An intelligent initial support construction wet shotcrete overconsumption analysis method and system for tunnels

Abstract

The present invention discloses a method and system for analyzing over-consumption in wet spraying during intelligent initial support construction of tunnels. The method includes: obtaining the spatial distribution data of the spraying trajectory through a three-dimensional motion capture system installed on the spraying unit, and combining it with the material flow fluctuation curve generated by a dynamic weighing system to construct a three-dimensional feature tensor including material flow state characteristics, spraying process characteristics, and environmental interference characteristics; extracting the theoretical volume of the initial support structure based on the geological BIM model, integrating the data of overbreak and underbreak of the surrounding rock obtained by point cloud scanning, and establishing a dynamic usage benchmark model considering rebound loss through fluid mechanics simulation to generate a theoretical usage interval value with a confidence interval. The present invention constructs a three-dimensional feature map and a closed-loop optimization system by integrating multi-physical field dynamic data, solves the problems of inaccurate dynamic adaptation of geological conditions and the non-analyzable over-consumption mode caused by multi-factor coupling in wet spraying construction, and realizes the dynamic and precise control of material usage and the interpretable diagnosis of the cause of over-consumption.

Description

Technical Field

[0001] This application relates to the technical field of tunnel construction, and particularly to an intelligent wet shotcrete overconsumption analysis method and system for primary support construction in tunnels. Background Art

[0002] In the fields of underground engineering, tunnel construction, and mine exploitation, wet shotcrete has become one of the important materials for support structures due to its good bonding performance, strength, and durability. The wet shotcreting technology sprays the slurry onto the construction surface under high pressure, which can not only quickly form a support layer but also effectively prevent geological disasters such as collapses and debris flows. However, in the actual construction process, overconsumption of wet shotcrete usually occurs, which not only leads to material waste but also brings many challenges to project management and cost control.

[0003] The overconsumption phenomenon mainly stems from multiple factors, including improper material mixing ratio, unreasonable setting of construction equipment parameters, changes in environmental conditions, and insufficient operation experience of construction personnel. When using wet shotcrete, construction personnel can only rely on visual inspection and experience to judge the amount of concrete in most cases, which is easily affected by personal subjective factors and leads to improper use of materials. At the same time, during the construction process, the combined action of factors such as the real-time change of the overbreak and underbreak shape of the surrounding rock, the fluctuation of environmental temperature and humidity, and the difference in the permeability of the surrounding rock makes it difficult to dynamically correct the theoretical material usage benchmark. Existing fluid mechanics models are mostly constructed based on static geological parameters, without considering the spatio-temporal heterogeneity of the adsorption capacity of the surrounding rock and the time-varying characteristics of the material setting process, resulting in a low confidence level in the theoretical usage interval and being unable to accurately guide on-site construction; moreover, loss paths such as material rebound and slurry segregation are often jointly caused by factors such as mechanical movement trajectory deviation, material flow matching disharmony, and environmental interference. Existing technologies lack the ability to conduct multi-dimensional correlation analysis of process parameters, material flow states, and environmental physical fields, and it is difficult to extract interpretable overconsumption causes from composite data, resulting in insufficient pertinence of optimization measures. Summary of the Invention

[0004] To solve the above problems, an embodiment of the present invention provides an intelligent wet shotcrete overconsumption analysis method for primary support construction in tunnels, and the method includes:

[0005] Obtain the spatial distribution data of the spraying trajectory through a three-dimensional motion capture system installed on the spraying unit, and combine it with the material flow fluctuation curve generated by the dynamic weighing system to construct a three-dimensional feature tensor including material flow state characteristics, spraying process characteristics, and environmental interference characteristics;

[0006] Extract the theoretical volume of the primary support structure based on the geological BIM model, integrate the overbreak and underbreak data of the surrounding rock obtained by point cloud scanning, and establish a dynamic usage benchmark model considering rebound loss through fluid mechanics simulation to generate a theoretical usage interval value with a confidence interval;

[0007] Dynamically compare the material flow fluctuation curve with the theoretical usage interval value, extract the material flow state deviation degree, the jet trajectory coincidence degree, and the environmental interference intensity, input the material flow state deviation degree, the jet trajectory coincidence degree, and the environmental interference intensity into a convolutional neural network, and generate a three-dimensional feature map of the over-consumption pattern through the convolutional neural network;

[0008] Based on the three-dimensional feature map of the over-consumption pattern, use the multi-layer perceptron algorithm to match the preset typical over-consumption pattern library, generate a diagnostic report including the material loss path, process deviation nodes, and environmental interference intensity, and output an optimization suggestion set for parameters;

[0009] Input the optimization suggestion set into the jet parameter self-learning model, dynamically adjust the jet pressure, the nozzle movement speed, and the material flow matching coefficient through the reinforcement learning algorithm, form a closed-loop control loop, and precipitate it into the knowledge graph database.

[0010] Furthermore, the environmental interference characteristics include three physical field data of the surrounding rock permeability gradient field, the air flow disturbance intensity field, and the temperature gradient field. Construct an environmental interference characteristic subspace through the multi-physical field coupling analysis method and map it to the third dimension of the three-dimensional feature tensor.

[0011] Furthermore, in the construction steps of the dynamic usage benchmark model, introduce a time series analysis module, establish a spatio-temporal correlation relationship between the over-break and under-break data of the surrounding rock and the movement trajectory of the spraying unit, and generate a dynamic benchmark surface with time stamps.

[0012] Furthermore, the calculation of the material flow state deviation degree adopts a dual-channel convolution kernel structure. The first channel extracts the time-frequency characteristics of the material flow fluctuation curve, and the second channel captures the spatial distribution characteristics of the jet trajectory. Generate a composite deviation index through feature fusion.

[0013] Furthermore, the typical over-consumption pattern library adopts a hierarchical topological structure. The top-level nodes are classified according to the material loss path, the middle-level nodes are associated with the process deviation types, and the bottom-level nodes store the environmental interference intensity parameters.

[0014] Furthermore, in the process of generating the parameter optimization suggestion set, introduce a constraint satisfaction algorithm, establish a dynamic constraint relationship between the jet pressure, the movement speed, and the material flow matching coefficient, and generate an optimization solution set that satisfies the multi-objective balance.

[0015] Furthermore, the knowledge graph database adopts a spatio-temporal dual-index structure. The time dimension divides the knowledge units according to the construction stage, and the space dimension divides the knowledge nodes according to the tunnel section grid.

[0016] Furthermore, the closed-loop control loop is provided with a three-level optimization mechanism, which includes a primary optimization layer, an intermediate optimization layer, and a high-level optimization layer; the primary optimization layer adjusts the moving speed of the nozzle, the intermediate optimization layer corrects the material flow matching coefficient, and the high-level optimization layer reconstructs the injection pressure gradient field to form a progressive parameter optimization system.

[0017] A method for analyzing over-consumption of wet shotcreting in intelligent primary support construction of tunnels further includes:

[0018] The adsorption rate of the surrounding rock is monitored in real time through an embedded moisture sensor array, and the permeability parameters of the hydrodynamic model are dynamically corrected using an extended Kalman filter to generate a dynamic reference surface update mechanism with geological adaptability; the dielectric constant change curve of the sprayed concrete is measured using time domain reflectometry, and the hydration reaction process of the material is inverted to establish a dosage compensation coefficient considering the time-varying characteristics of the material, realizing the online self-calibration of the theoretical dosage interval value.

[0019] An over-consumption analysis system for intelligent primary support construction of tunnels, the system includes:

[0020] A multi-source data fusion module, which obtains the spatial distribution data of the spraying trajectory through a three-dimensional motion capture system installed on the spraying unit, and combines the material flow fluctuation curve generated by the dynamic weighing system to construct a three-dimensional feature tensor including material flow state characteristics, spraying process characteristics, and environmental interference characteristics;

[0021] A reference model construction module, which extracts the theoretical volume of the primary support structure based on the geological BIM model, integrates the over-excavation and under-excavation data of the surrounding rock obtained by point cloud scanning, and establishes a dynamic dosage reference model considering rebound loss through hydrodynamic simulation to generate a theoretical dosage interval value with a confidence interval;

[0022] A pattern construction module, which dynamically compares the material flow fluctuation curve with the theoretical dosage interval value, extracts the deviation degree of the material flow state, the coincidence degree of the spraying trajectory, and the environmental interference intensity, inputs the deviation degree of the material flow state, the coincidence degree of the spraying trajectory, and the environmental interference intensity into a convolutional neural network, and generates a three-dimensional feature map of the over-consumption pattern through the convolutional neural network;

[0023] A decision generation module, which based on the three-dimensional feature map of the over-consumption pattern, uses a multi-layer perceptron algorithm to match a preset typical over-consumption pattern library, generates a diagnostic report including the material loss path, process deviation nodes, and environmental interference intensity, and outputs a set of parameter optimization suggestions;

[0024] A closed-loop feedback optimization module, which inputs the set of optimization suggestions into a spraying parameter self-learning model, and dynamically adjusts the spraying pressure, the moving speed of the nozzle, and the material flow matching coefficient through a reinforcement learning algorithm to form a closed-loop control loop, and deposits it into the knowledge graph database.

[0025] The technical effects and advantages of a wet shotcrete overconsumption analysis method and system for intelligent primary support construction in tunnels provided by the present invention are as follows:

[0026] By integrating multi-physical field dynamic data to construct a three-dimensional feature map and a closed-loop optimization system, the present invention solves the problems of inaccurate dynamic adaptation of geological conditions and non-analyzable multi-factor coupling overconsumption patterns in wet shotcrete construction, realizes dynamic and precise control of material usage, and enables interpretable diagnosis of the causes of overconsumption. The present invention integrates geological BIM models, over-excavation and under-excavation point clouds of surrounding rocks, and multi-physical field environment data, and generates a four-dimensional dynamic reference surface through a spatio-temporal interpolation algorithm to realize real-time calibration of the theoretical usage interval value. By introducing online identification of the permeability gradient field and monitoring of the dielectric properties of materials, and combining the extended Kalman filter to dynamically correct the hydrodynamic boundary conditions, the theoretical usage benchmark synchronously responds to the evolution of geological conditions and material properties. Based on a three-dimensional feature tensor (material flow state / process trajectory / environmental interference) and a two-channel convolutional network, a three-dimensional feature map of space-time-intensity is generated, breaking through the limitations of traditional single-dimensional data analysis. A hierarchical overconsumption pattern library is constructed, and through a three-level reasoning link of material loss path → process deviation node → environmental interference intensity, accurate positioning of the causes of overconsumption is realized. A parameter self-learning model driven by reinforcement learning is adopted, combined with a spatio-temporal dual-index knowledge graph, to form a full-process closed loop of "perception-diagnosis-optimization-precipitation", and realize the continuous evolution of spraying parameters under complex working conditions. The three-level optimization mechanism (speed leveling → material flow shaping → pressure plastic field) progresses layer by layer, and generates a non-dominated solution set through multi-objective constraint solving, reducing the comprehensive rebound rate. Brief Description of the Drawings

[0027] Figure 1 It is a flow chart of a wet shotcrete overconsumption analysis method for intelligent primary support construction in tunnels in Embodiment 1;

[0028] Figure 2 It is a flow chart of a wet shotcrete overconsumption analysis method for intelligent primary support construction in tunnels in Embodiment 2;

[0029] Figure 3 It is a schematic connection diagram of a wet shotcrete overconsumption analysis system for intelligent primary support construction in tunnels in Embodiment 3. Detailed Embodiments

[0030] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] Embodiment 1: Please refer to Figure 1As shown in the figure, an embodiment of the present invention provides a method for analyzing the over-consumption of wet spraying in intelligent primary support construction of tunnels. The method includes:

[0032] Obtain the spatial distribution data of the spraying trajectory through a three-dimensional motion capture system installed on the spraying unit, and combine it with the material flow fluctuation curve generated by the dynamic weighing system to construct a three-dimensional feature tensor including material flow state characteristics, spraying process characteristics, and environmental interference characteristics;

[0033] Extract the theoretical volume of the primary support structure based on the geological BIM model, fuse the over-excavation and under-excavation data of the surrounding rock obtained by point cloud scanning, establish a dynamic usage benchmark model considering rebound loss through fluid mechanics simulation, and generate a theoretical usage interval value with a confidence interval;

[0034] Dynamically compare the material flow fluctuation curve with the theoretical usage interval value, extract the deviation degree of the material flow state, the coincidence degree of the spraying trajectory, and the environmental interference intensity, input the deviation degree of the material flow state, the coincidence degree of the spraying trajectory, and the environmental interference intensity into the convolutional neural network, and generate a three-dimensional feature map of the over-consumption pattern through the convolutional neural network;

[0035] Based on the three-dimensional feature map of the over-consumption pattern, use the multi-layer perceptron algorithm to match the preset typical over-consumption pattern library, generate a diagnostic report including the material loss path, process deviation nodes, and environmental interference intensity, and output an optimization suggestion set for the parameters;

[0036] Input the optimization suggestion set into the spraying parameter self-learning model, dynamically adjust the spraying pressure, nozzle movement speed, and material flow matching coefficient through the reinforcement learning algorithm, form a closed-loop control loop, and precipitate it into the knowledge graph database.

[0037] The environmental interference characteristics include the three physical field data of the surrounding rock permeability gradient field, air flow disturbance intensity field, and temperature gradient field. The environmental interference characteristic subspace is constructed through the multi-physical field coupling analysis method and mapped to the third dimension of the three-dimensional feature tensor.

[0038] In the process of implementing multi-source data perception and fusion, the movement trajectory of the nozzle in the three-dimensional space is captured in real time through a three-dimensional motion capture device, and combined with the concrete material flow fluctuation data collected by the dynamic weighing sensor, a three-dimensional feature tensor reflecting the comprehensive state of the spraying process is constructed; among them, the quantitative analysis of the environmental interference characteristics adopts the multi-physical field coupling modeling method, that is, the surrounding rock permeability gradient field data is obtained in real time through the geological sensor array arranged on the tunnel working surface to characterize the difference in the adsorption capacity of the surrounding rock in different sections for the sprayed concrete; the air flow disturbance intensity field in the spraying operation area is monitored by means of an air velocity sensor network to reflect the interference degree of the air flow caused by the ventilation system and mechanical movement on the spraying trajectory; at the same time, the temperature gradient field data of the spraying surface is collected by using an infrared thermal imaging device to depict the influence law of the environmental temperature distribution on the material setting speed.

[0039] After the dimensionality reduction of features of these three physical field datasets by the principal component analysis method, they are integrated into an environmental interference feature subspace and finally mapped to the third dimension of the three-dimensional feature tensor in the form of tensor encoding, forming a three-dimensional data representation structure that includes the material flow state features (the first dimension), the spraying trajectory features (the second dimension), and the environmental interference features (the third dimension); and

[0040] This composite feature tensor provides comprehensive data support covering geological conditions, mechanical motion, and environmental factors for the construction of the subsequent dynamic dosage benchmark model.

[0041] In the construction steps of the dynamic dosage benchmark model, a time series analysis module is introduced to establish a spatio-temporal correlation relationship between the overbreak and underbreak data of the surrounding rock and the motion trajectory of the spraying unit, generating a dynamic benchmark surface with time stamps.

[0042] During the implementation of the construction of the dynamic dosage benchmark model, the dynamic coupling relationship between geological data and the construction process is strengthened through spatio-temporal correlation modeling technology. Specifically, during implementation, the time series analysis module synchronizes the millimeter-level overbreak and underbreak data obtained by the laser scanning device on the surface of the surrounding rock and the six-degree-of-freedom trajectory data recorded by the motion capture system of the spraying unit in space and time, that is, by attaching spatial coordinate tags to each frame of point cloud data through a high-precision Beidou positioning device, and at the same time generating millisecond-level time stamps using the motion control signals output by the spraying robotic arm controller, establishing a dual mapping relationship of "spatial deformation - time trajectory".

[0043] Based on the spatio-temporal interpolation algorithm, the discrete overbreak and underbreak point cloud data and the continuous spraying trajectory are three-dimensionally matched to generate a dynamically continuous dynamic benchmark surface. The dynamic benchmark surface is based on the theoretical primary support contour of the BIM model and superimposes the real-time change amount of the surrounding rock morphology along the time axis of the spraying trajectory, forming a four-dimensional data model (three-dimensional space + time dimension) that can dynamically represent the interaction between geological conditions and construction status. When performing dynamic material consumption simulation on this surface through a fluid dynamics simulation engine, the environmental interference feature data (such as the permeability gradient field and the temperature field) under different time slices are synchronously fused, and finally a theoretical dosage interval value with a confidence interval that matches the construction progress is output, providing a spatio-temporally correlated quantitative benchmark for the dynamic comparison of overconsumption characteristics.

[0044] The calculation of the material flow state deviation degree adopts a dual-channel convolution kernel structure. The first channel extracts the time-frequency features of the material flow fluctuation curve, and the second channel captures the spatial distribution features of the spraying trajectory, generating a composite deviation index through feature fusion.

[0045] The first channel accesses the material flow fluctuation curve transmitted in real time by the dynamic weighing system. Through time-frequency transformation and sliding window analysis, the periodic fluctuation characteristics and instantaneous mutation characteristics of the material flow velocity are extracted. After the feature vectors output by the two channels are normalized, they are dynamically weighted and fused through a learnable weight matrix to form a composite deviation index representing the matching degree between the actual flow state of the material and the process parameters.

[0046] The second channel constructs a spatial distribution feature matrix from the spray trajectory point cloud data collected by the three-dimensional motion capture system, and uses multi-scale convolution kernels to extract spatial morphological characteristics such as the continuity and coverage density of the spray trajectory, and captures phenomena such as uneven spray layer thickness or local accumulation.

[0047] The material flow state deviation degree, the spray trajectory coincidence degree, and the environmental temperature and humidity interference intensity jointly constitute a three-dimensional feature vector, which is input into the deep convolutional network for feature dimensionality increase and pattern mapping, and finally a super-consumption pattern feature map with three-dimensional attributes of space-time-intensity is generated.

[0048] The typical super-consumption pattern library adopts a hierarchical topological structure. The top-level nodes are classified according to the material loss path, the middle-level nodes are associated with the process deviation types, and the bottom-level nodes store the environmental interference intensity parameters.

[0049] The typical super-consumption pattern library organizes super-consumption characteristics in a three-level architecture, including the top level, the middle level, and the bottom level; the top-level nodes are divided into main categories such as aggregate rebound, slurry segregation, and interfacial adhesion failure according to the material loss path. Under each main category, middle-level nodes directly associated with the spray process parameters are mounted, including process deviation types such as spray pressure mismatch, nozzle movement speed deviation, and accelerator dosage abnormality; the bottom-level nodes are then bound with the quantified parameters of the environmental interference intensity, such as dynamic monitoring data such as the surrounding rock seepage rate, air temperature and humidity gradient, and equipment vibration amplitude.

[0050] When the three-dimensional feature tensor is input into the multi-layer perceptron model, the multi-layer perceptron model first filters the main abnormal direction through the activation function of the material loss path of the top-level node, then calculates the deviation weight between the feature vector and the preset process threshold along the middle-level node, and finally combines the environmental parameters of the bottom-level node to correct the interference intensity, and outputs a multi-dimensional diagnostic label such as "aggregate rebound-nozzle movement speed overload-seepage interference>45%".

[0051] The three-level architecture supports drilling down layer by layer from the macroscopic loss type to the microscopic process parameters, while retaining the corrective effect of environmental factors on decision-making, and provides an interpretable reasoning path for the generation of the parameter optimization suggestion set.

[0052] During the generation of the parameter optimization suggestion set, a constraint satisfaction algorithm is introduced to establish dynamic constraint relationships among the spray pressure, movement speed, and material flow matching coefficient, and an optimization solution set that meets multi-objective balance is generated.

[0053] Based on the process deviation nodes and environmental interference intensity parameters in the diagnostic report, first establish a dynamic constraint matrix for the injection pressure, nozzle movement speed, and material flow matching coefficient. The establishment method includes:

[0054] Associate the injection pressure threshold with the over-excavation and under-excavation data of the surrounding rock to form a spatial constraint condition. Establish a kinematic constraint between the nozzle movement speed interval and the trajectory coincidence degree in the three-dimensional feature tensor. The material flow matching coefficient constitutes a mass conservation constraint through the data of the dynamic weighing system and the fluid mechanics model.

[0055] The constraint satisfaction algorithm iteratively solves the Pareto front of the injection pressure, nozzle movement speed, and material flow matching coefficient to seek a balance point among suppressing aggregate rebound (material loss path), ensuring injection density (process target), and controlling equipment energy consumption (environmental constraint);

[0056] The feasible solutions generated in each iteration are simulated for injection efficiency through a reinforcement learning agent, and indicators such as the reduction gradient of the rebound rate, the improvement coefficient of the injection efficiency, and the increment of the interface density are calculated. Finally, a parameter optimization scheme including 3 - 5 sets of non-dominated solutions is output. Each set of solutions is marked with a "pressure - speed - material flow" collaborative adjustment strategy and expected improvement indicators. For example, the scheme of "increasing the movement speed by 12% + dynamically matching the material flow coefficient" corresponds to a predicted effect of an 8.2% reduction in the rebound rate and a 15% increase in the injection efficiency, for the operator to select and execute according to the on-site working conditions, and the optimization results are fed back to the knowledge graph database to improve the constraint relationship model.

[0057] The knowledge graph database adopts a spatio-temporal dual-index structure. In the time dimension, knowledge units are divided according to the construction stage, and in the space dimension, knowledge nodes are divided according to the tunnel cross-section grid.

[0058] The time dimension index is based on the tunnel construction stage as the division benchmark. The injection parameters, environmental monitoring data, and process optimization schemes are segmented into dynamic knowledge units according to the construction process stages such as initial shotcrete support, secondary shotcrete reinforcement, and invert construction. Each dynamic knowledge unit embeds a construction time sequence chain to record the evolution trajectory of the process parameters.

[0059] The space dimension index constructs a cross-section grid coordinate system based on the tunnel three-dimensional laser scanning data. Each 1m×1m grid cell is used as an independent knowledge node to store spatial feature data such as the historical injection trajectory point cloud, material loss heat map, and execution effect of the optimization parameters in this area.

[0060] When conducting quality traceability or parameter tuning, multi-dimensional retrieval is achieved through the spatio-temporal association engine. Exemplary:

[0061] Input the spatio-temporal coordinates of "Re-spray reinforcement stage + mileage stake number K32+150". When it is detected that there is insufficient local density in the inverted arch section at mileage stake number K32+150, the operator inputs the spatio-temporal coordinate parameters through the man-machine interface: select "Re-spray reinforcement stage (2023 / 05 / 12 - 05 / 15)" in the time dimension, and locate to the 7th circumferential partition and longitudinal grid G-235 in the space dimension. The spatio-temporal correlation engine performs the following multi-dimensional retrieval;

[0062] Retrieve the accelerator admixture dosage records (fluctuation range of 3.8% - 4.2%) and corresponding interface strength test data (average value of 28 MPa) of this grid in the initial spraying stage (2023 / 05 / 08), and it is found that the admixture dosage in the re-spraying stage is increased to 4.5% but the strength does not reach the expectation (31 MPa compared with the design value of 35 MPa);

[0063] Retrieve the time series data of the vibration amplitude of the spraying robotic arm (0.15 - 0.25 g) of adjacent grids G-234 and G-236, and it is found that there is an abnormal vibration peak (0.38 g @ 2023 / 05 / 13 14:23) in grid G-235 during re-spraying;

[0064] Combined with the equipment operation and maintenance records in the knowledge graph, locate the hydraulic pump pressure fluctuation event corresponding to the abnormal vibration period, and associate the failure mode of "vibration amplitude > 0.3 g resulting in slurry segregation" in the historical case library, and automatically generate an optimization suggestion with spatio-temporal tags: "During the re-spraying operation of grid G-235, lower the spraying pressure by 8% - 10% to compensate for the impact of hydraulic fluctuations, simultaneously increase the accelerator admixture dosage to 5.0% and extend the curing time by 2 hours"; this plan is displayed through a three-dimensional visualization interface, where the red highlighted area shows the current problem grid, the blue semi-transparent layer superimposes and shows the vibration data trend line of adjacent grids, and the yellow arrow marks the adjustment direction of the optimization parameters. The quality re-inspection data after execution (such as 34.7 MPa) is automatically transmitted back to the knowledge node to form a closed-loop learning link.

[0065] The spatio-temporal dual-index structure supports construction personnel to quickly locate the "spatio-temporal blind area", such as the material segregation mode caused by equipment vibration at the intersection of the inverted arch and the side wall, and generate a targeted process optimization plan through cross-stage knowledge transfer; all newly generated abnormal diagnosis results and optimization parameter sets are automatically archived according to spatio-temporal tags, forming a knowledge network with spatio-temporal evolution characteristics, providing spatio-temporally relevant experience support for the intelligent decision-making of subsequent construction sections.

[0066] The closed-loop control loop sets a three-level optimization mechanism, and the three-level optimization mechanism includes a primary optimization layer, a secondary optimization layer, and a tertiary optimization layer; the primary optimization layer adjusts the moving speed of the nozzle, the secondary optimization layer corrects the material flow matching coefficient, and the tertiary optimization layer reconstructs the spraying pressure gradient field to form a progressive parameter optimization system.

[0067] The primary speed optimization layer drives the dynamic compensation of the nozzle movement speed through the spray trajectory deviation data obtained by the laser scanner; when the three-dimensional point cloud analysis shows an over-excavated contour area, speed adjustment instructions are generated according to the principle of "under-spray deceleration - over-excavation acceleration". For example, at the vault joint, the translation speed of the nozzle is automatically reduced to enhance the material accumulation effect. At the same time, the vibration spectrum of the robotic arm is monitored in real time through the inertial measurement unit (IMU) to prevent equipment resonance caused by sudden speed changes.

[0068] On the basis of speed adjustment, the intermediate material flow optimization layer corrects the material flow matching coefficient by using the fuzzy control algorithm according to the aggregate consumption rate feedback by the dynamic weighing system and the slurry rheological property data of the pressure sensor; when it is monitored that the slurry viscosity increases after the addition of the accelerator, the coupling coefficient of the rotation speed of the conveying screw and the air supply volume of the air compressor is automatically increased to ensure that the sprayed material flow maintains a stable jet shape during the acceleration setting stage, and the infrared thermal imager is used to detect the interface temperature field distribution to verify the improvement effect of the material flow matching on the uniformity of the hydration reaction.

[0069] Based on the spatio-temporal parameter set formed by the first two levels of optimization, the advanced pressure field optimization layer constructs a dynamic pressure gradient field model through computational fluid dynamics simulation; the dynamic pressure gradient field model divides the tunnel section into pressure control domains, and applies a gradient pressurization strategy in the weak surrounding rock area (such as increasing the pressure value by 12% - 15% at the arch foot to form a compaction and anchorage area). At the same time, combined with the development direction of the rock mass fissures detected by the geological radar, an intelligent pressure vector distribution map is generated to guide the multi-axis coordinated movement of the nozzle in different spatial postures. The optimized pressure field parameters are verified by virtual spraying through the digital twin system, and finally a global pressure configuration plan with weight coefficients is generated.

[0070] The three-level optimization mechanism forms a progressive optimization path of "speed leveling - material flow shape stabilization - pressure field shaping" by sharing the process constraint conditions in the spatio-temporal index knowledge nodes. The vibration spectrum, material consumption rate and interface strength data generated by each spraying operation are all fed back to the knowledge graph to drive the adaptive update of the constraint matrix, realizing the continuous evolution of the process parameters in the spatio-temporal dimension.

[0071] Example 2: As Figure 2 shown, this embodiment further improves the design on the basis of Embodiment 1. The difference is that in the actual operation of Embodiment 1, it is found that there is a confidence drift phenomenon in the theoretical dosage interval of the dynamic dosage benchmark model in the complex bedding geological section, mainly due to the anisotropy of the surrounding rock permeability leading to the distortion of the fluid simulation boundary conditions. Based on this, a wet spraying over-consumption analysis method for intelligent primary support construction of tunnels also includes:

[0072] The adsorption rate of surrounding rock is monitored in real time through an embedded moisture sensor array, and the permeability parameters of the hydrodynamic model are dynamically corrected using an extended Kalman filter to generate a dynamic reference surface update mechanism with geological adaptability; the dielectric constant change curve of the sprayed concrete is measured using time domain reflectometry to invert the hydration reaction process of the material, and a dosage compensation coefficient considering the time-varying characteristics of the material is established to realize the online self-calibration of the theoretical dosage interval value.

[0073] The dynamic correlation correction mechanism between the adsorption rate of surrounding rock and the permeability gradient field strengthens the adaptability of the hydrodynamic simulation model to complex geological conditions; while the time-varying compensation model of materials inverted based on dielectric properties constructs a closed-loop feedback link between the evolution of material properties and the dosage reference; the two work together, enabling the dynamic reference surface to synchronously respond to the change in the adsorption capacity of the surrounding rock and the attenuation of the performance of the sprayed material. Finally, through the spatio-temporal index structure of the knowledge graph database, the corrected reference data is compared with historical construction parameters across stages to form a dosage prediction system with self-evolution ability.

[0074] Example 3: As Figure 3 shown, based on the same inventive concept as a wet shotcrete overconsumption analysis method for intelligent primary support construction in a tunnel in the foregoing embodiment, the present application provides a wet shotcrete overconsumption analysis system for intelligent primary support construction in a tunnel. The system in the embodiment of the present application and the method embodiment are based on the same inventive concept. Among them, the system includes:

[0075] A multi-source data fusion module. The multi-source data fusion module obtains the spatial distribution data of the spraying trajectory through a three-dimensional motion capture system installed on the spraying unit, and combines the material flow fluctuation curve generated by the dynamic weighing system to construct a three-dimensional feature tensor including material flow state characteristics, spraying process characteristics, and environmental interference characteristics;

[0076] A reference model construction module. The reference model construction module extracts the theoretical volume of the primary support structure based on the geological BIM model, integrates the overbreak and underbreak data of the surrounding rock obtained by point cloud scanning, and establishes a dynamic dosage reference model considering rebound loss through hydrodynamic simulation to generate a theoretical dosage interval value with a confidence interval;

[0077] A map construction module. The map construction module dynamically compares the material flow fluctuation curve with the theoretical dosage interval value, extracts the deviation degree of the material flow state, the coincidence degree of the spraying trajectory, and the environmental interference intensity, and inputs the deviation degree of the material flow state, the coincidence degree of the spraying trajectory, and the environmental interference intensity into a convolutional neural network to generate a three-dimensional feature map of the overconsumption pattern through the convolutional neural network;

[0078] A decision generation module. The decision generation module is based on the three-dimensional feature map of the overconsumption pattern, uses a multi-layer perceptron algorithm to match a preset typical overconsumption pattern library, generates a diagnostic report including the material loss path, process deviation nodes, and environmental interference intensity, and outputs a set of parameter optimization suggestions;

[0079] The closed-loop feedback optimization module inputs the optimization suggestion set into the injection parameter self-learning model, dynamically adjusts the injection pressure, the moving speed of the nozzle, and the material flow matching coefficient through the reinforcement learning algorithm, forms a closed-loop control loop, and precipitates to the knowledge graph database.

[0080] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

[0081] The above-mentioned are only the preferred specific embodiments of the embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application, according to the technical solution and its concept of the present application, makes equivalent substitutions or changes, and should be covered by the protection scope of the present application.

Claims

1. A method for analyzing the over-consumption of wet spraying in the intelligent initial support construction of tunnels, characterized in that, Including: Obtain the spatial distribution data of the spraying trajectory through a three-dimensional motion capture system installed on the spraying unit, and combine it with the material flow fluctuation curve generated by the dynamic weighing system to construct a three-dimensional feature tensor including material flow state characteristics, spraying process characteristics, and environmental interference characteristics; Extract the theoretical volume of the primary support structure based on the geological BIM model, fuse the over-excavation and under-excavation data of the surrounding rock obtained by point cloud scanning, and establish a dynamic consumption benchmark model considering rebound loss through fluid mechanics simulation to generate a theoretical consumption interval value with a confidence interval; Dynamically compare the material flow fluctuation curve with the theoretical consumption interval value, extract the material flow state deviation degree, spraying trajectory coincidence degree, and environmental interference intensity, and input the material flow state deviation degree, spraying trajectory coincidence degree, and environmental interference intensity into a convolutional neural network to generate a three-dimensional feature map of the over-consumption pattern through the convolutional neural network; Based on the three-dimensional feature map of the over-consumption pattern, use a multi-layer perceptron algorithm to match a preset typical over-consumption pattern library, generate a diagnostic report including the material loss path, process deviation nodes, and environmental interference intensity, and output an optimization suggestion set of parameters; Input the optimization suggestion set into the spraying parameter self-learning model, and dynamically adjust the spraying pressure, nozzle movement speed, and material flow matching coefficient through a reinforcement learning algorithm to form a closed-loop control loop and deposit it into the knowledge graph database.

2. The wet shotcrete overconsumption analysis method for intelligent primary support construction of a tunnel according to claim 1, characterized in that The environmental interference characteristics include the data of three physical fields: the surrounding rock permeability gradient field, the air flow disturbance intensity field, and the temperature gradient field. A subspace of environmental interference characteristics is constructed through a multi-physics field coupling analysis method and mapped to the third dimension of the three-dimensional feature tensor.

3. A wet shotcrete overconsumption analysis method for intelligent primary support construction in tunnels according to claim 1, characterized in that, In the construction steps of the dynamic consumption benchmark model, a time series analysis module is introduced to establish a spatio-temporal correlation relationship between the over-excavation and under-excavation data of the surrounding rock and the movement trajectory of the spraying unit, and generate a dynamic benchmark surface with a timestamp.

4. A wet shotcrete overconsumption analysis method for intelligent primary support construction in tunnels according to claim 1, characterized in that: The calculation of the material flow state deviation degree adopts a dual-channel convolution kernel structure. The first channel extracts the time-frequency characteristics of the material flow fluctuation curve, and the second channel captures the spatial distribution characteristics of the spraying trajectory. Generate a composite deviation index through feature fusion.

5. A wet shotcrete overconsumption analysis method for intelligent primary support construction in tunnels according to claim 1, characterized in that: The typical over-consumption pattern library adopts a hierarchical topological structure. The top-level nodes are classified according to the material loss path, the middle-level nodes are associated with the process deviation types, and the bottom-level nodes store the environmental interference intensity parameters.

6. The wet shotcrete overconsumption analysis method for intelligent primary support construction of a tunnel according to claim 1, characterized in that: In the generation process of the parameter optimization suggestion set, a constraint satisfaction algorithm is introduced to establish a dynamic constraint relationship between the spraying pressure, movement speed, and material flow matching coefficient, and generate an optimization solution set that meets the multi-objective balance.

7. A wet shotcrete overconsumption analysis method for intelligent primary support construction in tunnels according to claim 1, characterized in that: The knowledge graph database adopts a spatio-temporal dual-index structure. The time dimension divides knowledge units according to the construction stage, and the spatial dimension divides knowledge nodes according to the tunnel section grid.

8. A wet shotcrete overconsumption analysis method for intelligent primary support construction in tunnels according to claim 1, characterized in that: The closed-loop control loop is set with a three-level optimization mechanism, which includes a primary optimization layer, a middle-level optimization layer, and a high-level optimization layer; the primary optimization layer adjusts the nozzle movement speed, the middle-level optimization layer corrects the material flow matching coefficient, and the high-level optimization layer reconstructs the spraying pressure gradient field to form a progressive parameter optimization system.

9. A wet shotcrete overconsumption analysis method for intelligent primary support construction of tunnels according to claim 1, characterized in that, Also including: Real-time monitor the adsorption rate of surrounding rock through an embedded moisture sensor array, dynamically correct the permeability parameters of the hydrodynamic model using an extended Kalman filter, and generate a dynamic benchmark surface update mechanism with geological adaptability; measure the dielectric constant change curve of the sprayed concrete using time-domain reflectometry, invert the hydration reaction process of the material, establish a dosage compensation coefficient considering the time-varying characteristics of the material, and achieve online self-calibration of the theoretical dosage interval value.

10. An intelligent wet shotcrete overconsumption analysis system for tunnel initial support construction, characterized in that, The system includes: A multi-source data fusion module that obtains the spatial distribution data of the spraying trajectory through a three-dimensional motion capture system installed on the spraying unit, and combines the material flow fluctuation curve generated by the dynamic weighing system to construct a three-dimensional feature tensor containing the material flow state characteristics, spraying process characteristics, and environmental interference characteristics. A benchmark model construction module that extracts the theoretical volume of the primary support structure based on the geological BIM model, integrates the over-excavation and under-excavation data of the surrounding rock obtained by point cloud scanning, and establishes a dynamic dosage benchmark model considering rebound loss through hydrodynamic simulation to generate a theoretical dosage interval value with a confidence interval. A pattern construction module that dynamically compares the material flow fluctuation curve with the theoretical dosage interval value, extracts the material flow state deviation degree, spraying trajectory coincidence degree, and environmental interference intensity, inputs the material flow state deviation degree, spraying trajectory coincidence degree, and environmental interference intensity into a convolutional neural network, and generates a three-dimensional feature map of the over-consumption pattern through the convolutional neural network. A decision generation module that, based on the three-dimensional feature map of the over-consumption pattern, uses a multi-layer perceptron algorithm to match a preset typical over-consumption pattern library, generates a diagnostic report containing the material loss path, process deviation nodes, and environmental interference intensity, and outputs a set of parameter optimization suggestions. A closed-loop feedback optimization module that inputs the set of optimization suggestions into the spraying parameter self-learning model, dynamically adjusts the spraying pressure, nozzle movement speed, and material flow matching coefficient through a reinforcement learning algorithm to form a closed-loop control loop, and deposits it in the knowledge graph database.

Images