Rock mass joint mechanical property simulation test system and test method based on physical field equivalence principle

Abstract

The invention belongs to the technical field of sensors and machine learning, and provides a rock mass joint mechanical property simulation test system and method based on the physical force field equivalence principle, and the method mainly comprises the steps: constructing a geomechanical similar model based on the physical force field equivalence principle, when the substrate moves relative to the interface of the rock mass physical model, the friction effect between the substrate and the bottom surface of the rock mass physical model generates uniformly distributed shear stress, and the shear stress generates equivalent physical strength in the rock mass physical model; collecting multi-dimensional sensing data when the substrate moves relative to the interface of the rock physical model; on the basis of multi-dimensional sensing data, an improved U-Net machine learning model is used for automatically recognizing a rock mass joint extension track, displacement field data is obtained through combination of improved particle image velocity measurement and an optical flow algorithm, and a Bayesian reasoning inversion model is used for inverting mechanical parameters of a joint surface. And the uncertainty of an inversion result is evaluated through a Markov chain Monte Carlo method.

Description

Technical Field

[0001] This invention relates to the fields of sensor and machine learning technology, and in particular to a simulation test system and test method for the mechanical properties of rock joints based on the principle of body force field equivalence. Background Technology

[0002] The evaluation of rock mass stability in engineering relies heavily on the accurate understanding of joint mechanical properties. However, current research mainly relies on traditional experimental equipment such as large-scale direct shear testing machines and true triaxial testing systems, which have several major technical shortcomings: These large-scale hydraulic servo control systems are not only expensive, with a single set typically costing several million yuan; they also require a large footprint and a professional maintenance team, resulting in high research and development costs; in terms of sample preparation, both the severe disturbance during undisturbed sampling and the serious distortion of key parameters such as joint surface morphology, roughness JRC, and filling material characteristics during the preparation of reconstructed samples make it difficult to truly reflect the structural characteristics of natural rock masses; at the same time, due to the limited space of the equipment, the sample size is usually less than 1 cubic meter, which cannot effectively study the synergistic effect of joint networks and the overall instability mechanism in rock masses of tens to hundreds of meters in engineering scale; in addition, traditional tests usually take several days or even weeks from sample preparation and installation to the completion of a single test, which seriously restricts the conduct of parameter sensitivity analysis and a large number of comparative studies.

[0003] Based on a systematic understanding of the long-standing contradiction between theoretical deficiencies and engineering practice needs in the field of rock mass physical simulation, the applicant identified three key technical bottlenecks during in-depth research into the technical foundations of existing physical simulation methods: First, traditional physical simulation methods suffer from serious deficiencies at the theoretical level. Most existing simulation techniques are based on empirical analogies and lack rigorous physical field theory support. This reliance on experience leads to a fragmented system of similarity criteria in the simulation process, making it impossible to establish a complete mechanical mapping relationship from the model to the prototype. In particular, when simulating large-scale rock mass engineering, the quantitative conversion relationship between model response and prototype behavior has never been rigorously proven, which has led to widespread questioning of the scientific validity and reliability of the experimental results.

[0004] Secondly, existing technologies have fundamental limitations in their implementation mechanisms. Traditional loading methods, whether using gravity fields, boundary forces, or centrifugal force fields, struggle to reproduce the complex stress states in real rock masses on a laboratory scale. Gravity field simulations are limited to a 1g environment and cannot effectively simulate deep rock conditions. Boundary force loading introduces unrealistic stress concentration phenomena. While centrifuge simulations can increase gravity levels, their equipment is complex, costly, and makes it difficult to observe dynamic processes. These technological limitations severely restrict the development of rock mechanics research.

[0005] Then, through systematic theoretical analysis and basic experimental research, the existing simulation methods have failed to find a physical mechanism that can both accurately reproduce the distribution characteristics of the body force field and achieve convenient control and quantitative adjustment. Summary of the Invention

[0006] The purpose of this invention is to provide a new simulation test system and test method for the mechanical properties of rock joints based on the principle of equivalent body force field.

[0007] The technical solution adopted by this invention to solve its technical problem is as follows: On the one hand, the present invention provides a simulation test system for the mechanical properties of rock joints based on the principle of equivalent force field, comprising: The geomechanical similarity model construction unit is used to construct a geomechanical similarity model based on the principle of equivalent body force field. The geomechanical similarity model includes a base and a rock mass physical model. When the base moves relative to the interface of the rock mass physical model, the friction between the base and the bottom surface of the rock mass physical model generates uniformly distributed shear stress. The shear stress generates equivalent body force strength inside the rock mass physical model. A multi-dimensional sensor data acquisition unit is used to acquire multi-dimensional sensing data when the base moves relative to the interface of the rock mass physical model. The data processing and inversion unit is used to automatically identify the rock mass joint propagation trajectory based on multidimensional sensor data and an improved U-Net machine learning model. It also uses an improved particle image velocimetry combined with optical flow algorithm and displacement field data to invert the mechanical parameters of the joint surface using a Bayesian inference inversion model. Finally, it evaluates the uncertainty of the inversion results using the Markov chain Monte Carlo method.

[0008] In some embodiments, the substrate is disposed on a translation stage, and the substrate moves at a constant speed, with the speed range of the substrate being 0.1-100 mm / s; The interface material of the rock mass physical model interface is replaceable, and different replaceable interface materials correspond to different friction coefficients.

[0009] In some embodiments, the geomechanical similarity model satisfies geometric similarity, mechanical similarity, and boundary condition similarity; The rock mass physical model in the geomechanical similarity model has an elastic modulus ranging from 10 MPa to 1 GPa.

[0010] In some embodiments, the multidimensional sensor data acquisition unit includes a high-speed camera and a laser displacement sensor array; The high-speed camera has a maximum acquisition frequency of 1000fps and is used to acquire image data of the rock mass physical model. The laser displacement sensor array consists of 30 laser displacement sensors arranged in a 5×6 matrix, evenly distributed on the surface of the rock mass physical model with a horizontal spacing of 15cm and a vertical spacing of 20cm. Each laser displacement sensor is perpendicular to the surface of the rock mass physical model. The laser spot diameter of the laser displacement sensor is 2-3mm, and the spatial resolution is 0.1mm. The laser displacement sensor array is used to acquire displacement data.

[0011] In some embodiments, the high-speed camera and laser displacement sensor array are integrated into a digital image correlation deformation testing system. The digital image correlation deformation testing system uses an adaptive mesh partitioning algorithm to perform multi-source data weighted fusion processing on the image data and displacement data of the rock mass physical model to obtain rock mass joint images, which are used as the dataset for the improved U-Net machine learning model.

[0012] In some embodiments, the improved U-Net machine learning model has a model structure including an encoder, a decoder, and an output layer; The encoder uses ResNet-50 as the backbone network to extract multi-scale feature maps, and the output feature map sizes are 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the original rock mass joint image, respectively. The decoder upsamples progressively through transposed convolution and makes skip connections with the feature maps of the corresponding layers of the encoder. An attention gating module is introduced to enhance the features of the joint region. The output layer uses 1x1 convolution + Sigmoid activation to output a joint segmentation mask.

[0013] In some embodiments, the automatic identification of rock mass joint propagation trajectories using the improved U-Net machine learning model includes the following steps: The rock mass joint images were input into the trained improved U-Net machine learning model, and tensile strain > 0.005 and shear strain > 0.003 were set as joint initiation thresholds; Trajectory tracking is performed on the continuous frame recognition results, the joint extension coordinate sequence is fitted, and the extension rate and extension length are calculated.

[0014] In some embodiments, the improved particle image velocimetry combined with optical flow algorithm and displacement field data acquisition includes the following steps: The rock mass joint image is preprocessed, and cross calculation is performed using a multi-grid window. Subpixel fitting is then performed using cubic spline interpolation to complete the PIV calculation. The Horn-Schunck optical flow method was used to iteratively solve the optical flow equation, with a smoothing coefficient of 0.1 and a maximum number of iterations of 100, to complete the optical flow calculation. Set a local image gradient threshold and calculate the local image gradient magnitude. Use optical flow calculation results for the region where the gradient magnitude is greater than the gradient threshold, and use PIV calculation results for the region where the gradient magnitude is less than the gradient threshold to obtain the final displacement field data.

[0015] In some embodiments, the process of using a Bayesian inference inversion model to derive the mechanical parameters of the joint surface and evaluating the uncertainty of the inversion results using a Markov chain Monte Carlo method includes the following steps: A forward model is established with mechanical parameters as input and displacement field or strain field as output. The mechanical parameters include cohesion, internal friction angle and deformation modulus. Define the prior range of the mechanical parameters and assign them to the prior probability distribution; A likelihood function is constructed, and the displacement field data obtained by combining the improved particle image velocimetry with the optical flow algorithm is used as the observation value. The simulated value output by the forward model is calculated, and the sum of squared errors between the observed value and the simulated value is calculated. The Markov chain Monte Carlo method is used for iterative sampling. The parameter values ​​are adjusted according to the likelihood probability in each sampling, and finally the posterior distribution of mechanical parameters that satisfies the condition of "prior distribution + minimum observation error" is obtained. The mean of the parameters is extracted as the optimal value from the posterior distribution of the mechanical parameters, and the error is obtained by calculating the standard deviation of the parameters. The inversion result is then obtained. The inversion results are verified, and the sampling results are obtained by iterative sampling based on the Markov chain Monte Carlo method. The standard deviation and coefficient of variation of the posterior distribution of the parameters are calculated to quantify the uncertainty of the mechanical parameter estimation and output the uncertainty level.

[0016] On the other hand, the present invention also provides a simulation test method for the mechanical properties of rock joints based on the principle of body force field equivalence, applied to the aforementioned simulation test system for the mechanical properties of rock joints based on the principle of body force field equivalence, comprising the following steps: A geomechanical similarity model based on the principle of equivalent body force field is constructed. The geomechanical similarity model includes a base and a rock mass physical model. When the base moves relative to the interface of the rock mass physical model, the friction between the base and the bottom surface of the rock mass physical model generates uniformly distributed shear stress. The shear stress generates equivalent body force strength inside the rock mass physical model. Collect multidimensional sensing data when the base moves relative to the interface of the rock mass physical model; Based on multidimensional sensing data, and using an improved U-Net machine learning model, the propagation trajectory of rock joints is automatically identified. An improved particle image velocimetry combined with optical flow algorithm and displacement field data is used to invert the mechanical parameters of the joint surface using a Bayesian inference inversion model. The uncertainty of the inversion results is evaluated using the Markov chain Monte Carlo method.

[0017] The beneficial effects of this invention are: At the theoretical level, this invention has achieved a fundamental shift in physical simulation from experience to theory by establishing a complete mathematical and physical model based on the "equivalence principle of body force field". This breakthrough is due to the fact that, for the first time, starting from the basic principles of mechanics, a strict mathematical relationship between the interaction between the interface of the moving base and the equivalent body force field has been derived. A complete similarity criterion system including geometric similarity, mechanical similarity and dynamic similarity has been established, laying a solid foundation for the scientific interpretation and application of experimental results.

[0018] In terms of economics, this invention demonstrates significant cost advantages. The manufacturing cost of the entire system is only 5%-10% of that of traditional large-scale testing equipment, and energy consumption is reduced by more than 80%. This effect is achieved by adopting a standardized and modular system design, replacing the complex hydraulic servo system with a precision transmission system, which greatly reduces the equipment manufacturing and maintenance costs. At the same time, the system does not require special hydraulic oil source rooms, large reaction frames and other auxiliary facilities to operate, reducing site requirements and investment in supporting facilities, enabling research institutions and enterprises to conduct high-quality rock mechanics tests with a lower threshold.

[0019] Regarding the improvement in experimental efficiency, the cycle of a single experiment has been shortened from several days in the traditional method to several hours. This effect is due to the optimization of the experimental process, which has achieved standardization of model preparation, automation of the loading process, and intelligent data acquisition. Through parallel experimental design and rapid model change system, researchers can conduct multiple comparative experiments on the same platform, which improves research efficiency by dozens of times and greatly accelerates the scientific research process and engineering optimization decisions.

[0020] In terms of simulation capabilities, this invention can easily achieve continuous testing of large-scale models of 3m×2m. This technical effect is due to the developed distributed loading system and flexible boundary condition simulation technology, which can accurately reproduce the entire process of complex geological phenomena such as slope sliding, cavern collapse, and fault activation. The system achieves high-fidelity simulation of rock mass deformation and failure processes under complex geological conditions by precisely controlling the spatial distribution and time sequence of the equivalent force field.

[0021] In terms of data quality, this invention obtains massive amounts of continuous deformation data through non-contact full-field measurement. This breakthrough stems from the construction of a multi-source sensor fusion observation system, which organically combines high-speed photography, laser scanning, and digital image correlation technologies to achieve multi-scale observation from macroscopic deformation to microscopic fracture. The data density and quality collected by the system far exceed those of traditional point measurement methods, providing a rich information foundation for in-depth research on rock mass failure mechanisms. Attached Figure Description

[0022] Figure 1 This is a mechanical schematic diagram illustrating the principle of equivalent force field in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram comparing the recognition results in scenario 1 of embodiment 2 of the present invention; Figure 3 This is a schematic diagram comparing the recognition results in scenario 2 of embodiment 2 of the present invention; Figure 4 This is a schematic diagram comparing the recognition results in scenario 3 of embodiment 2 of the present invention; Figure 5 This is a flowchart of a simulation test method for the mechanical properties of rock joints based on the principle of equivalent body force field in Embodiment 5 of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] Example 1

[0025] This embodiment provides a simulation test system for the mechanical properties of rock joints based on the principle of body force field equivalence, including: The geomechanical similarity model construction unit is used to construct a geomechanical similarity model based on the principle of equivalent body force field. The geomechanical similarity model includes a base and a rock mass physical model. When the base moves relative to the interface of the rock mass physical model, the friction between the base and the bottom surface of the rock mass physical model generates uniformly distributed shear stress. The shear stress generates equivalent body force strength inside the rock mass physical model. A multi-dimensional sensor data acquisition unit is used to acquire multi-dimensional sensing data when the base moves relative to the interface of the rock mass physical model. The data processing and inversion unit is used to automatically identify the rock mass joint propagation trajectory based on multidimensional sensor data and an improved U-Net machine learning model. It also uses an improved particle image velocimetry combined with optical flow algorithm and displacement field data to invert the mechanical parameters of the joint surface using a Bayesian inference inversion model. Finally, it evaluates the uncertainty of the inversion results using the Markov chain Monte Carlo method.

[0026] See Figure 1 In this embodiment, through the interface interaction between the moving base and the rock mass physical model, an equivalent distributed body force field with similar mechanical effects to the gravitational field is dynamically generated inside the model. At this time: when the base moves at a constant velocity... v During movement, the friction between it and the bottom surface of the model will generate uniformly distributed shear stress. τ The equivalent body strength induced by this shear stress inside the modelg eq It can be represented as g eq = τ / ( ρ h )= μ σ / ( ρ h ),in ρ For the model material density, h For model feature height, μ Let σ be the interfacial friction coefficient and σ be the normal stress. Establishing this mathematical relationship allows for the quantitative control of the equivalent body force field through precise control of the base motion parameters.

[0027] In practical applications, in this embodiment, the substrate can be placed on a translation platform. The substrate moves at a constant speed, with a speed range of 0.1-100 mm / s. A servo motor combined with a precision planetary reduction mechanism can be used to drive a high-strength fiber-reinforced rubber conveyor belt or a surface-hardened anodized aluminum alloy translation platform to achieve uniform speed movement. Through a digital control interface based on a PID control algorithm, the substrate's movement speed can be precisely adjusted continuously within the range of 0.1-100 mm / s, achieving a speed control accuracy of ±0.5%. Simultaneously, the interface material of the rock mass physical model interface is replaceable, and different replaceable interface materials correspond to different friction coefficients. For example, the interface material can be quickly changed from smooth polytetrafluoroethylene (friction coefficient approximately 0.04) to sandpaper of different grits (friction coefficient 0.3-1.2), achieving continuous control of the interface friction characteristics and ensuring that the control accuracy of the equivalent body strength is better than ±5%.

[0028] For geomechanical similarity models, which are based on rigorous similarity theory, it is required to simultaneously satisfy geometric similarity, mechanical similarity, and boundary condition similarity. To this end, this embodiment uses a special rock mass similarity material system with quartz sand as aggregate, gypsum as binder, and barite powder as modifier. This material can achieve precise control of the elastic modulus of the rock mass physical model within the range of 10 MPa-1 GPa by adjusting the proportions. In addition, in terms of model manufacturing process, a joint network pre-embedded system based on 3D printing technology can be used. Complex joint morphologies can be accurately formed through soluble support materials. Layered air pressure compaction technology is used to ensure that the density uniformity error of the model is less than 3%. A laser engraving system is introduced to achieve accurate reproduction of the joint surface roughness JRC value, with an accuracy of ±0.5.

[0029] It should be noted that, in this embodiment, the multi-dimensional sensor data acquisition unit includes a high-speed camera and a laser displacement sensor array; The high-speed camera has a maximum acquisition frequency of 1000fps, and together with the large field-of-view deep-center lens, it ensures clear capture of the entire field of view of the dynamic deformation process, and is used to acquire image data of the rock mass physical model. The laser displacement sensor array consists of 30 laser displacement sensors arranged in a 5×6 matrix, evenly distributed on the surface of the rock mass physical model with a horizontal spacing of 15cm and a vertical spacing of 20cm. Each laser displacement sensor is perpendicular to the surface of the rock mass physical model. The laser spot diameter of the laser displacement sensor is 2-3mm, and the spatial resolution is 0.1mm. The laser displacement sensor array is used to acquire displacement data. In this embodiment, the laser displacement sensor array has a measurement accuracy of ±0.01mm, which can realize real-time monitoring of the three-dimensional deformation of the model.

[0030] Furthermore, in this embodiment, the high-speed camera and laser displacement sensor array are integrated into a digital image correlation deformation testing system. This system employs an adaptive mesh generation algorithm to perform multi-source data weighted fusion processing on the image data and displacement data of the rock mass physical model, obtaining rock mass joint images, which serve as the dataset for the improved U-Net machine learning model. The specially introduced digital image correlation (DIC) system, using an adaptive mesh generation algorithm, achieves a spatiotemporal resolution of 0.1 mm and 0.1 s for full-field strain analysis, effectively capturing the multi-scale evolution process from macroscopic deformation to microscopic fracture.

[0031] It should be noted that, in this embodiment, the improved U-Net machine learning model has a model structure including an encoder, a decoder, and an output layer; The encoder uses ResNet-50 as the backbone network to extract multi-scale feature maps, and the output feature map sizes are 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the original rock mass joint image, respectively. The decoder upsamples progressively through transposed convolution and makes skip connections with the feature maps of the corresponding layers of the encoder. An attention gating module is introduced to enhance the features of the joint region. The output layer uses 1x1 convolution + sigmoid activation to output a joint segmentation mask.

[0032] In addition, the automatic identification of rock mass joint propagation trajectories using the improved U-Net machine learning model includes the following steps: The rock mass joint images were input into the trained improved U-Net machine learning model, and tensile strain > 0.005 and shear strain > 0.003 were set as joint initiation thresholds; Trajectory tracking is performed on the continuous frame recognition results, the joint extension coordinate sequence is fitted, and the extension rate and extension length are calculated.

[0033] In addition, the improved particle image velocimetry combined with optical flow algorithm and displacement field data includes the following steps: The rock mass joint image is preprocessed, and cross calculation is performed using a multi-grid window. Subpixel fitting is then performed using cubic spline interpolation to complete the PIV calculation. The Horn-Schunck optical flow method was used to iteratively solve the optical flow equation, with a smoothing coefficient of 0.1 and a maximum number of iterations of 100, to complete the optical flow calculation. Set a local image gradient threshold and calculate the local image gradient magnitude. Use optical flow calculation results for the region where the gradient magnitude is greater than the gradient threshold, and use PIV calculation results for the region where the gradient magnitude is less than the gradient threshold to obtain the final displacement field data.

[0034] Finally, the mechanical parameters of the joint surface are inverted using the Bayesian inference inversion model, and the uncertainty of the inversion results is evaluated using the Markov chain Monte Carlo method, including the following steps: A forward model is established with mechanical parameters as input and displacement field or strain field as output. The mechanical parameters include cohesion, internal friction angle and deformation modulus. Define the prior range of the mechanical parameters and assign them to the prior probability distribution; A likelihood function is constructed, and the displacement field data obtained by combining the improved particle image velocimetry with the optical flow algorithm is used as the observation value. The simulated value output by the forward model is calculated, and the sum of squared errors between the observed value and the simulated value is calculated. The Markov chain Monte Carlo method is used for iterative sampling. The parameter values ​​are adjusted according to the likelihood probability in each sampling, and finally the posterior distribution of mechanical parameters that satisfies the condition of "prior distribution + minimum observation error" is obtained. The mean of the parameters is extracted as the optimal value from the posterior distribution of the mechanical parameters, and the error is obtained by calculating the standard deviation of the parameters. The inversion result is then obtained. The inversion results are verified, and the sampling results are obtained by iterative sampling based on the Markov chain Monte Carlo method. The standard deviation and coefficient of variation of the posterior distribution of the parameters are calculated to quantify the uncertainty of the mechanical parameter estimation and output the uncertainty level.

[0035] Example 2

[0036] Based on Example 1, this example provides a detailed description of the improved U-Net machine learning model.

[0037] The improved U-Net machine learning model in this embodiment adopts an improved U-Net architecture, combining an attention mechanism and residual connections, with the following specific structure: Encoder: Uses ResNet-50 as the backbone network to extract multi-scale feature maps; the output feature map sizes are 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the original image, respectively.

[0038] Decoder: It upsamples progressively through transposed convolution and skips connections with the feature maps of the corresponding layers of the encoder. It also introduces an attention gating module to enhance the features of the joint region.

[0039] Output layer: Uses 1x1 convolution + sigmoid activation to output a joint segmentation mask.

[0040] Parameter size: The total number of parameters is approximately 28.5 million.

[0041] In practical applications, it can include the following processing steps: (1) Data preprocessing: The input image is uniformly scaled to 512×512 pixels and normalized to [0,1].

[0042] Data augmentation: random rotation (±30°), horizontal flip, brightness jitter (±20%).

[0043] (2) Training process: Loss function: A combination of Dice Loss and Focal Loss is used: Dice Loss: improves boundary segmentation accuracy; Dice Loss: improves boundary segmentation accuracy; Dice Loss: Improves boundary segmentation accuracy; Training period: 100 epochs, batch size = 8.

[0044] (3) Reasoning stage: The input image is forward propagated through the model, and the output is a probability map.

[0045] Post-processing: Thresholding (0.5) → Morphological opening operation to remove noise → Output joint contour.

[0046] In this embodiment, when constructing the dataset, the sample size is as follows: a total of 1,850 rock joint images, with 1,480 images (80%) in the training set, 185 images (10%) in the validation set, and 185 images (10%) in the test set. After constructing the dataset, the following annotation process was performed: pixel-level semantic segmentation annotation was performed using the LabelMe tool; annotation categories: background (0), joint region (1). Furthermore, the annotation lines were required to be continuous and smooth, with a joint width ≥ 3 pixels.

[0047] Here, during model training, the learning rate is initially 1e-4, and cosine annealing scheduling is used. The number of iterations is 100 epochs, and the batch size is 8.

[0048] After the model training is complete, see [link / reference]. Figures 2-4 This embodiment provides a comparative schematic diagram of the recognition results for three sets of complex joint scenes, where: Scene 1: Intersecting joints (intersecting angle 60°); Scene 2: Bending joints (JRC=18); Scene 3: Dense network joints. Figures 2-4 As can be seen, in scenario 1, the existing RC-FCN model identifies the intersection as broken and mistakenly identifies shadows as joints, while the improved U-Net of this application identifies the intersection as continuous and complete with no shadow misjudgment. Therefore, it can be proved that this embodiment has strong robustness to complex topological structures. In scenario 2, the existing RC-FCN model identifies the joint path as discontinuous and roughness is lost, while the improved U-Net of this application identifies the bending path as completely restored and JRC morphology is preserved. Therefore, it can be proved that this embodiment can achieve high-precision boundary segmentation and morphology preservation. In scenario 3, the existing RC-FCN model identifies the partial omission of small joints and regional adhesion, while the improved U-Net of this application identifies all joints as clearly separated with no adhesion or missed detection. Therefore, it can be proved that this embodiment has superior multi-target resolution capability.

[0049] Additionally, refer to Table 1 for a comparison chart and table between this embodiment and existing models. The comparison object is the existing RC-FCN model, and the comparison indicators are: accuracy, intersection-over-union ratio (IoU), recall, and F1 score.

[0050] Table 1

[0051] As shown in Table 1, this embodiment significantly improves the recognition ability of complex joint morphology (such as intersections and bends) through attention mechanism and residual connection, with an accuracy improvement of about 3.9% and an IoU improvement of about 8.2%, thus it can be applied to large-scale, multi-joint scenarios. Example

[0052] Based on Example 1, this example provides a detailed description of how displacement field data was obtained by combining improved particle image velocimetry with optical flow algorithm in Example 1.

[0053] In this embodiment, the initial subpixel displacement estimation of PIV (Particle Image Velocimetry) is combined with the continuous motion field optimization of the optical flow algorithm, and high-precision displacement field quantization is achieved through a weighted fusion strategy.

[0054] In practical applications, displacement field data is obtained by combining improved particle image velocimetry with optical flow algorithms, which can be achieved through the following steps: Image preprocessing: Gaussian filtering for noise reduction, and histogram equalization to enhance contrast.

[0055] PIV calculation: Cross-calculation is performed using a multi-grid window (64×64 → 32×32 → 16×16 pixels). Subpixel fitting: Cubic spline interpolation is used, with an accuracy of 0.1 pixels.

[0056] Optical flow calculation: The Horn-Schunck optical flow method was used, with a smoothing coefficient λ=0.1; the optical flow equation was solved iteratively, with a maximum number of iterations of 100.

[0057] Fusion and optimization: Calculate the gradient magnitude of the local image as the weighting basis; prioritize the use of optical flow results in regions with high gradients and prioritize the use of PIV results in regions with low gradients; output the fused displacement field.

[0058] Here, compared with the traditional improved PIV method (for oil-water two-phase flow), this embodiment provides comparative data for various comparative indicators as shown in Table 2. Among them, the comparative indicators include: root mean square error of displacement field (RMSE), sub-pixel accuracy, and computational efficiency (frame processing time).

[0059] Table 2 As shown in Table 2, the fusion algorithm in this embodiment can significantly reduce displacement field error (RMSE reduced by 61%) while maintaining sub-pixel accuracy. Therefore, it is suitable for non-uniform deformation fields of rock masses and can effectively capture micro-fracture processes.

[0060] In addition, in this embodiment, the fusion strategy for PIV and optical flow fusion adopts a weighted adaptive fusion strategy: First, the displacement field of PIV is calculated separately. displacement field of optical flow Secondly, calculate the average gradient magnitude for each small region (e.g., 16x16 pixels) in the image. As the basis for fusion weights, the final displacement field is: Among them, weight From gradient Decision: In regions with flat textures (small G), trust PIV; in regions with rich textures and large gradients (large G), trust optical flow.

[0061] Furthermore, this embodiment can use a high-precision translation stage to create a known standard displacement field (such as an overall translation of 0.5 pixels or 1.0 pixels). The calibration process is as follows: First, apply this algorithm to the standard displacement image to calculate the average displacement of the entire field. Second, compare the calculated value with the true value to obtain the system error. Then, introduce a compensation term for the error into the algorithm output for correction.

[0062] After calibration, the results can be verified. As shown in Table 3, the standard displacement field verification data shows that after calibration, the root mean square error (RMSE) of the algorithm in the standard displacement field is ≤ 0.08 pixels, which meets the accuracy requirement of ≤ 0.1 pixels.

[0063] Table 3

[0064] Example 4 Based on Example 1, this example provides a detailed explanation of how to use a Bayesian inference inversion model to invert the mechanical parameters of joint surfaces and how to evaluate the uncertainty of the inversion results using the Markov chain Monte Carlo method.

[0065] First, mechanical parameter inversion is performed. The goal is to invert key mechanical parameters of the rock mass and joint surfaces based on micromechanical information, achieving the transformation from "data" to "parameters". This can be achieved through the following steps: Step 1: Establish a forward model: Based on the theory of elasticity of jointed rock mass, construct the "mechanical parameters (cohesion)" model. internal friction angle Deformation modulus The forward model of displacement / strain field is as follows: Input a set of mechanical parameters to calculate the corresponding simulated displacement / strain field. Step 2: Set the prior distribution: Based on similar material mix design tests and engineering experience, set the prior range of parameters: ∈[0,50] kPa、 ∈[20°, 45°]、 ∈[10MPa,1GPa], and assigned a uniform prior probability distribution; Step 3: Construct the likelihood function: Using the "global displacement / strain field output in step 4" as the observed value, calculate the sum of squared errors between the simulated value and the observed value output by the forward model. The smaller the error, the greater the likelihood probability (i.e., the closer the set of parameters is to the true value). Step 4: MCMC sampling to solve for the posterior distribution: The Markov Chain Monte Carlo (MCMC) algorithm is used for iterative sampling (iterations ≥ 10000). Each sampling adjusts the parameter values ​​according to the likelihood probability, and finally obtains the parameter posterior distribution that satisfies "prior distribution + minimum observation error". Step 5: Output the optimal parameters: Extract the mean of the parameters from the posterior distribution as the optimal values, and calculate the standard deviation of the parameters to obtain the error. Error ±2kPa Error ±0.5° Error ≤ 5%.

[0066] Then, the inversion results are verified and uncertainties are assessed to ensure the reliability of the results. The goal is to verify the rationality of the inversion parameters, quantify the uncertainty of the results, and provide a reliable basis for engineering decisions.

[0067] For result verification: the optimal mechanical parameters can be substituted into the forward model to calculate the simulated displacement / strain field, and then compared with the observed values ​​point by point to ensure that the overall error is ≤5%; Comparing the simulated joint extension trajectory with the observed trajectory, a coincidence rate of ≥90% is considered acceptable; For uncertainty assessment: First, based on the MCMC sampling results, calculate the standard deviation and coefficient of variation of the posterior distribution of the parameters to quantify the uncertainty of the parameter estimation; Second, output the uncertainty level (low / medium / high) and corresponding engineering application suggestions (e.g., low uncertainty parameters can be directly used in design, while high uncertainty parameters need to be supplemented with experimental verification); Then, finally output: mechanical parameter inversion report (including optimal parameters, errors, and verification results), uncertainty assessment report, and engineering application guidance.

[0068] For the Bayesian inference inversion model in this embodiment, its model structure includes a parameter space, a likelihood function, and posterior sampling, wherein: The parameter space includes: cohesion internal friction angle Deformation modulus ; Prior distributions include: The likelihood function is: ,in This represents the standard deviation of the observation error.

[0069] Posterior sampling: Adaptive Metropolis-Hastings algorithm is used for MCMC sampling.

[0070] In practical applications, the algorithm processing steps in this embodiment are as follows: Step 1: Parameter Initialization: Randomly draw an initial parameter vector from the prior distribution. .

[0071] Step 2, MCMC Sampling: Proposal Distribution: Multivariate normal distribution, covariance matrix is ​​adaptively updated every 1000 steps; Acceptance Probability: Calculate the posterior density ratio to determine whether to accept a new sample; Number of Sampling Chains: 4 independent chains, each chain is 50,000 iterations long.

[0072] Step 3, Convergence Diagnosis: Use the Gelman-Rubin statistic (R-hat) to assess convergence, requiring R-hat < 1.05.

[0073] Step 4: Parameter estimation and uncertainty assessment: The mean of the posterior sample is used as the parameter estimate; the 95% confidence interval is used as the uncertainty interval.

[0074] Furthermore, this embodiment also provides a comparison chart and table with existing models as shown in Table 4. The comparison objects include: traditional least squares inversion method (deterministic inversion), and the comparison indicators are: parameter inversion error, inversion time, and whether uncertainty assessment is provided.

[0075] Table 4 As shown in Table 4, this embodiment can achieve multi-parameter synchronous inversion with an error reduction of more than 60%, thus providing complete posterior distribution and uncertainty quantification to support engineering risk decision-making.

[0076] Furthermore, this embodiment also provides a comparison table of inversion parameters and measured values ​​for three sets of engineering cases, as shown in Table 5. The verification results from the three independent engineering cases in Table 5 show that the relative error between the inversion parameters and the measured values ​​is generally less than 3%, and for the first time, a quantitative uncertainty interval (95% confidence interval) is provided, which can provide reliable data support for engineering risk assessment and decision-making.

[0077] Example 5 Based on Example 1, this example provides a simulation test method for the mechanical properties of rock mass joints based on the principle of body force field equivalence. The flowchart is shown below. Figure 5 The method may include the following steps: S1. Construct a geomechanical similarity model based on the principle of equivalent body force field. The geomechanical similarity model includes a base and a rock mass physical model. When the base moves relative to the interface of the rock mass physical model, the friction between the base and the bottom surface of the rock mass physical model generates uniformly distributed shear stress. The shear stress generates equivalent body force strength inside the rock mass physical model. S2. Collect multi-dimensional sensing data when the base moves relative to the interface of the rock mass physical model; S3. Based on multidimensional sensing data, the improved U-Net machine learning model is used to automatically identify the rock mass joint propagation trajectory. The improved particle image velocimetry combined with optical flow algorithm and displacement field data is used to invert the mechanical parameters of the joint surface using a Bayesian inference inversion model. The uncertainty of the inversion results is evaluated by the Markov chain Monte Carlo method.

[0078] As can be seen from the description of Embodiment 1, the application scenario and implementation principle of this embodiment are the same as those of Embodiment 1, so they will not be repeated here.

[0079] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A simulation test system for the mechanical properties of rock mass joints based on the principle of equivalent force field, characterized in that, include: The geomechanical similarity model construction unit is used to construct a geomechanical similarity model based on the principle of equivalent body force field. The geomechanical similarity model includes a base and a rock mass physical model. When the base moves relative to the interface of the rock mass physical model, the friction between the base and the bottom surface of the rock mass physical model generates uniformly distributed shear stress. The shear stress generates equivalent body force strength inside the rock mass physical model. A multi-dimensional sensor data acquisition unit is used to acquire multi-dimensional sensing data when the base moves relative to the interface of the rock mass physical model. The data processing and inversion unit is used to automatically identify the rock mass joint propagation trajectory based on multidimensional sensor data and an improved U-Net machine learning model. It also uses an improved particle image velocimetry combined with optical flow algorithm and displacement field data to invert the mechanical parameters of the joint surface using a Bayesian inference inversion model. Finally, it evaluates the uncertainty of the inversion results using the Markov chain Monte Carlo method.

2. The rock mass joint mechanical property simulation test system based on the principle of body force field equivalence as described in claim 1, characterized in that, The substrate is placed on a translation stage and moves at a constant speed, with the speed range being 0.1-100 mm / s. The interface material of the rock mass physical model interface is replaceable, and different replaceable interface materials correspond to different friction coefficients.

3. The rock mass joint mechanical property simulation test system based on the principle of body force field equivalence as described in claim 1, characterized in that, The geomechanical similarity model satisfies geometric similarity, mechanical similarity, and boundary condition similarity. The rock mass physical model in the geomechanical similarity model has an elastic modulus ranging from 10 MPa to 1 GPa.

4. The rock mass joint mechanical property simulation test system based on the principle of body force field equivalence as described in claim 1, characterized in that, The multi-dimensional sensor data acquisition unit includes a high-speed camera and a laser displacement sensor array; The high-speed camera has a maximum acquisition frequency of 1000fps and is used to acquire image data of the rock mass physical model. The laser displacement sensor array consists of 30 laser displacement sensors arranged in a 5×6 matrix, evenly distributed on the surface of the rock mass physical model with a horizontal spacing of 15cm and a vertical spacing of 20cm. Each laser displacement sensor is perpendicular to the surface of the rock mass physical model. The laser spot diameter of the laser displacement sensor is 2-3mm, and the spatial resolution is 0.1mm. The laser displacement sensor array is used to acquire displacement data.

5. The rock mass joint mechanical property simulation test system based on the principle of body force field equivalence according to claim 4, characterized in that, The high-speed camera and laser displacement sensor array are integrated into the digital image correlation deformation testing system. The digital image correlation deformation testing system uses an adaptive mesh partitioning algorithm to perform multi-source data weighted fusion processing on the image data and displacement data of the rock mass physical model to obtain rock mass joint images, which are used as the dataset for the improved U-Net machine learning model.

6. The rock mass joint mechanical property simulation test system based on the principle of body force field equivalence as described in claim 5, characterized in that, The improved U-Net machine learning model has a model structure including an encoder, a decoder, and an output layer; The encoder uses ResNet-50 as the backbone network to extract multi-scale feature maps, and the output feature map sizes are 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the original rock mass joint image, respectively. The decoder upsamples progressively through transposed convolution and makes skip connections with the feature maps of the corresponding layers of the encoder. An attention gating module is introduced to enhance the features of the joint region. The output layer uses 1x1 convolution + Sigmoid activation to output a joint segmentation mask.

7. The rock mass joint mechanical property simulation test system based on the principle of body force field equivalence as described in claim 5, characterized in that, The method of automatically identifying rock mass joint propagation trajectories using the improved U-Net machine learning model includes the following steps: The rock mass joint images were input into the trained improved U-Net machine learning model, and tensile strain > 0.005 and shear strain > 0.003 were set as joint initiation thresholds; Trajectory tracking is performed on the continuous frame recognition results, the joint extension coordinate sequence is fitted, and the extension rate and extension length are calculated.

8. The rock mass joint mechanical property simulation test system based on the principle of body force field equivalence as described in claim 5, characterized in that, The improved particle image velocimetry combined with optical flow algorithm and displacement field data includes the following steps: The rock mass joint image is preprocessed, and cross calculation is performed using a multi-grid window. Subpixel fitting is then performed using cubic spline interpolation to complete the PIV calculation. The Horn-Schunck optical flow method was used to iteratively solve the optical flow equation, with a smoothing coefficient of 0.1 and a maximum number of iterations of 100, to complete the optical flow calculation. Set a local image gradient threshold and calculate the local image gradient magnitude. Use optical flow calculation results for the region where the gradient magnitude is greater than the gradient threshold, and use PIV calculation results for the region where the gradient magnitude is less than the gradient threshold to obtain the final displacement field data.

9. A simulation test system for the mechanical properties of rock joints based on the principle of equivalent force field as described in claims 5-8, characterized in that, The process of using a Bayesian inference inversion model to derive the mechanical parameters of the joint surface and evaluating the uncertainty of the inversion results using the Markov chain Monte Carlo method includes the following steps: A forward model is established with mechanical parameters as input and displacement field or strain field as output. The mechanical parameters include cohesion, internal friction angle and deformation modulus. Define the prior range of the mechanical parameters and assign them to the prior probability distribution; A likelihood function is constructed, and the displacement field data obtained by combining the improved particle image velocimetry with the optical flow algorithm is used as the observation value. The simulated value output by the forward model is calculated, and the sum of squared errors between the observed value and the simulated value is calculated. The Markov chain Monte Carlo method is used for iterative sampling. The parameter values ​​are adjusted according to the likelihood probability in each sampling, and finally the posterior distribution of mechanical parameters that satisfies the condition of "prior distribution + minimum observation error" is obtained. The mean of the parameters is extracted as the optimal value from the posterior distribution of the mechanical parameters, and the error is obtained by calculating the standard deviation of the parameters. The inversion result is then obtained. The inversion results are verified, and the sampling results are obtained by iterative sampling based on the Markov chain Monte Carlo method. The standard deviation and coefficient of variation of the posterior distribution of the parameters are calculated to quantify the uncertainty of the mechanical parameter estimation and output the uncertainty level.

10. A simulation test method for the mechanical properties of rock joints based on the principle of body force field equivalence, applied to the simulation test system for the mechanical properties of rock joints based on the principle of body force field equivalence as described in any one of claims 1-9, characterized in that, Includes the following steps: A geomechanical similarity model based on the principle of equivalent body force field is constructed. The geomechanical similarity model includes a base and a rock mass physical model. When the base moves relative to the interface of the rock mass physical model, the friction between the base and the bottom surface of the rock mass physical model generates uniformly distributed shear stress. The shear stress generates equivalent body force strength inside the rock mass physical model. Collect multidimensional sensing data when the base moves relative to the interface of the rock mass physical model; Based on multidimensional sensing data, and using an improved U-Net machine learning model, the propagation trajectory of rock joints is automatically identified. An improved particle image velocimetry combined with optical flow algorithm and displacement field data is used to invert the mechanical parameters of the joint surface using a Bayesian inference inversion model. The uncertainty of the inversion results is evaluated using the Markov chain Monte Carlo method.

Images