious shortcomings: First, most of them are purely data-driven "black box" models that do not incorporate the material constitutive laws of rubber sand and the basic principles of geotechnical mechanics, which easily output prediction results that violate physical common sense, and the physical rationality and generalization ability of the models are extremely poor. Second, they only have a single positive prediction function and have not formed a complete technical closed loop of "accurate prediction - interpretable analysis - multi-objective optimization", which cannot directly output feasible engineering design solutions and has insufficient engineering practicality. Third, the real three-dimensional morphological features of rubber particles, rubber strips and sand particles obtained by computed tomography (CT) have not been used as model input parameters, which makes it impossible to capture the key influence of particle morphology on dynamic characteristics, further limiting the prediction accuracy of the model.

[0004] Therefore, developing a technical method that can integrate CT three-dimensional topography characterization, take into account both prediction accuracy and physical rationality, and realize the entire process from predicting the dynamic characteristics of rubber sand to engineering optimization design has become an urgent need to promote the large-scale resource utilization of waste tires, ensure the long-term service safety of seismic isolation roadbeds, and help implement the "dual carbon" strategy. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for predicting and optimizing the dynamic load characteristics of rubber sand. It is applicable to geotechnical engineering scenarios involving long-term cyclic dynamic loads, such as roadbeds, municipal roadbeds, railway roadbeds, airport runways, and subway vibration isolation layers. It can also be used for static load filling projects such as slope reinforcement and foundation pit backfilling. This invention effectively solves industry pain points such as low interfacial bonding strength of waste tire rubber sand, difficulty in coordinating strength and damping performance, and neglect of the influence of particle morphology on mechanical behavior. It achieves high-precision, physically reasonable prediction and engineering-based multi-objective optimization of the dynamic characteristics of rubber sand under cyclic dynamic loads. It can be used for engineering and scientific research work such as rubber sand mechanical property evaluation, long-term service stability verification of seismic isolation roadbeds, optimization of rubber sand engineering design parameters, and large-scale resource utilization of waste tire solid waste.

[0006] To achieve the above objectives, in a first aspect, the present invention provides a method for predicting and optimizing the dynamic load characteristics of rubber sand. The method includes the following steps: obtaining core control variables; setting the core control variables based on the material property boundaries of the rubber sand to construct a multi-condition consolidated-drained-cycle dynamic triaxial test; obtaining rubber particle samples from waste tires; using the rubber particle samples to obtain a standard dynamic triaxial sample of the rubber sand; conducting the multi-condition consolidated-drained-cycle dynamic triaxial test on the standard dynamic triaxial sample and collecting the original test dataset; preprocessing the original test dataset to obtain a standard test dataset; dividing the standard test dataset into a training set and a test set; constructing a dual-branch time-series fusion machine learning model; obtaining a dynamic load characteristic prediction model of the rubber sand based on the training set and the test set; obtaining performance prediction results using the dynamic load characteristic prediction model; and constructing a multi-objective engineering optimization model based on the performance prediction results to obtain the optimal engineering solution. This invention achieves accurate prediction and optimization of the dynamic load characteristics of rubber sand, solves the problem of physically unreasonable pure data-driven models, provides reliable technical support for roadbed engineering, and improves the efficiency of waste tire resource utilization.

[0007] Optionally, the step of obtaining the core control variables and setting them based on the material property boundaries of the rubber sand to construct a multi-condition consolidated drained cyclic dynamic triaxial test includes: obtaining the core control variables, including rubber particle-specific material parameters, sand particle CT scan morphology and gradation parameters, engineering condition parameters, and cyclic loading parameters; the material property boundaries include rubber particle dosage thresholds, particle size thresholds, and morphology effect boundaries; and setting the core control variables at multiple gradient levels based on the common engineering value ranges of the core control variables and the material property boundaries to construct the multi-condition consolidated drained cyclic dynamic triaxial test. This invention ensures that the test conditions closely match engineering realities and the material properties of rubber sand, reduces test dispersion, guarantees data representativeness, provides a scientifically sound foundation for subsequent model training, and improves prediction accuracy.

[0008] Optionally, obtaining rubber particle samples from waste tires and using these samples to obtain standard dynamic triaxial specimens of the rubber sand includes: standardizing the crushing, screening, and surface modification pretreatment of the waste tires to obtain rubber particle samples with different admixture amounts, particle sizes, morphological proportions, and modification methods; obtaining subgrade sand from the target engineering site, mixing the rubber particle samples with the subgrade sand according to the design mix ratio, and adjusting the mixture with water to obtain the optimal moisture content; and using a layered compaction method to prepare standard cylindrical specimens based on the mixture as the standard dynamic triaxial specimens. This invention ensures the uniformity and standardization of the specimens, closely matches the characteristics of the subgrade materials at the site, makes the test results more relevant to engineering applications, and provides a stable and reliable specimen foundation for dynamic load tests.

[0009] Optionally, the step of conducting the multi-condition consolidated drained cyclic dynamic triaxial test on the standard dynamic triaxial specimen and collecting the original test dataset includes: collecting test data in real time at a fixed sampling frequency throughout the entire process of the multi-condition consolidated drained cyclic dynamic triaxial test, including axial stress, axial displacement, and pore water pressure data; simultaneously recording the load-displacement response curve and pore water pressure dissipation curve throughout the entire test to generate full-cycle curve data, including the dynamic shear modulus-shear strain full curve, damping ratio-shear strain full curve, and hysteresis loop evolution curve; post-processing the test data and the full-cycle curve data to extract the core dynamic characteristic indicators of the standard dynamic triaxial specimen; and constructing the original test dataset by combining the test data, the full-cycle curve data, the core dynamic characteristic indicators, and the multi-condition test variables. This invention fully captures the dynamic mechanical evolution law of rubber sand, constructs a comprehensive dataset, accurately reflects the stiffness, damping, and deformation characteristics under dynamic load, and provides high-quality data support for model training.

[0010] Optionally, the core dynamic characteristic indicators include: dynamic strength, cumulative plastic strain sequence, dynamic shear modulus, damping ratio, and critical dynamic stress ratio. This invention provides a comprehensive evaluation of mechanical behavior and long-term stability under dynamic loads, resulting in more complete predictions.

[0011] Optionally, the preprocessing of the original experimental dataset to obtain a standard experimental dataset, and the division of the standard experimental dataset into a training set and a test set, includes: performing outlier removal, smoothing and noise reduction, and dimension normalization on the original experimental dataset to obtain the standard experimental dataset; constructing a dual partitioning rule by combining the non-overlapping dimensions of independent working conditions and the priority isolation of rubber batches; and partitioning the standard experimental dataset based on the dual partitioning rule to obtain the training set and the test set. This invention improves the quality of the dataset through data processing, ensuring the predictive reliability of subsequent models.

[0012] Optionally, the step of constructing a two-branch temporal fusion machine learning model to obtain a dynamic load characteristic prediction model for the rubber sand based on the training set and the test set includes: the two-branch temporal fusion machine learning model includes a static feature branch, a temporal feature branch, and a global fusion output layer; physical constraints are layered and embedded into the two-branch temporal fusion machine learning model, and the two-branch temporal fusion machine learning model is trained using the training set; the trained two-branch temporal fusion machine learning model is validated based on the test set to obtain the dynamic load characteristic prediction model. This invention takes into account both nonlinear mapping and temporal evolution laws, embedding physical constraints to ensure the rationality of model prediction, and improving model accuracy and generalization.

[0013] Optionally, the hierarchical embedding of physical constraints into the dual-branch temporal fusion machine learning model includes: embedding the constitutive constraints of the rubber sand material into the node splitting stage of the static feature branch; embedding the leaf node result correction into the result output stage of the static feature branch; and embedding the geotechnical dynamic property constraints into the global fusion output layer. This invention, through physical constraints, ensures that the prediction conforms to the material constitutive and geotechnical mechanical laws, thereby improving the physical rationality and engineering credibility of the model.

[0014] Optionally, the step of obtaining performance prediction results using the dynamic load characteristic prediction model and constructing a multi-objective engineering optimization model based on the performance prediction results to obtain the optimal engineering solution includes: acquiring actual engineering parameters; inputting the actual engineering parameters into the dynamic load characteristic prediction model to output the performance prediction results; quantifying the contribution of the core control variables to the dynamic load characteristics of the rubber sand based on the performance prediction results; acquiring decision variables, optimization objectives, and constraints based on the contribution to construct the multi-objective engineering optimization model; and solving the multi-objective engineering optimization model using a multi-objective evolutionary algorithm to obtain the optimal engineering solution. This invention forms a complete closed loop from prediction to optimization, significantly improving engineering design efficiency and the practicality of the solution.

[0015] Secondly, this invention provides a system for predicting and optimizing the dynamic load characteristics of rubber sand. The system executes the method for predicting and optimizing the dynamic load characteristics of rubber sand provided by this invention. The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions, and the processor is configured to call the program instructions. This invention, through the collaboration of high-performance hardware, ensures the stable execution of the method and facilitates business deployment and expansion. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for predicting and optimizing the dynamic load characteristics of rubber sand according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the equipment connection of the multi-condition dynamic triaxial testing system according to an embodiment of the present invention; Figure 3 This is a partial detailed schematic diagram of the multi-condition dynamic triaxial testing system according to an embodiment of the present invention; Figure 4 This is a framework diagram of a rubber sand dynamic load characteristic prediction and optimization system according to an embodiment of the present invention; Figure labeling: 1. X-ray emission source; 2. Rubber particle sample; 3. High-precision rotating stage; 4. Image acquisition and scanning detector; 5. Dynamic triaxial pressure chamber; 6. Axial dynamic loading rod; 7. Industrial CT three-dimensional morphology characterization unit; 8. Dynamic triaxial cyclic loading test unit; 9. Central control and data acquisition system; 10. Data processing and visualization terminal; 11. Deformation and strain acquisition unit. Detailed Implementation

[0017] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0018] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0019] Please see Figure 1 An embodiment of the present invention provides a method for predicting and optimizing the dynamic load characteristics of rubber sand, the method comprising the following steps: S1. Obtain the core control variables and set the core control variables based on the material property boundaries of the rubber sand to construct a multi-condition consolidated drainage cyclic dynamic triaxial test.

[0020] In this embodiment, combined with the cyclic dynamic load service scenario of traffic roadbed engineering, the core influencing factors affecting the dynamic load dynamic characteristics of waste tire rubber sand are systematically sorted out. The four categories of rubber particle-specific material parameters (including CT scan three-dimensional morphological features), sand particle CT scan morphology and gradation parameters, engineering working condition parameters, and cyclic loading parameters are taken as core control variables. Taking into account the common value range of engineering and the material characteristic boundaries such as the rubber particle's own dosage threshold, particle size threshold, and morphological effect boundary, multiple gradient levels are set for each variable to complete the parameter grouping and multi-condition test scheme design, so as to construct a multi-condition consolidated drainage cyclic dynamic triaxial test.

[0021] Specifically, taking into account both the common value ranges in engineering and the material property boundaries of rubber sand itself, multiple gradient levels are set for each variable, including the following: Rubber quality dosage: covering the low, medium and high dosage range commonly used in engineering, taking into account both vibration isolation and damping effects and engineering load-bearing capacity requirements.

[0022] Rubber particle size: divided into three particle size grades: fine, medium and coarse, covering the commonly used particle size range of roadbed fillers.

[0023] Morphology ratio: Set a gradient of mass ratios between different types of spherical rubber particles and elongated rubber strips to cover all possible morphological combinations that may be produced by the crushing process.

[0024] Surface modification methods: including unmodified, water-washed modified, Mainstream engineering modification processes such as alkali treatment modification.

[0025] Confining pressure: The stress level corresponding to different depths of the roadbed.

[0026] Initial compaction degree: Meets the compaction degree requirements of the current "Highway Subgrade Design Specification".

[0027] Dynamic stress ratio: corresponds to different traffic load levels.

[0028] Loading frequency: Simulates the driving speed of different vehicles.

[0029] In this embodiment, considering that the number of full factorial design conditions is too large, orthogonal experimental design is used to screen out representative conditions. Three parallel samples are set for each condition, and the average value is taken as the final test result to eliminate random errors in sample preparation and testing.

[0030] S2. Obtain rubber particle sample 2 from waste tires, and use the rubber particle sample 2 to obtain the standard dynamic triaxial sample of the rubber sand.

[0031] Specifically, S2 includes the following steps: S21. The waste tires are subjected to standardized crushing, screening and surface modification pretreatment to obtain rubber particle samples 2 with different dosages, particle sizes, morphology ratios and modification methods.

[0032] In this embodiment, the tread portion of scrapped passenger car tires is selected as raw material. Impurities such as steel wires and cords are removed. First, the tread is cut into strips using a tire strip cutter. After being coarsely crushed by a hammer crusher, it is sent to a rubber fine crusher for further crushing. The rubber mixture with different particle sizes is obtained by screening through a standard vibrating screen.

[0033] Furthermore, industrial CT scanning equipment was used to perform three-dimensional scanning of rubber mixtures with different particle sizes and crushing processes. After scanning, the scan data was processed using three-dimensional reconstruction software. Individual rubber particles and sand particles were separated using a threshold segmentation algorithm to quantitatively characterize their morphological features. The sphericity of spherical rubber particles, the aspect ratio of elongated rubber particles, and the true morphology and gradation characteristics of sand particles were quantitatively characterized, including the following: Spherical rubber particles: characterized by a sphericity index, the calculation formula of which satisfies the following relationship:

[0034] in, It represents the sphericity (with a value ranging from 0 to 1, the closer it is to a standard sphere). Pi The projected area of ​​the particle. Let be the projected perimeter of the particle.

[0035] It should be noted that this embodiment defines sphericity. The particles are spherical rubber particles.

[0036] Long strip rubber strips: characterized by the aspect ratio index, the calculation formula satisfies the following relationship:

[0037] in, The aspect ratio is... The length of the largest bounding rectangle of the particle (i.e., the particle length); The width of the largest bounding rectangle of the particle (i.e., the particle width).

[0038] It should be noted that the aspect ratio is defined in this embodiment. The particles are long, strip-shaped rubber strips.

[0039] Using the above method, the sieved rubber mixture is mixed according to the preset morphology ratio to obtain rubber particle samples 2 with different proportions of spherical / strip-shaped particles.

[0040] Meanwhile, the true morphology and gradation characteristics of sand particles are characterized by industrial CT scanning equipment, completing the CT scanning gradation and morphology characterization of sand particles, as well as the basic physical and mechanical properties tests such as rubber density, particle hardness, and rubber-sand interface friction coefficient.

[0041] In this embodiment, some rubber particle sample 2 undergoes surface modification treatment to remove surface impurities and improve the rubber-sand interface friction coefficient. The modification methods include: Water washing modification: Place rubber granule sample 2 in clean water and stir and rinse. Remove residual mold release agent, oil, and dust impurities from the rubber production process; rinse repeatedly until the effluent is clear, then remove and place in... Drying in an oven Cool to room temperature for later use. This method is simple to operate, inexpensive, and can significantly improve the mechanical interlocking force between particles.

[0042] Alkali treatment modification: using a concentration of approximately of The solution was used as a surface treatment agent, and the rubber particle sample 2 was completely immersed in the solution. During the process, stir continuously to ensure the solution reacts fully with the rubber surface; after removal, rinse repeatedly with clean water to remove any residual alkali solution on the surface; drain excess water, place in an oven to dry, and cool to room temperature for later use.

[0043] S22. Obtain the subgrade sand at the target project site, mix the rubber particle sample 2 with the subgrade sand according to the design ratio, and add water to adjust the mixture to obtain the optimal moisture content.

[0044] In this embodiment, the soil used for the test was taken from the roadbed sand at the target project site. The soil was air-dried, crushed, and passed through a 2mm sieve to remove large particles of impurities.

[0045] Furthermore, rubber particle samples 2 with different dosages, particle sizes, morphological proportions, and modification methods were mixed with roadbed sand according to the design ratio. First, the particles were dry-mixed to disperse them evenly. Then, an appropriate amount of water was added to adjust the moisture content to the optimal level. After mixing, the mixture was sealed and allowed to fully penetrate the soil to obtain a mixture.

[0046] S23. Using the layered compaction method, a standard cylindrical specimen is prepared based on the mixture as the standard dynamic triaxial specimen.

[0047] In this embodiment, standard dynamic triaxial specimens of rubber sand are prepared by layering and compacting according to preset rubber parameters and initial compaction requirements.

[0048] Specifically, a standard cylindrical specimen is prepared in a dynamic triaxial specimen mold using a layered compaction method. Each layer is compacted in layers, and the surface is roughened with a wire brush after each layer is compacted to ensure good interlayer bonding and control the dry density to achieve the target compaction degree. After the specimen is prepared, it is wrapped with plastic wrap and placed in a standard curing room for curing. After the specimen condition is stable, a standard dynamic triaxial specimen is obtained for testing.

[0049] In an optional embodiment, a standard dynamic triaxial specimen is fixed in the pressure chamber of a dynamic triaxial apparatus, the accuracy of the axial and radial load sensors and displacement sensors is calibrated, and core loading parameters such as confining pressure, dynamic stress ratio (used only when describing load levels) and loading frequency are set. The consolidation-drainage cyclic loading mode is clearly adopted, and the multi-condition dynamic triaxial test system is completed.

[0050] Please see Figure 2 The diagram shows the equipment connection of a multi-condition dynamic triaxial testing system. It consists of an industrial CT three-dimensional topography characterization unit 7, a dynamic triaxial cyclic loading test unit 8, a central control and data acquisition system 9, and a data processing and visualization terminal 10 connected in sequence. The industrial CT three-dimensional topography characterization unit 7 is equipped with an X-ray emission source 1, a rubber particle sample 2, and a high-precision rotating stage 3. The dynamic triaxial cyclic loading test unit 8 is equipped with an image acquisition and scanning detector 4, a dynamic triaxial pressure chamber 5, and an axial dynamic loading rod 6.

[0051] Please see Figure 3 The figure shows a partial detail of the multi-condition dynamic triaxial test system; specifically, the details of the dynamic triaxial pressure chamber 5, showing the connection relationship between the deformation strain acquisition unit 11 and the sample in the pressure chamber. The deformation strain acquisition unit 11 is connected to different positions (side and bottom) of the cylindrical sample in the dynamic triaxial pressure chamber 5 through multiple sets of wires.

[0052] Specifically, the prepared standard dynamic triaxial sample is placed in a rubber diaphragm, with permeable stones and filter paper placed at both ends, and then installed in the pressure chamber of the dynamic triaxial apparatus; the axial displacement sensor, radial displacement sensor, and pore water pressure sensor are installed in sequence, and the zero point and range accuracy of each sensor are calibrated.

[0053] Furthermore, a consolidated drained (CD) cyclic loading mode is adopted. First, an isotropic confining pressure is applied for consolidation. The consolidation stability criteria are: pore water pressure dissipation greater than 95% and axial displacement change not greater than 0.01 mm within 1 hour. After consolidation, a sinusoidal cyclic deviatoric stress is applied to simulate the repeated action of traffic load.

[0054] S3. Perform the multi-condition consolidation-drainage-cycle dynamic triaxial test on the standard dynamic triaxial specimen and collect the original test data set.

[0055] In this embodiment, a multi-condition dynamic triaxial test system is activated, and a multi-condition consolidated drained cyclic dynamic triaxial test is carried out according to a preset loading scheme. Test data throughout the entire test process is collected in real time at a fixed sampling frequency, including axial stress, axial displacement, and pore water pressure data. Load-displacement response curves and pore water pressure dissipation curves are recorded simultaneously during the test process to generate full-cycle curve data, including dynamic shear modulus-shear strain curves, damping ratio-shear strain curves, and hysteresis loop evolution curves. This fully records the coupled evolution law of rubber sand stiffness decay, damping enhancement, and deformation accumulation with the cyclic loading process. Core dynamic characteristic indicators for different cyclic stages are extracted, where dynamic strength is defined as the dynamic stress level corresponding to reaching a specified cumulative plastic strain, the cumulative plastic strain sequence is defined as the sequence that changes with the number of cyclic loading cycles, and dynamic shear modulus, damping ratio, and critical dynamic stress ratio are defined according to geotechnical engineering industry standards.

[0056] It should be noted that the test termination condition for each working condition is: the cumulative axial strain of the specimen reaches the level specified in the "Highway Subgrade Design Code". The corruption value, or the number of loop loads, has been reached. Second-rate.

[0057] Specifically, after the experiment, the collected experimental data and full-cycle curve data were post-processed to extract the following five core dynamic characteristic indicators, including the following: Dynamic strength: defined as the dynamic stress level corresponding to the cumulative plastic strain reaching a specified value, denoted as .

[0058] Cumulative plastic strain sequence: recording different number of cycles The cumulative plastic strain value under the condition is denoted as .

[0059] Dynamic shear modulus: The ratio of the elastic shear strain amplitude to the dynamic shear stress amplitude, calculated using the following formula:

[0060] in, For dynamic shear modulus, The dynamic shear stress amplitude (in a dynamic triaxial test), ), This represents the elastic shear strain amplitude (i.e., the strain of the recovered portion).

[0061] Damping ratio: Calculated from the hysteresis loop area, it reflects the energy dissipation capacity of the rubber sand. The formula satisfies the following relationship:

[0062] in, For the damping ratio, Let be the area enclosed by a hysteresis loop within a cycle. Pi Let OAB be the area of ​​triangle OAB.

[0063] It should be noted that for triangle OAB, where O is the origin and A is the positive vertex of the hysteresis loop (coordinates (...)...) , ), corresponding to the point of maximum shear stress), B is the negative vertex of the hysteresis loop (coordinates ( , (corresponding to the point of minimum shear stress).

[0064] Critical dynamic stress ratio: defined as the maximum dynamic stress ratio at which the specimen does not fail within a specified number of cycles, denoted as . .

[0065] Furthermore, the collected test data, full-cycle curve data, core dynamic characteristic indicators and multi-condition test variables are correlated and matched one by one to construct a multi-dimensional original test dataset covering the full dynamic mechanical behavior of rubber sand.

[0066] S4. Preprocess the original experimental dataset to obtain a standard experimental dataset, and divide the standard experimental dataset into a training set and a test set.

[0067] In this embodiment, the original experimental dataset is subjected to a three-step standardization preprocessing to eliminate the impact of data noise and dimensional differences on model training, thereby obtaining a standard experimental dataset.

[0068] Outlier removal: using The criteria eliminate outliers caused by factors such as sensor drift, sample disturbance, and loading error. The judgment formula satisfies the following relationship:

[0069] in, For the first Measured values ​​of a sample The arithmetic mean of all samples for this feature. is the standard deviation of all samples for this feature.

[0070] Smoothing noise reduction: adopts The point moving average method is used to smooth the stress-strain time series curves, completing the 5-point moving average smoothing and noise reduction of the data, eliminating high-frequency noise interference during the experiment. The formula satisfies the following relationship:

[0071] in, For the smoothed first The value of each point, These are the values ​​of 5 adjacent points in the original data.

[0072] Dimensional normalization: Min-max normalization is used to map all input features and output targets to the [0,1] interval, and the formula satisfies the following relationship:

[0073] in, The normalized value. The original value, The maximum value of this feature in the entire dataset. This is the minimum value of this feature in the entire dataset.

[0074] In this embodiment, a dual partitioning rule is adopted, which involves non-overlapping independent working conditions and priority isolation of rubber batches. A unique combination of [rubber content + rubber particle size + spherical / strip-shaped ratio + surface modification method + confining pressure + dynamic stress ratio + loading frequency + initial compaction degree] is defined as one independent working condition. All working conditions of the same rubber batch and the same modification scheme are preferentially included in the training set or test set. Rubber particles are prepared using waste tires from multiple different batches. The full-cycle data of 80% of the independent working conditions is used as the training set, and the full-cycle data of the remaining 20% ​​of independent working conditions with no overlapping variables is used as the test set. The training set and the test set have no overlap in rubber batches and working condition parameters, thus completely avoiding the risk of data leakage caused by time-series data leakage and material discreteness.

[0075] Furthermore, the training set is further divided into a validation set at an 8:2 ratio for hyperparameter adjustment and early stopping detection during model training.

[0076] In practical engineering applications, in order to obtain the optimal prediction accuracy for a specific rubber material used in the target project, all working condition data corresponding to the material can be used as the basic dataset, and the model can be retrained and validated in a ratio of 80% / 20%.

[0077] In an optional embodiment, when the rubber material used in the target project differs significantly from the test material, 5-10 sets of targeted test data can be added to fine-tune the model.

[0078] It should be noted that, compared with the traditional random partitioning method, this partitioning method can truly evaluate the model's ability to generalize to new material batches that have never been seen before, which is more in line with actual engineering application scenarios. The batches of rubber granules used in engineering are often different from the experimental batches, and the model needs to have the ability to predict across batches.

[0079] S5. Construct a dual-branch time-series fusion machine learning model to obtain a dynamic load characteristic prediction model for the rubber sand based on the training set and the test set.

[0080] In this embodiment, a bi-branch temporal fusion machine learning model with embedded material constitutive and geotechnical dynamic constraints is constructed. The model's basic training is completed using the dynamic strength, cumulative plastic strain sequence, dynamic shear modulus, damping ratio, and critical dynamic stress ratio of rubber sand as joint prediction targets. The model adopts a mature and open-source bi-branch fusion architecture: the static feature branch uses a multi-output gradient boosting tree model to specifically learn the nonlinear mapping relationship between rubber particle material parameters, sand particle parameters, engineering working condition parameters, and the basic dynamic characteristics of rubber sand; the temporal feature branch uses a recurrent neural network, inputting the dynamic characteristic data of previous cycles within the same working condition, specifically capturing the temporal evolution law of modulus weakening and plastic accumulation of rubber sand under cyclic dynamic loading; the outputs of the two branches are fused through a fully connected layer of the global fusion output layer to synchronously output five core dynamic characteristic indicators.

[0081] Specifically, the architecture of the dual-branch fusion machine learning model combines the tree model's strong ability to fit static parameters with the recurrent neural network's advantage in processing time-series data.

[0082] Static feature branch: The static feature branch adopts a multi-output gradient boosting tree model. The inputs are static parameters such as rubber content, rubber particle size, spherical / strip-shaped ratio, surface modification method, sand particle size distribution, confining pressure, initial compaction degree, dynamic stress ratio, and loading frequency. Through ensemble learning of multiple decision trees, the nonlinear mapping relationship between static parameters and the dynamic characteristics of rubber sand foundation is explored.

[0083] Temporal feature branch: A recurrent neural network is adopted, with the input being the time series of cumulative plastic strain and dynamic shear modulus of the previous cycle within the same working condition. The time series evolution law of modulus weakening and plastic accumulation of rubber sand under cyclic dynamic load is captured by the gating mechanism, which makes up for the deficiency of pure tree model in processing time series data.

[0084] Global fusion output layer: The output of the static feature branch and the output of the temporal feature branch are concatenated. Through a fully connected layer, five core dynamic characteristic indicators, namely dynamic strength, cumulative plastic strain sequence, dynamic shear modulus, damping ratio and critical dynamic stress ratio, are predicted simultaneously.

[0085] In this embodiment, the optimization objective is to minimize the joint prediction error of multiple objectives. The grid search method is used to globally optimize and screen the core hyperparameters of the model, and the training set is used to train the dual-branch time-series fusion machine learning model.

[0086] Purely data-driven machine learning models are prone to producing predictions that violate physical common sense. Therefore, physical constraints are embedded in layers during the model training phase, and the physical rationality of the prediction results is ensured through a triple mechanism: the constitutive constraints of rubber sand-specific materials are embedded in the node splitting stage of the static feature branch; the leaf node result correction is embedded in the result output stage of the static feature branch; and the general geotechnical dynamic property constraints (based on the Mohr-Coulomb strength criterion and critical state theory) are embedded in the global fusion output layer. The mature triple mechanism of node splitting pre-verification + leaf node result correction + global output verification is used to ensure the physical rationality of the prediction results.

[0087] The pre-split node verification corresponds to the constitutive constraints specific to rubber sand, the global output verification corresponds to the general geotechnical dynamic property constraints, and the leaf node result correction serves as an intermediate supplementary step between the two, as detailed below: Pre-split check for nodes (constitutive constraints of rubber sand material): During each node split in the gradient boosting tree, check whether the predicted values ​​of the left and right child nodes after the split satisfy the following rubber sand-specific physical constraints. If not, the split is prohibited to avoid unreasonable predictions from the source. The following relationship must be satisfied:

[0088] in, The dynamic shear modulus of plain sand. This is the predicted value of the dynamic shear modulus of rubber sand. This is the predicted value of the damping ratio of the rubber sand. For the dynamic strength of plain sandy soil, This is the predicted value of the dynamic strength of rubber sand.

[0089] It should be noted that the above boundaries are typical engineering thresholds for general sandy soil. In practical applications, they can be calibrated and adjusted according to test results from different regions, different sand types, and different rubber sources.

[0090] Leaf node result correction: If the dynamic shear modulus output by the leaf node is lower than 50% or higher than 120% of that of plain sand, it will be forcibly corrected to the corresponding boundary value; if the damping ratio is lower than 0.05 or higher than 0.3, it will be forcibly corrected to the corresponding boundary value.

[0091] Global output verification (geotechnical dynamic property constraints): Ensure that the predicted value of the critical dynamic stress ratio satisfies the Mohr-Coulomb strength criterion and the following relationship:

[0092] in, This is the predicted value of the critical dynamic stress ratio. For effective confining pressure, The effective cohesion of rubber sand under drainage conditions, The effective internal friction angle of rubber sand under drainage conditions.

[0093] It should be noted that, and All can be calibrated through conventional static triaxial tests; if the dynamic characteristics of soil and rock are not satisfied, the original predicted value is replaced by the constraint boundary value.

[0094] In this embodiment, a verification system is constructed based on the prediction accuracy of the test set to verify the trained dual-branch time-series fusion machine learning model, complete the evaluation of the model's generalization ability, and finally obtain a dynamic load characteristic prediction model that is adapted to real engineering scenarios.

[0095] Specifically, the weighted multi-objective mean squared error is used as the loss function of the model, while simultaneously optimizing the prediction accuracy of five dynamic characteristic indices. The formula satisfies the following relationship:

[0096] in, This is the total loss value. To predict the target index, For the first Weighting coefficients for each prediction target (with equal weights). For the first The number of training samples for each target For sample index, For the first The first sample The measured value of each predicted target. For the first The first sample The model prediction value for each prediction target.

[0097] Furthermore, the core hyperparameters of the model (number of decision trees, learning rate, number of hidden layer neurons, and minimum number of samples in leaf nodes) are globally optimized using a grid search method. An early stopping strategy is adopted during training, and training is stopped early when the loss on the validation set no longer decreases for several consecutive rounds to prevent the model from overfitting.

[0098] In this embodiment, the trained model has a high coefficient of determination and a low prediction error on the test set, which meets the accuracy requirements for engineering applications.

[0099] S6. Obtain performance prediction results using the dynamic load characteristic prediction model, and construct a multi-objective engineering optimization model based on the performance prediction results to obtain the optimal engineering solution.

[0100] In this embodiment, actual engineering parameters are obtained, including soil parameters at the target engineering site, confining pressure corresponding to different depths of the subgrade, dynamic stress ratio and loading frequency corresponding to traffic load, and subgrade design compaction parameters. The actual engineering parameters are input into the dynamic load characteristic prediction model, and the full-cycle dynamic characteristic evolution law and long-term service performance prediction results of the rubber sand under this working condition are output.

[0101] In an optional embodiment, feature importance analysis is performed based on Shapley Additive Explanations (SHAP) to quantify the contribution of each core control variable to the dynamic load characteristics of the rubber sand. The global average SHAP value of each feature is calculated according to the following relationship:

[0102] in, For the first The global average SHAP value of each feature. The total number of samples, For sample index, Not including the first Any subset of features, The set of all input features. For feature index, For subset The number of features, The total number of input features in the set. The factorial of the corresponding numerical value. To use feature subsets Add the When the feature is , the model is for the . The predicted value for each sample, For the first One sample, To use only a subset of features Time model for the first The predicted value for each sample.

[0103] The analysis results show that the rubber content is the core factor determining the vibration isolation and damping effect of rubber sand; the morphology of rubber particles (the ratio of spherical to elongated shapes) has a significant impact on the dynamic shear modulus and cumulative plastic strain; the confining pressure and dynamic stress ratio are the main factors controlling long-term deformation. These analysis results are consistent with the basic laws of geotechnical mechanics and further verify the rationality of the model.

[0104] In this embodiment, a multi-objective engineering optimization model is constructed based on the performance prediction results. The optimization objectives are to achieve the best vibration isolation and damping effect, the lowest engineering cost, and the largest amount of solid waste disposal. The constraints are the requirements of the current roadbed design specifications and the threshold values ​​of rubber sand material properties.

[0105] Specifically, multi-objective engineering optimization is performed based on the Non-dominated Sorting Genetic Algorithm II (NSGA-II) with an elitist strategy. The multi-objective optimization problem can be expressed in standard mathematical form and satisfies the following relationship:

[0106] in, Indicates minimization. For the decision variable vector, Let be the feasible region of the decision variable. For the objective function vector (maximizing the objective is transformed into minimizing its negative value). For the first original objective function, For the second original objective function, For the third original objective function, These are conjunctions that introduce constraints. For the first Inequality constraint functions, For constraint indexes.

[0107] To adapt to the NSGA-Ⅱ algorithm, the surface modification method is mapped to integer values: 0 - unmodified, 1 - water washing modification, 2 - NaOH alkali treatment modification.

[0108] For the specific working conditions of the target project, a multi-objective engineering optimization model is established to achieve the best overall optimization in terms of vibration isolation and damping effect, lowest project cost, and maximum solid waste disposal, while meeting the subgrade design requirements. This includes the following: The decision variables satisfy the following relationship:

[0109] in, For the decision variable vector, The mass content of rubber particles, For rubber particle size, The proportion of spherical rubber particles, This is a surface modification method.

[0110] The optimization objective satisfies the following relationship:

[0111] in, This means maximizing the first original objective function (which achieves the best vibration isolation and damping effect). For decision variable vectors The corresponding maximum damping ratio (obtained from the dynamic load characteristic prediction model). This means minimizing the second original objective function (minimizing engineering cost). Cost per unit mass of waste tire rubber granules The mass content of rubber particles, To account for the unit mass cost of the corresponding surface modification method, Cost per unit mass of sand This means maximizing the third original objective function (maximizing the amount of solid waste disposed of). This represents the dry density of the soil.

[0112] The constraints satisfy the following relationships (in order: satisfying the long-term deformation condition of the subgrade, avoiding excessive strength degradation, and satisfying the bearing capacity requirements of the subgrade):

[0113] in, For decision variable vectors The cumulative plastic strain sequence under specified cyclic loading (obtained from the dynamic load characteristic prediction model). To specify the number of cyclic load applications, The allowable long-term deformation limit of the subgrade as specified in the "Specifications for Design of Highway Subgrade". The mass content of rubber particles, To the maximum permissible rubber content, For decision variable vectors The dynamic strength of the rubber sand. The dynamic strength of plain sandy soil.

[0114] In this embodiment, a multi-objective evolutionary algorithm commonly used in the engineering field is used to solve the multi-objective engineering optimization model to obtain the Pareto optimal solution set. Then, the optimal engineering scheme adapted to different road sections and different depths of the project is automatically output from the Pareto optimal solution set, including the optimal combination of rubber content, particle size, morphology ratio and surface modification scheme.

[0115] Specifically, the non-dominated sorting genetic algorithm (NSGA-II) with an elitist strategy is used to solve the above problem. The core process is as follows: Hybrid coding and population initialization: Continuous variables (dosage, particle size, morphology proportion) are encoded with real numbers, while discrete variables (modification method) are encoded with integers. The initial population is randomly generated within the feasible region.

[0116] Fitness calculation: The trained machine learning model is called to predict the dynamic characteristics of each individual, the objective function value is calculated, and the penalty function method is used to handle infeasible solutions that do not meet the constraints.

[0117] Fast non-dominated ordination and crowding calculation: The population is divided into different non-dominated layers according to the Pareto dominance relation. Crowding is calculated for individuals within the same layer to maintain population diversity. The crowding calculation formula satisfies the following relationship:

[0118] in, For the first The degree of crowding of individuals For individual indexes, The index of the objective function. For the first The first individual The objective function value, For the first The first individual The objective function value, For the first non-dominated layer The maximum value of each objective function. For the first non-dominated layer The minimum value of an objective function.

[0119] Genetic manipulation and population renewal: A binary tournament selection method is adopted, combined with simulated binary crossover and polynomial mutation operators to generate offspring populations; the parent generation and offspring generation are merged, and elite individuals are selected to form a new generation population according to the principle of "non-dominated layer priority, crowding degree second".

[0120] Termination condition: The iteration continues until the preset maximum number of iterations or the Pareto optimal solution set converges, and the final Pareto optimal solution set is output.

[0121] In this embodiment, after the algorithm iterates to convergence, a Pareto optimal solution set is obtained. No solution in this set can improve the performance of a certain objective without reducing the performance of other objectives.

[0122] Furthermore, the optimal engineering solution is selected from the Pareto optimal solution set based on the actual needs of the project. For example, the recommended solution for a typical traffic roadbed project can simultaneously meet the requirements of vibration isolation and damping, bearing capacity, cost, and environmental protection; alternatively, different preferred solutions can be selected from the Pareto optimal solution set depending on the project's focus, such as prioritizing construction cost or prioritizing solid waste disposal.

[0123] Please see Figure 4In an optional embodiment, the present invention provides a system for predicting and optimizing the dynamic load characteristics of rubber sand. The system includes an input device, an output device, a processor, and a memory, all interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions and execute specific steps as described in the embodiments of the rubber sand dynamic load characteristic prediction and optimization method provided by the present invention. The rubber sand dynamic load characteristic prediction and optimization system provided by the present invention has a complete structure, is objective and stable, and improves the overall applicability and practical application capability of the present invention.

[0124] In summary, this invention provides a method and system for predicting and optimizing the dynamic load characteristics of rubber sand. This invention addresses the industry pain points of high pressure and low resource utilization rate in the disposal of waste tire solid waste, as well as existing research on rubber sand that fails to consider the influence of the true morphological characteristics (spherical / elongated) of rubber-sand particles obtained from CT scans on interfacial interactions, focuses only on static mechanical properties of a single parameter, and fails to reveal the coupling law between stiffness decay and damping evolution under traffic cycle dynamic loads. Existing machine learning prediction methods suffer from purely data-driven black-box models that easily violate material constitutive and geotechnical mechanisms, lack a physical constraint system adapted to the characteristics of rubber sand, and lack accurate forward-looking predictions. This invention addresses the core technical shortcomings of reverse engineering optimization of a complete closed loop. Its core objective is to leverage the unique material properties of waste tire rubber particles, integrating CT scanning 3D reconstruction technology to obtain the true morphology, gradation, and porosity characteristics of rubber particles (quasi-spherical / elongated) and sand particles. A bi-branch time-series fusion machine learning model, embedded with material constitutive and geotechnical mechanics constraints, is constructed to achieve multi-objective, synchronous, and high-precision prediction of core dynamic characteristics of rubber sand under cyclic dynamic loading, such as dynamic strength, dynamic shear modulus, damping ratio, cumulative plastic strain, and critical dynamic stress ratio. Simultaneously, an engineering-oriented multi-objective optimization design closed loop is established, balancing the physical rationality and engineering practicality of the prediction results. This invention's method is easy to understand, computationally simple, requires minimal workload, and is readily applicable in engineering, providing a theoretical foundation and technical support for the further development of intelligent geotechnical engineering technology.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for predicting and optimizing the dynamic load characteristics of rubber sand, characterized in that, Includes the following steps: The core control variables are obtained and set based on the material property boundaries of the rubber sand to construct a multi-condition consolidated drainage circulation dynamic triaxial test. Rubber particle samples were obtained from waste tires, and the standard dynamic triaxial sample of the rubber sand was obtained from the rubber particle samples. The standard dynamic triaxial specimen was subjected to the multi-condition consolidated drainage cycle dynamic triaxial test, and the original test data was collected. The original experimental dataset is preprocessed to obtain a standard experimental dataset, which is then divided into a training set and a test set. A dual-branch time-series fusion machine learning model is constructed to obtain a dynamic load characteristic prediction model for the rubber sand based on the training set and the test set. The performance prediction results are obtained by using the dynamic load characteristic prediction model, and a multi-objective engineering optimization model is constructed based on the performance prediction results to obtain the optimal engineering solution.

2. The method for predicting and optimizing the dynamic load characteristics of rubber sand according to claim 1, characterized in that, The acquisition of core control variables, and the setting of these core control variables based on the material property boundaries of the rubber sand to construct a multi-condition consolidated drained cyclic dynamic triaxial test, includes: The core control variables are obtained, including the rubber particle-specific material parameters, the CT scan morphology and gradation parameters of the sand particles, the engineering working condition parameters, and the cyclic loading parameters. The material property boundaries include the rubber particle doping threshold, particle size threshold, and morphology effect boundary. Based on the common engineering value range of the core control variables and the material property boundaries, the core control variables are set with multiple gradient levels to construct the multi-condition consolidated drainage cyclic dynamic triaxial test.

3. The method for predicting and optimizing the dynamic load characteristics of rubber sand according to claim 1, characterized in that, The process of obtaining rubber particle samples from waste tires and using these rubber particle samples to obtain a standard dynamic triaxial sample of the rubber sand includes: The waste tires are subjected to standardized crushing, screening and surface modification pretreatment to obtain rubber particle samples with different dosages, particle sizes, morphology ratios and modification methods. Obtain the subgrade sand from the target project site, mix the rubber particle sample with the subgrade sand according to the design ratio, and add water to adjust the moisture content of the mixture to obtain the optimal moisture content. A standard cylindrical specimen was prepared based on the mixture using a layered compaction method as the standard dynamic triaxial specimen.

4. The method for predicting and optimizing the dynamic load characteristics of rubber sand according to claim 1, characterized in that, The process of conducting the multi-condition consolidated drained cyclic dynamic triaxial test on the standard dynamic triaxial specimen and collecting the original test data set includes: Throughout the multi-condition consolidation-drainage cyclic dynamic triaxial test, test data, including axial stress, axial displacement, and pore water pressure data, were collected in real time at a fixed sampling frequency. The load-displacement response curve and pore water pressure dissipation curve were recorded synchronously throughout the entire test process to generate full-cycle curve data, including the dynamic shear modulus-shear strain full curve, the damping ratio-shear strain full curve, and the hysteresis loop evolution curve. The test data and the full-cycle curve data are post-processed to extract the core dynamic characteristic indicators of the standard dynamic triaxial specimen. The original test dataset is constructed by combining the test data, the full-cycle curve data, the core dynamic characteristic indicators, and the multi-condition test variables.

5. The method for predicting and optimizing the dynamic load characteristics of rubber sand according to claim 4, characterized in that, The core dynamic performance indicators include: The core dynamic characteristic indicators include dynamic strength, cumulative plastic strain sequence, dynamic shear modulus, damping ratio, and critical dynamic stress ratio.

6. The method for predicting and optimizing the dynamic load characteristics of rubber sand according to claim 1, characterized in that, The process of preprocessing the original experimental dataset to obtain a standard experimental dataset, and then dividing the standard experimental dataset into a training set and a test set, includes: The original experimental dataset is subjected to outlier removal, smoothing and noise reduction, and dimension normalization to obtain the standard experimental dataset. A dual partitioning rule is constructed by combining the non-overlapping dimensions of independent working conditions and the priority isolation of rubber batches. The standard experimental dataset is divided into the training set and the test set based on the dual partitioning rule.

7. The method for predicting and optimizing the dynamic load characteristics of rubber sand according to claim 1, characterized in that, The construction of the dual-branch time-series fusion machine learning model, based on the training set and the test set, to obtain a prediction model for the dynamic load characteristics of the rubber sand includes: The dual-branch temporal fusion machine learning model includes a static feature branch, a temporal feature branch, and a global fusion output layer; Physical constraints are hierarchically embedded into the dual-branch temporal fusion machine learning model, and the training set is used to train the dual-branch temporal fusion machine learning model; The trained dual-branch time-series fusion machine learning model is validated based on the test set to obtain the dynamic load characteristic prediction model.

8. The method for predicting and optimizing the dynamic load characteristics of rubber sand according to claim 7, characterized in that, The hierarchical embedding of physical constraints into the dual-branch temporal fusion machine learning model includes: Embed the constitutive constraints of the rubber sand material into the node splitting stage of the static feature branch, and embed the leaf node results into the result output stage of the static feature branch; The constitutive constraints of the rubber sand material satisfy the following relationship: in, The dynamic shear modulus of plain sand. This is the predicted value of the dynamic shear modulus of rubber sand. This is the predicted value of the damping ratio of the rubber sand. For the dynamic strength of plain sandy soil, This is the predicted value of the dynamic strength of rubber sand; The constraints on geotechnical dynamic properties are embedded in the global fusion output layer.

9. The method for predicting and optimizing the dynamic load characteristics of rubber sand according to claim 1, characterized in that, The process of obtaining performance prediction results using the dynamic load characteristic prediction model and constructing a multi-objective engineering optimization model based on the performance prediction results to obtain the optimal engineering solution includes: Obtain actual engineering parameters and input the actual engineering parameters into the dynamic load characteristic prediction model to output the performance prediction result; Based on the performance prediction results, the contribution of the core control variables to the dynamic load characteristics of the rubber sand is quantified; Based on the contribution level, decision variables, optimization objectives, and constraints are obtained to construct the multi-objective engineering optimization model. The optimization objective satisfies the following relationship: in, This means maximizing the first original objective function (which achieves the best vibration isolation and damping effect). For decision variable vectors The corresponding maximum damping ratio (obtained from the dynamic load characteristic prediction model). This means minimizing the second original objective function (minimizing engineering cost). Cost per unit mass of waste tire rubber granules The mass content of rubber particles, To account for the unit mass cost of the corresponding surface modification method, Cost per unit mass of sand This means maximizing the third original objective function (maximizing the amount of solid waste disposed of). This refers to the dry density of the soil. The multi-objective evolutionary algorithm is used to solve the multi-objective engineering optimization model to obtain the optimal engineering solution.

10. A system for predicting and optimizing the dynamic load characteristics of rubber sand, characterized in that, The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the method for predicting and optimizing the dynamic load characteristics of rubber sand as described in any one of claims 1-9.

Images