CN-122021354-A - Rock mechanical property intelligent prediction and parameter optimization method based on digital twin

CN122021354ACN 122021354 ACN122021354 ACN 122021354ACN-122021354-A

Abstract

The invention provides a digital twinning-based rock mechanical property intelligent prediction and parameter optimization method, which belongs to the technical field of geotechnical engineering and intelligent computation and comprises the steps of S1, obtaining data, S2, mapping and unified expression of scales, S3, modal division, S4, extracting characteristic data, S5, characteristic coding and fusion, S6, constructing a mechanical behavior prediction and parameter optimization model, S7, outputting a prediction result and/or an optimization result, and S8, self-adaptive closed-loop optimization. According to the invention, through the intelligent analysis scheme of experiment-numerical simulation fusion and combination of cross-scale data modeling and physical priori constraint, the rapid and accurate prediction of the mechanical response of the rock contact surface, the intelligent inversion and optimization of key mechanical parameters of the contact surface are realized, and finally the calculation efficiency, the prediction precision and the engineering applicability of the mechanical analysis of the rock contact surface are comprehensively improved.

Inventors

YU PENGCHENG
LUO CHAO
ZHAN QUANYU
ZHANG YINGBIN
LI DEJIAN
LI HAOJIA

Assignees

西南交通大学

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The intelligent rock mechanical property prediction and parameter optimization method based on digital twinning is characterized by comprising the following steps of: s1, acquiring experimental data and numerical simulation data of a rock contact surface, and correspondingly matching material parameters, working condition parameters and contact surface image raw data; s2, performing microscopic-macroscopic three-level cross-scale division on all the data acquired in the S1, constructing a scale conversion matrix based on a homogenization theory to finish scale mapping and unified expression of data features, and obtaining a cross-scale data set with unified scale; S3, performing modal division on the cross-scale data set according to a physical expression form to obtain parameter modal data, curve modal data, image modal data and time sequence modal data; s4, processing image mode data through a multi-scale image feature extraction model, and extracting to obtain cross-scale image mode feature data; S5, preprocessing, encoding and fusing parameter mode data, curve mode data, time sequence mode data and cross-scale image mode feature data through a multi-mode feature processing model to obtain experimental-numerical simulation cross-scale fusion feature representation; S6, constructing a mechanical behavior prediction and parameter optimization model, taking experimental-numerical simulation trans-scale fusion characteristics as input, taking experimental data and numerical simulation data as training labels, and introducing rock mechanical physical constraint to complete model training so as to obtain a prediction model after training; S7, inputting working condition parameters to be analyzed into a prediction model, and outputting a mechanical response prediction result and/or a mechanical parameter optimization result of the corresponding rock contact surface; and S8, comparing the verification data corresponding to the working condition to be analyzed with a mechanical response prediction result or a mechanical parameter optimization result, and adaptively updating the corresponding models of S4, S5 and S6 based on the comparison result to form an experiment-numerical simulation-artificial intelligence collaborative self-adaptive closed-loop optimization process.
2. The intelligent prediction and parameter optimization method based on digital twin rock mechanical characteristics according to claim 1, wherein the parameter mode data in the step S3 comprises at least one of normal stress, shear displacement, contact stiffness and friction coefficient under different scales, the curve mode data is used for representing at least one of a shear stress-displacement relation curve, a normal displacement evolution curve and a cyclic loading response curve under different scales or different loading levels, the image mode data comprises at least one of a rock contact surface morphology image, a scanning contour map and a contact state distribution map under microscopic scales, microscopic scales or macroscopic scales, and the time sequence mode data is used for representing the evolution process of a rock contact surface mechanical response over time under different scales or cross-scale coupling conditions.
3. The intelligent prediction and parameter optimization method based on digital twin rock mechanical characteristics according to claim 1, wherein the multiscale image feature extraction model in S4 adopts a multiscale convolutional neural network structure, performs parallel feature extraction on rock contact surface images with different spatial resolutions and scale levels, forms uniform cross-scale image feature representation through a scale alignment or feature aggregation mode, and extracts geometric forms and surface roughness features related to the rock contact surface mechanical behaviors by combining a feature description mode of local texture statistical analysis.
4. The intelligent prediction and parameter optimization method based on digital twin rock mechanical characteristics according to claim 1, wherein in the step S5, the multi-mode feature processing model introduces an attention weighting mechanism based on the mechanical physical prior constraint of the rock contact surface in the feature encoding and fusion process, wherein the attention weight is initialized based on at least one mechanical prior information of the contact stiffness evolution rule, the shear-normal coupling relationship and the friction constitutive characteristics, and is adaptively corrected based on a data driving mode in the model training process.
5. The intelligent prediction and parameter optimization method based on digital twinning of rock mechanical properties according to claim 1, wherein the rock mechanical physical constraints in S6 include at least one of mechanical balance constraints, boundary condition constraints and material constitutive consistency constraints, the physical constraints participate in the process of constructing model loss functions or updating attention weights to realize guidance of physical prior information to model training process, the specific steps of model training include: Performing consistency processing and feature reconstruction on experimental data and numerical simulation data corresponding to the experimental-numerical simulation trans-scale fusion feature representation to form a feature vector for model input; dividing the feature vector into a training data set and a verification data set according to a preset proportion; Performing scale unification on the training data set and the verification data set to eliminate the influence of different physical quantity dimension differences on model training; Inputting the processed training data set into a mechanical behavior prediction and parameter optimization model, and iteratively updating network weight parameters under the condition of meeting the preset physical constraint by a back propagation algorithm; And outputting the prediction model after training when the prediction error of the model on the verification data set meets the preset convergence condition.
6. The intelligent prediction and parameter optimization method for rock mechanical properties based on digital twinning according to claim 1, wherein the mechanical response prediction result in S7 comprises at least one of shear strength, normal deformation, contact stiffness evolution and friction properties of a rock contact surface, and the mechanical parameter optimization result is output in the form of at least one of a parameter optimal value, a parameter candidate interval and a parameter probability distribution.
7. The intelligent prediction and parameter optimization method based on digital twin rock mechanical characteristics according to claim 1 is characterized in that the verification data corresponding to the working condition to be analyzed in S8 comprises experimental results, numerical simulation results or engineering monitoring results under the working condition, after comparison, the scene to be analyzed is classified based on at least one of working condition parameters, loading paths and contact surface state change characteristics, and then a corresponding model update strategy is selected in a self-adaptive mode according to the scene classification results and the comparison relation between prediction errors and preset error thresholds, wherein the preset error thresholds are dynamically adjusted according to at least one of scene types, historical prediction error distribution and engineering safety levels.
8. The intelligent prediction and parameter optimization method based on digital twin rock mechanical properties of claim 7, wherein the model updating strategy comprises at least one of incremental training, migration learning and model retraining, and the model updating process comprises at least one of model parameter adaptive adjustment, model structure reconfiguration and feature weight redistribution, wherein the feature weight redistribution process is simultaneously influenced by a combination of a mechanical physical prior constraint and a data-driven optimization result.
9. The intelligent rock mechanical property prediction and parameter optimization method based on digital twinning according to claim 1 is characterized by comprising the steps of constructing a digital twinning body of a rock contact surface based on a prediction model obtained by training and obtained in the step S6 and combining three-dimensional scanning data and geological survey data of an engineering site, wherein the digital twinning body is used for mapping the geometric form, the mechanical state and the evolution process of a real rock contact surface, collecting sensor data of the engineering site in real time, and inputting the sensor data into the digital twinning body to realize real-time update of the state of the digital twinning body.
10. The intelligent prediction and parameter optimization method based on the digital twin rock mechanical characteristics according to claim 9, wherein at least one of a support structure adjustment scheme, a loading scheme optimization proposal and risk early warning information for guiding engineering decisions is output based on the real-time state of the digital twin, the mechanical response prediction result output by the S7 and/or the mechanical parameter optimization result in combination with a preset engineering safety threshold.

Description

Rock mechanical property intelligent prediction and parameter optimization method based on digital twin Technical Field The invention relates to the technical field of intersection of geotechnical engineering and intelligent computation, in particular to a method for intelligently predicting mechanical properties of rock mass and optimizing parameters based on digital twinning. Background The mechanical behaviors of contact surfaces such as a rock joint surface, a fracture surface, a structural surface and the like are core basic problems of rock mass engineering stability analysis and disaster prevention and control, and the shear strength, deformation characteristics and parameter evolution rules of the rock mass engineering stability analysis and disaster prevention and control are used for directly determining the safety and reliability of rock mass engineering such as side slopes, underground caverns, tunnels and the like. At present, the main stream research means in the industry has obvious bottleneck: The indoor mechanical test method can truly restore the mechanical response characteristics of the rock contact surface through a direct shear test, a cyclic shear test and the like, is strictly limited by sample preparation, loading working conditions and equipment capacity, is difficult to cover multiple parameters and multiple working conditions in complex engineering scenes, has high test cost, cannot form a data expression form with uniform physical significance under the multiple-scale and multiple working conditions, and severely limits the application of the data driving method in engineering prediction and parameter inversion. The numerical simulation method is characterized in that a rock mass block and contact relation model is constructed, so that the shearing sliding and deformation process of a rock contact surface can be flexibly simulated, but the calculated result is highly sensitive to key parameters such as contact stiffness, friction coefficient, cohesive force and the like. The parameters are generally determined by experience value taking or manual repeated trial calculation, and an intelligent unified mapping and inversion mechanism between experimental data and numerical simulation results is lacked, so that the parameter inversion efficiency is extremely low, and the consistency and stability of the parameter value taking under different working conditions are difficult to maintain. In the prior art, experimental data and numerical simulation results are used in a mutually independent mode, and a complete technical scheme capable of carrying out multi-mode feature expression and fusion modeling on two types of data under a unified physical constraint condition and simultaneously realizing intelligent prediction of mechanical behaviors and parameter self-adaptive optimization is not formed, so that the core requirements of complex rock mass engineering fine analysis and efficient calculation cannot be met. Disclosure of Invention The invention provides a rock mechanical property intelligent prediction and parameter optimization method based on digital twinning, which aims to overcome the core defects of the conventional rock contact surface mechanical analysis technology, namely, the key mechanical parameters depend on manual experience to take values, the numerical simulation calculation efficiency is low, experimental data and numerical simulation data are difficult to effectively fuse under a unified physical frame, and mechanical information with different scales is difficult to cooperatively model. By providing an intelligent analysis scheme of experiment-numerical simulation fusion and combining cross-scale data modeling and physical priori constraint, the rapid and accurate prediction of the mechanical response of the rock contact surface, the intelligent inversion and optimization of key mechanical parameters of the contact surface are realized, and finally the calculation efficiency, the prediction precision and the engineering applicability of the mechanical analysis of the rock contact surface are comprehensively improved. In order to achieve the above purpose, the invention adopts the following technical scheme: A rock mechanical property intelligent prediction and parameter optimization method based on digital twinning comprises the following steps: s1, acquiring experimental data and numerical simulation data of a rock contact surface, and correspondingly matching material parameters, working condition parameters and contact surface image raw data; s2, performing microscopic-macroscopic three-level cross-scale division on all the data acquired in the S1, constructing a scale conversion matrix based on a homogenization theory to finish scale mapping and unified expression of data features, and obtaining a cross-scale data set with unified scale; s3, performing modal division on the cross-scale data set according to a physical expression form to obtain four types of