Tunneling machine cooperative operation real-time optimization method and system based on digital twinning

Abstract

The application discloses a tunneling machine cooperative operation real-time optimization method and system based on digital twinning, and relates to the technical field of digital twinning. The method comprises the following steps: collecting multi-source sensing data in real time, and constructing a neural Gaussian force field digital twin; combining force prediction dynamic analysis of multi-machine operation state to generate cooperative path instructions; using adaptive impedance control to adjust hydraulic pressure to obtain cooperative parameters; and optimizing tunneling machine cluster operation according to the cooperative parameters. The technical problems of lacking real-time perception and prediction ability of rock layer physical properties in the existing tunneling machine cluster cooperative operation process, leading to path planning lag, unreasonable hydraulic torque distribution and uneven tool wear are solved, the technical effects of realizing rock hardness online identification and future force prediction based on digital twinning and neural Gaussian force field, improving multi-machine cooperative path planning accuracy and impedance control adaptive ability, and realizing tunneling efficiency improvement, operation safety enhancement and tool wear balance are achieved.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, specifically to a real-time optimization method and system for collaborative operation of tunneling machines based on digital twins. Background Technology

[0002] With the continuous expansion of underground mining, tunnel engineering, and large-scale underground space construction, tunnel boring machines (TBMs) are gradually developing towards clustering and collaboration to improve construction efficiency and resource utilization. In complex geological environments, the coordinated operation of multiple TBMs has become an important means of accelerating project progress. However, rock strata structures exhibit significant heterogeneity and uncertainty. The hardness, joint and fracture development, and stress distribution of rock strata vary significantly in different areas, leading to large fluctuations in cutter load, uneven wear, and increased risk of equipment interference during tunneling. In severe cases, this can even cause equipment failure and construction safety issues.

[0003] Existing tunneling machine control systems are mostly based on single-machine closed-loop control modes, relying primarily on preset parameters or simple feedback signals for adjustment. They lack the ability to accurately perceive the physical properties of rock strata in real time, making it difficult to achieve forward-looking predictions for complex working conditions. Furthermore, in multi-machine collaborative operation scenarios, traditional path planning methods are typically based on static environmental models, failing to comprehensively consider the dynamic state of the equipment and future stress changes. This easily leads to spatiotemporal conflicts and interference risks, resulting in low collaborative efficiency. Summary of the Invention

[0004] This application provides a real-time optimization method and system for collaborative tunneling machine operations based on digital twins, which solves the technical problems of lack of real-time perception and prediction capabilities of rock strata physical properties during existing collaborative tunneling machine cluster operations, resulting in path planning lag, unreasonable hydraulic torque distribution, and uneven tool wear.

[0005] The first aspect of this application provides a real-time optimization method for collaborative operation of tunneling machines based on digital twins, the method comprising:

[0006] Multi-source sensor data from a tunneling machine cluster is collected in real time using multiple sensors, and a digital twin is constructed based on a neural Gaussian force field. The multi-machine operating status is dynamically analyzed within this digital twin, extracting the propulsion speed characteristic set and cutter load characteristic set of the tunneling machine cluster. Combined with force prediction, the future force field distribution is obtained, and a spatiotemporal conflict interference risk analysis is performed to adjust the path and generate collaborative path planning instructions. An adaptive impedance control algorithm is used to adjust the hydraulic system pressure of each tunneling machine in the cluster, obtaining impedance collaborative control parameters. These parameters are dynamically adjusted based on the online rock hardness identification results output from the explicit force field representation in the digital twin. The collaborative operation of the tunneling machine cluster is optimized based on the impedance collaborative control parameters and the collaborative path planning instructions.

[0007] A second aspect of this application provides a real-time optimization system for collaborative operation of tunneling machines based on digital twins, the system comprising:

[0008] The module comprises the following components: a digital twin construction module, a path planning module, and an operation twin. The latter uses a multi-source sensor to collect real-time multi-source sensor data from the tunneling machine cluster and constructs a digital twin based on a neural Gaussian force field. The former dynamically analyzes the multi-machine operation status within the digital twin, extracts the propulsion speed characteristic set and cutter load characteristic set of the tunneling machine cluster, combines force prediction to obtain the future force field distribution, performs spatiotemporal conflict interference risk analysis, adjusts the path, and generates collaborative path planning instructions. The latter uses an adaptive impedance control algorithm to adjust the hydraulic system pressure of each tunneling machine in the cluster, obtaining impedance collaborative control parameters. These parameters are dynamically adjusted based on the online identification results of rock hardness output from the explicit force field representation in the digital twin. The latter optimizes the collaborative operation of the tunneling machine cluster based on the impedance collaborative control parameters and the collaborative path planning instructions.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] First, multi-source sensors are used to collect real-time data on the operating status of the tunneling machine cluster, cutter stress, and rock strata information. A high-fidelity digital twin reflecting the rock-cutter interaction is then constructed using a neural Gaussian force field model. Within this digital twin environment, the operating status of multiple tunneling machines is dynamically analyzed, extracting propulsion speed and cutter load characteristics. Force prediction based on an explicit force field model yields the future force field distribution, generating multi-machine collaborative path planning instructions to proactively avoid interference risks. Next, an adaptive impedance control algorithm adjusts the hydraulic system pressure of each tunneling machine, forming impedance collaborative control parameters. These parameters are adjusted in real-time based on the online rock hardness identification results output by the digital twin, achieving adaptive response to different lithological conditions. Finally, based on the optimized impedance collaborative control parameters, collaborative operation control is implemented for the tunneling machine cluster, thereby improving overall tunneling efficiency and achieving cutter load balancing. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram of the real-time optimization method for collaborative operation of tunneling machines based on digital twins, provided in an embodiment of this application.

[0013] Figure 2 A schematic diagram of the structure of a real-time optimization system for collaborative tunneling machine operation based on digital twins, provided in an embodiment of this application.

[0014] Figure labeling: Twin construction module 11, path planning module 12, impedance control module 13, job optimization module 14. Detailed Implementation

[0015] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0016] Example 1, as Figure 1 As shown, this application provides a real-time optimization method for collaborative operation of tunneling machines based on digital twins, wherein the method includes:

[0017] A digital twin is constructed based on a neural Gaussian force field by collecting multi-source sensor data of a tunneling machine cluster in real time using multi-source sensors.

[0018] In this embodiment, multi-source sensors are used to perceive the tunneling machine cluster and its operating environment in real time, collecting multi-source sensor data for constructing a digital twin. This multi-source sensor data includes at least image information acquired by a multi-view RGB-D camera, tool load information acquired by a fiber optic strain sensor, and rock vibration information acquired by a triaxial accelerometer. During the acquisition process, the data from each sensor are uniformly timestamped and calibrated, and noise filtering, anomaly removal, and scale normalization are performed. Then, image, pose, load, and vibration information are synchronously integrated to form multi-source sensor data describing "equipment status - cutting load - rock mass response". Subsequently, based on the obtained multi-source sensor data, a digital twin with a neural Gaussian force field as its core is constructed. Specifically, a three-dimensional scene of the rock wall and its surrounding environment is reconstructed based on the images obtained by the RGB-D camera. A coordinate system and a geometric model of the tunneling machine consistent with the physical world are established in the twin space, and real-time pose data drives the virtual tunneling machine to move synchronously. Next, a neural Gaussian representation is introduced to continuously model the space, representing the force distribution of the rock-tool interaction as a learnable explicit force field, thereby achieving high-fidelity mapping and real-time updating of the force state in the cutting contact area.

[0019] Subsequently, to ensure that the digital twin not only possesses the ability to map geometry and stress but also reflects the influence of rock material properties on stress and vibration, a material property space for the rock strata is explicitly modeled within the digital twin. This material property space includes at least four types of parameters: density, Young's modulus, internal friction angle, and friction coefficient. These parameters are assigned and updated in the digital twin space using voxel meshes, continuous field functions, or layered region models. Density characterizes the rock mass's inertia and vibration propagation characteristics; Young's modulus characterizes the rock mass's elastic deformation and stiffness; and the internal friction angle and friction coefficient characterize the rock mass's failure and contact friction behavior. These material parameters can be initially assigned based on prior geological survey results and then corrected and updated using real-time acquired tool loads, vibration responses, and force field errors, ensuring mutual constraint between the material property space and the neural Gaussian force field. When there is a deviation between the predicted force field and the measured load, the material property parameters and force field network parameters are adjusted in reverse by minimizing the deviation, achieving adaptive updates that maintain consistency between geometry, material, and stress. The digital twin constructed in this way can synchronously present the status of the tunneling machine cluster, the spatial distribution of rock material, and the changes in the contact force field of the cutting tools under real-time driving, providing an executable data and model foundation for subsequent collaborative path planning, online hardness identification, and control parameter optimization.

[0020] For example, at a sampling time t = 12.580s, a multi-view RGB-D camera deployed at the tunneling face acquires color image frames (resolution 1920×1080 pixels) of the rock wall area and corresponding depth maps (depth range 0.5m to 8m, accuracy ±2mm). Simultaneously, pose data is output, such as the spatial position of tunneling machine No. 1 being x = 12.436m, y = -3.215m, z = 1.872m, with attitude quaternions of 0.923, 0.102, -0.356, and 0.105, corresponding to an angle of 8.7° between the advancing direction and the rock wall normal; and the spatial position of tunneling machine No. 2 being x = 14.012m, y = -2.864m, z = 1.855m, with an attitude yaw angle of 3.2°. This pose data is continuously output at a frequency of 100Hz to drive the synchronous movement of the virtual tunneling machines in the digital twin. At the same time, the fiber optic strain sensor acquired tool load data, including an axial load of 215 kN, a radial load of 168 kN, and an instantaneous torque of 32.5 kN·m. The load data for each distributed tool unit were: tool 1 normal force 18.6 kN, tool 2 normal force 21.3 kN, and tool 3 normal force 19.8 kN. This type of load data was sampled at a frequency of 1 kHz and stored in time series format; for example, 2000 sets of continuous force values ​​were recorded within a time window of 12.560 s to 12.580 s. Simultaneously, the triaxial accelerometer output rock vibration data, such as instantaneous accelerations of 3.25 m / s² in the X, Y, and Z directions. 2 -1.84m / s 2and 5.62m / s 2 Within the 0–500 Hz frequency band, the main vibration frequency was found to be 142 Hz through fast Fourier transform analysis, and the spectral energy was concentrated in the 120–180 Hz range.

[0021] Furthermore, the digital twin includes an object-centric explicit force field representation to describe the physical interaction mechanism between the rock strata and the cutting tool.

[0022] Preferably, the constructed digital twin includes an object-centric explicit force field representation for computable and updatable modeling of the physical interaction between the rock layer object and the tool object. Specifically, rock layer objects and tool objects are established separately in the digital twin space. The rock layer object is represented by a voxel mesh obtained from RGB-D reconstruction, with an attached material parameter field; the tool object is represented by a geometric model of structures such as the cutterhead and cutting teeth, and its position and attitude are updated in real time driven by the tunneling machine's pose sensing data. Object-centric means that the calculation and expression of the force field is not based on pure spatial sampling of global coordinates, but rather organized around the object's local coordinate system and contact area. That is, a local coordinate system is established for each tool object, defining its cutting direction, normal direction, and lateral direction, and an explicit force field unit is constructed in the local neighborhood covered by the tool sweep trajectory. This explicit force field unit can be a local voxel block or a continuous set of Gaussian elements. As the tool object moves within the digital twin, the system performs a neighborhood query on the rock strata object within its current position and the expected cutting area to obtain a set of contact candidate points. Based on material parameters and tool kinematics, it calculates contact state quantities, including normal indentation, relative slip velocity, and shear direction. Through an object-centric explicit force field representation, the digital twin can provide the local force distribution and interaction intensity between each tool object and the rock strata object at any given time. This can be directly used to predict future forces and analyze interference risks, and can also serve as executable input for subsequent impedance control and hydraulic pressure distribution, enabling feasible modeling of the physical interaction mechanism between the rock strata and the tool.

[0023] Furthermore, by acquiring multi-source sensor data from a tunneling machine cluster in real time using multi-source sensors, and constructing a digital twin based on a neural Gaussian force field, including:

[0024] The geometric texture images of the rock wall and the pose images of the tunnel boring machine (TBM) cluster are captured using a multi-view RGB-D camera in a multi-source sensor suite to obtain a set of rock wall geometric texture images and a set of TBM pose images. A fiber optic strain sensor in the multi-source sensor suite senses the cutterhead bearing to obtain cutter load data. A triaxial accelerometer in the multi-source sensor suite senses the root of the cutting teeth to obtain rock vibration data. The multi-source sensor data is then integrated with the rock wall geometric texture image set, the TBM pose image set, the cutter load data, and the rock vibration data to obtain the multi-source sensor data. This multi-source sensor data is then co-encoded, and combined with a neural Gaussian force field and an architecture based on neural operators, the explicit force field representation of the TBM cluster is determined, thus constructing the digital twin.

[0025] Preferably, multi-view RGB-D cameras are deployed at the tunneling face and key parts of the tunneling machine. The cameras undergo intrinsic and extrinsic parameter calibration. Intrinsic parameters include focal length, principal point, and distortion coefficient; extrinsic parameters include the rigid transformation matrix between the camera and the tunneling machine's coordinate system. A unified global coordinate system is then established with the initial cutterhead center position as the origin. Next, fiber optic strain sensors undergo zero-point calibration and temperature compensation calibration to establish a "wavelength change-strain-load" transformation function. Then, triaxial accelerometers undergo static and sensitivity calibration to establish a linear mapping relationship between voltage output and acceleration. All sensors are connected to the system via industrial Ethernet or CAN bus and synchronized at millisecond levels using a unified clock source. During data acquisition, the multi-view RGB-D cameras acquire RGB images and depth maps at a frequency of at least 30 frames per second. The pixel coordinates are converted into 3D point cloud data using the depth map back-projection formula. These point clouds are registered using the ICP algorithm to form a complete 3D point cloud model of the rock wall, preserving texture mapping information. The tunneling machine's pose is determined by fusion localization through image feature matching combined with IMU data, and state estimation is performed using extended Kalman filtering. The output is a pose data sequence containing position vectors and attitude quaternions.

[0026] Fiber optic strain sensors acquire wavelength variation data at a frequency of at least 1 kHz and demodulate it into strain values. The corresponding load is then calculated by multiplying the strain value by the material's elastic modulus and cross-sectional area, and further converted into cutterhead torque and equivalent cutting resistance. Triaxial accelerometers acquire vibration data at a frequency of at least 5 kHz. After bandpass filtering to remove low-frequency drift and high-frequency noise, short-time Fourier transform is performed to obtain frequency domain characteristic parameters, including dominant frequency, band energy, and spectral centroid, serving as dynamic characteristics of rock fragmentation and hardness changes. Subsequently, all data are aligned with a unified timestamp, and linear or spline interpolation is used to compensate for differences in sampling frequencies. The images and point clouds are scaled, and the load and vibration signals are standardized. Finally, each spatial sampling point in the 3D point cloud corresponding to the rock wall geometric texture image is correlated with the tunneling machine pose, cutter load, and rock vibration at the corresponding time, forming structured multi-source sensing data.

[0027] In the digital twin construction phase, the multi-source sensor data obtained above undergoes joint encoding processing. Geometric information is extracted using a point cloud encoding network, such as PointNet or a graph convolutional structure, to extract spatial features. Pose, load, and vibration data are extracted using a multilayer perceptron network to extract temporal features. These features are then fused through an attention mechanism to form a unified high-dimensional latent vector. Next, based on a neural Gaussian force field model, the workspace is divided into several learnable three-dimensional Gaussian primitives. Each Gaussian primitive contains a center, covariance, primitive intrinsic weights, and a force field vector. The neural network takes the unified high-dimensional latent vector and spatial coordinates as input and outputs the parameter update of the corresponding Gaussian primitive. The explicit force field value is calculated by spatially weighting and superimposing the force field vectors of multiple Gaussian primitives, representing the normal and tangential force distribution of the rock-tool interaction at that spatial point. The weights used for weighting are jointly determined by the primitive intrinsic weights and the spatial Gaussian kernel function. Simultaneously, a neural operator-based architecture is introduced. This architecture can be a Fourier neural operator or a graph neural operator, used to learn the operator mapping relationship from the input feature field to the output force field. This enables the model to predict the force response at any location in continuous space, without relying on a fixed grid resolution. During training or online updates, the actual force values ​​obtained from measured tool load and vibration inversion are used as supervision signals to construct the loss function L=||F pred- F real || 2 +λR, where F pred To predict the force field, F realThe network parameters and Gaussian element parameters are updated in real time through backpropagation to represent the actual force field obtained from actual measurement or inversion. λ is the regularization weight coefficient, and R is the regularization term. Finally, the real-time updated 3D geometric model, pose model, material property field, and explicit force field model are integrated into a unified digital twin. This twin can output the corresponding spatial force distribution and rock layer response results at any time upon input of the current pose and working condition data, achieving synchronous mapping and predictive calculation between the physical entity and the virtual model. This provides a computable physical basis for subsequent force prediction, path planning, and impedance control, improving the model's prediction accuracy and the real-time performance of control response.

[0028] The system dynamically analyzes the multi-machine operation status in the digital twin, extracts the propulsion speed feature set and the tool load feature set of the tunneling machine cluster, obtains the future force field distribution by combining force prediction, performs spatiotemporal conflict interference risk analysis, adjusts the path, and generates collaborative path planning instructions.

[0029] In one embodiment, within a constructed and real-time updated digital twin, dynamic operational status analysis is performed on a tunneling machine cluster. Specifically, the digital twin performs streaming computation on each tunneling machine in the cluster to obtain its corresponding propulsion speed characteristics; it analyzes the time-frequency domain characteristics of the tool load to obtain tool load characteristics; and then, by combining the obtained propulsion speed characteristics, tool load characteristics, and future force field distribution predicted based on explicit force field representation, it performs spatiotemporal conflict interference risk analysis on the real-time operational path planning of the tunneling machine cluster. This determines whether overlaps or insufficient safety distances occur within the same time window, thus outputting the interference risk analysis results for each tunneling machine. Finally, based on the obtained interference risk analysis results, real-time collaborative path planning adjustments are executed. Specifically, the local paths, speed profiles, and operational area allocations of each tunneling machine are jointly corrected to obtain a collaborative path planning instruction containing a target trajectory point sequence, allowable speed / acceleration upper limits, yield / avoidance timing, and operational area switching instructions. This instruction is then output to each tunneling machine for execution, thereby achieving dynamic collaboration and forward-looking planning optimization of the multi-machine operational status.

[0030] Furthermore, the multi-machine operation status is dynamically analyzed within the digital twin, extracting the propulsion speed feature set and tool load feature set of the tunneling machine cluster. Combined with force prediction, the future force field distribution is obtained, and spatiotemporal conflict interference risk analysis is performed to adjust the path and generate collaborative path planning instructions, including:

[0031] The tunneling machine cluster is traversed through the digital twin to perform streaming computation and obtain a set of propulsion speed characteristics; the time-frequency domain characteristics of the tool load in the digital twin are analyzed to obtain a set of tool load characteristics; force prediction is performed based on the explicit force field representation to obtain the future force field distribution; real-time operation path planning for the tunneling machine cluster is obtained; spatiotemporal conflict interference risk analysis is performed based on the set of propulsion speed characteristics, the set of tool load characteristics, and the future force field distribution to obtain the tunneling machine cluster interference risk analysis results; and real-time collaborative path planning adjustment is executed in conjunction with the tunneling machine cluster interference risk analysis results to obtain collaborative path planning instructions.

[0032] Optionally, after the digital twin is constructed and updated in real time, streaming computation is first performed on the tunneling machine cluster. Specifically, an independent state queue is established for each tunneling machine in the cluster. Its pose data and propulsion displacement data are continuously read according to time windows. Then, the instantaneous propulsion speed is obtained by differential calculation of the displacement change of the cutterhead center along the propulsion direction per unit time. Subsequently, statistical quantities such as speed mean, speed variance, acceleration, and speed fluctuation rate are calculated to form a propulsion speed feature vector. This vector is then traversed sequentially for all tunneling machines in the cluster to form a propulsion speed feature set, which is written to the state database of the digital twin in real time. Next, time-frequency domain analysis is performed on the tool load data synchronously stored in the digital twin. That is, the tool load time series is subjected to fast Fourier transform or continuous wavelet transform within a fixed time window to extract frequency domain features such as dominant frequency components, frequency band energy, spectral peak position, and harmonic amplitude. Time domain statistical features, including mean, peak value, root mean square value, kurtosis, and skewness, are also calculated. By concatenating the aforementioned time-domain and frequency-domain features to form a tool load feature vector, and then performing the same processing on each tunneling machine in the cluster, a tool load feature set is obtained. Subsequently, future force prediction is performed based on the explicit force field representation. Specifically, the tunneling machine pose images of each machine are acquired at the current moment, and the pose state is extracted. Then, based on the current explicit force field representation, the predicted force distribution is calculated. Finally, the force vectors of all sampling points within the contact area are integrated to obtain the future equivalent cutting resistance and torque distribution, thus forming the future force field distribution. Next, the current real-time operation path planning data of the tunneling machine cluster is acquired, including the motion trajectory point set of each machine at several future time steps. By projecting each trajectory in three dimensions onto a digital twin, a corresponding spatiotemporal occupancy volume is constructed, i.e., a motion envelope in the time-space dimension. Then, using the propulsion speed feature set, the tool load feature set, and the future force field distribution as constraints, continuous collision detection is performed between the tunneling machines and between the tunneling machines and the rock wall. This completes the spatiotemporal conflict interference risk analysis of the real-time operation path planning, obtaining the interference risk analysis results of the tunneling machine cluster. Then, real-time collaborative path planning adjustments are performed based on the interference risk analysis results.

[0033] Specifically, when the interference risk analysis results exceed a set threshold, a collaborative trajectory replanning method based on Model Predictive Control (MPC) is used to optimize the future trajectory of the tunneling machine cluster. At the current moment, the state vector of each tunneling machine is obtained. This state vector includes at least position coordinates, attitude angles, propulsion speed, and angular velocity. Simultaneously, a discrete time step sequence with a prediction time domain length of T is obtained. Then, the control quantity sequence of each tunneling machine within this prediction time domain is used as the optimization variable to construct the objective function. During the optimization process, the equipment dynamics model is introduced as an equality constraint condition; that is, the future trajectory is recursively predicted based on a simplified kinematic model. Inequality constraints are set, including the propulsion speed range, angular velocity range, and spatial boundary constraints, and a minimum safe distance constraint is introduced to avoid interference between equipment. Afterward, the above optimization problem is transformed into a constrained quadratic programming problem, which is solved iteratively in real time using a sequential quadratic programming (SQP) or interior-point method solver. Within each control cycle, only the first step of the optimized control quantity is executed, i.e., the current propulsion speed and orientation angle are updated, before entering the next control cycle for re-prediction and optimization, forming a rolling optimization mechanism. If the risk primarily stems from operational timing conflicts, a time delay variable is introduced within the same MPC framework to fine-tune the start-up time or advance rhythm of some tunneling machines. This allows some machines to enter the conflict zone a few time steps later, thus achieving timing decoupling. Finally, updated collaborative path planning instructions are output, including the target advance speed, target orientation angle, and necessary time delay parameters for each tunneling machine, and sent to the corresponding machine. Through this process, a complete closed-loop calculation process is achieved, from state feature extraction, future force prediction, interference risk assessment to path collaborative optimization, thereby improving the safety and efficiency of multi-machine collaborative operations.

[0034] Furthermore, based on the explicit force field representation, force prediction is performed to obtain the future force field distribution, including:

[0035] Obtain the set of tunnel boring machine pose images from the multi-source sensor data, and extract the set of tunnel boring machine pose state features; based on the explicit force field representation and the set of tunnel boring machine pose state features, predict the force field distribution to obtain the future force field distribution.

[0036] Optionally, a set of tunnel boring machine (TBM) pose images is first extracted from multi-source sensor data. These pose images are derived from real-time captures of the TBM cluster by multi-view RGB-D cameras. Using the same method described above, pose data matching is performed on the TBM pose image set, and feature extraction is performed on the pose data of the continuous time series to construct a set of TBM pose state features. Subsequently, based on the current pose state feature vector and a predetermined time step, the predicted pose for future moments is calculated using a motion model based on uniform velocity or model predictive control, thus obtaining the future predicted pose. Next, in the digital twin, the tool object is spatially projected according to the future predicted pose to determine the spatial neighborhood where the tool may contact the rock stratum. This neighborhood can be filtered by setting an envelope interval of a certain thickness around the tool sweep volume. For each spatial sampling point within the spatial neighborhood, its spatial coordinates, corresponding rock stratum material parameters, and pose state features are used as input variables and input into the neural Gaussian force field model. The model outputs a predicted force vector at a given point based on trained or online-updated parameters, representing the normal force and two orthogonal tangential force components. Then, the force vectors of all sampled points in the spatial neighborhood are weighted and superimposed using a Gaussian weighting function, and the area of ​​the contact surface is integrated to obtain the predicted total cutting force and predicted torque. Next, the set of force vectors from each sampled point in space and the integration results are combined to construct a three-dimensional force field distribution for future moments, including spatial location and corresponding force magnitude and direction information. Finally, the future force field distribution is output to a digital twin for subsequent interference risk analysis and collaborative path planning calculations. Through these steps, a complete computational process from pose image extraction and motion state modeling to explicit force field prediction is achieved, enabling the prediction of the tunnel boring machine's force state at the next moment, improving prediction accuracy and control foresight.

[0037] Furthermore, the real-time operation path planning of the tunneling machine cluster is obtained, and a spatiotemporal conflict interference risk analysis is performed based on the propulsion speed characteristic set, the tool load characteristic set, and the future force field distribution to obtain the interference risk analysis results of the tunneling machine cluster, including:

[0038] Based on the real-time operation path planning, the motion trajectory is projected into the digital twin to construct a spatiotemporal occupancy volume; using the propulsion speed feature set, the tool load feature set, and the future force field distribution as constraints, and combining the spatiotemporal occupancy volume, a continuous collision detection algorithm is used to perform spatiotemporal conflict interference risk analysis to obtain the tunneling machine cluster interference risk analysis results.

[0039] Optionally, firstly, the real-time operational path plan for each tunneling machine within the predicted future time interval is obtained. This real-time operational path plan is a discrete trajectory point sequence, including spatial position, attitude matrix, and corresponding timestamp. In the digital twin, the digital twin is invoked, and the 3D model of the tunneling machine in the digital twin is projected using rigid body transformation at each trajectory point to obtain the spatial occupancy geometry at that moment. Subsequently, the time dimension is discretized, and within each time slice, the corresponding spatial occupancy geometry is enveloped to generate voxelized spatial occupancy cells or convex hull representations, forming a time-related spatial occupancy set. Then, the spatial occupancy volumes of all time slices are superimposed along the time axis to construct a four-dimensional representation. By repeating the above steps for all tunneling machines in the cluster, a spatiotemporal occupancy volume set is obtained. Afterward, for any two tunneling machines, the minimum distance between their spatial occupancy volumes is calculated using Euclidean distance within the time interval. Continuous collision detection is performed using algorithms based on the Separating Axis Theorem (SAT) or Bounding Box Hierarchy (BVH) for fast solution. If a time t results in the minimum distance being less than 0, it is considered a physical collision. If the distance is less than a preset safety distance threshold, it is considered a potential interference risk. A distance risk factor is constructed by subtracting the preset safety distance threshold from the minimum distance and dividing by the preset safety distance threshold, then using the maximum value between the calculated result and 0. Next, for equipment pairs showing a tendency to approach each other, the relative velocity vector is calculated. If the relative velocity directions point towards each other and the relative velocity is large, the risk weight is increased, and the risk weight is multiplied by the ratio of the relative velocity to the maximum relative velocity to construct a velocity risk factor. If a tunneling machine currently exhibits a high RMS value or a peak value approaching the safety threshold in its load characteristics, its dynamic response capability is reduced. In this case, a load risk factor is constructed by multiplying the risk weight by the ratio of the load to the maximum load. Then, based on the predicted future force field distribution, the predicted equivalent cutting resistance is calculated within the potential proximity region. If the predicted equivalent cutting resistance exceeds the safety load threshold or a high stress concentration zone forms between the two machines, the risk weight is multiplied by the ratio of the predicted equivalent cutting resistance to the allowable force threshold to construct a force field risk factor. Then, for each pair of equipment, the interference risk analysis results of the tunneling machine cluster are obtained by weighting the distance risk factor, speed risk factor, load risk factor, and force field risk factor, and taking the maximum value or weighted average value for all pairs of equipment. Through the above process, a complete calculation flow from trajectory projection and spatiotemporal occupancy modeling to multi-constraint coupling risk assessment is realized, thereby enabling the early identification of potential interference and overload risks in multi-machine collaborative operations in the digital twin, providing a decision-making basis for subsequent collaborative path optimization.

[0040] An adaptive impedance control algorithm is used to adjust the hydraulic system pressure of each tunneling machine in the tunneling machine cluster to obtain impedance collaborative control parameters. The impedance collaborative control parameters are dynamically adjusted based on the online identification results of rock layer hardness output by the explicit force field representation in the digital twin.

[0041] In one embodiment, to ensure stable and efficient collaborative tunneling of a tunneling machine cluster under varying rock types and stress disturbances, an adaptive impedance control algorithm is employed to adjust the hydraulic system pressure of each tunneling machine in real time. Specifically, aiming to improve overall tunneling efficiency and balance cutter load, multi-objective optimization is performed on the online identification results of desired stiffness, desired damping, and explicit force field representation of rock hardness to obtain the impedance collaborative control parameters for each machine. For example, when a tunneling machine is in a high-hardness region or the predicted stress is too high, its damping can be increased or its expected propulsion amount can be decreased. Simultaneously, the expected propulsion amount of adjacent tunneling machines can be appropriately increased to achieve overall efficiency and load balance. Finally, the impedance collaborative control parameters are converted into executable hydraulic control quantities. Specifically, this can be achieved based on the relationship between the propulsion cylinder force and pressure. The controller calculates the target hydraulic cylinder force for each tunneling machine i. And obtain the pressure setpoint. Simultaneously, the proportional valve opening is calculated using a valve control model to form the final impedance-coordinated control parameters. In each control cycle, the system sends the pressure setpoint and proportional valve opening to each tunneling machine hydraulic station and proportional valve actuator. In the next cycle, it performs closed-loop correction based on newly acquired load and vibration data, thus achieving a real-time control closed loop of "online hardness identification - adaptive impedance parameter update - coordinated hydraulic pressure distribution." This ensures stable advancement even when rock hardness changes, while also balancing tunneling efficiency and cutter load.

[0042] Furthermore, by using an adaptive impedance control algorithm to adjust the hydraulic system pressure of each tunneling machine in the tunneling machine cluster, impedance coordination control parameters are obtained, including:

[0043] Based on the explicit force field representation and the rock stratum vibration data from the multi-source sensing data, the online identification result of rock stratum hardness is obtained; the desired stiffness and desired damping are determined based on the online identification result of rock stratum hardness; with the goal of improving the overall tunneling efficiency and tool load balance, multi-objective optimization is performed on the desired stiffness, desired damping, and the online identification result of rock stratum hardness to obtain impedance collaborative control parameters.

[0044] Preferably, during the collaborative operation of a tunneling machine cluster, the force field vector distribution within the contact neighborhood of the cutter and rock strata is first acquired based on the explicit force field representation of the digital twin at the current moment, and spatial sampling is performed on this neighborhood to establish a voxel sampling point set based on the cutter sweep volume. Subsequently, the force field gradient and its magnitude are calculated for each sampling point, and the component distribution of the gradient in the three principal directions is statistically analyzed. Furthermore, the anisotropy index of the force field is calculated to characterize the differences in the directionality of the force and the heterogeneity caused by joints or bedding. This anisotropy index can be obtained through the gradient tensor or structure tensor; a larger anisotropy index indicates a more concentrated force direction and a higher probability of well-developed rock joints or obvious bedding. Simultaneously, rock strata vibration data from multi-source sensor data are read, and the time-frequency features of the signals within the window are extracted after bandpass filtering, including root mean square, peak factor, dominant frequency, frequency band energy, and high-frequency energy ratio, to reflect the ease of rock fragmentation and impact intensity. Next, the explicit force field gradient statistics, anisotropy index and vibration characteristics are input into the online identification model. This online identification model can be a regression network or an incremental learning model, which is used to output a set of rock mechanical property parameters of the current cut area as the online identification result of rock hardness. This online identification result of rock hardness includes at least the uniaxial compressive strength of the rock, the degree of joint development and the friction angle.

[0045] After obtaining the online identification results of rock hardness, the system determines the desired stiffness and desired damping based on a preset impedance target mapping relationship. Specifically, for each tunneling machine, the system obtains the current average and fluctuation of the cutter load in the twin, and uses the rock hardness scalar and friction angle as inputs for the environmental equivalent stiffness or friction change to calculate the desired stiffness and desired damping. When the rock hardness scalar or the uniaxial compressive strength of the rock increases, the desired stiffness is increased to suppress pose errors and maintain cutting stability. When the root mean square of vibration or the proportion of high-frequency energy increases, the desired damping is increased to enhance energy dissipation and vibration suppression; when the anisotropy index or the degree of joint development increases, the desired stiffness is appropriately reduced and the desired damping is increased to reduce impact loads and the risk of cutter chipping. The above mapping can be implemented using piecewise linear or lookup table methods. Subsequently, with the goal of improving overall tunneling efficiency and tool load balance, multi-objective optimization solutions were performed on the desired stiffness, desired damping, and rock hardness identification results. By continuously optimizing and adjusting the initial impedance collaborative control parameters, the required impedance collaborative control parameters were obtained and sent to each tunneling machine controller for real-time adjustment of hydraulic system pressure and impedance characteristics. This enabled dynamic torque distribution and load balance control under different rock types and joint conditions, thus simultaneously balancing tunneling efficiency and tool load balance under conditions of varying rock hardness and structural heterogeneity.

[0046] For the online identification model, within each sliding time window, explicit force field gradient statistics, anisotropy index, and rock vibration characteristics are extracted from the digital twin as input variables. To enhance the model's ability to express nonlinear relationships, the above features are combined and expanded at the input end, including the squared terms of individual features, the cross terms between gradients and vibration characteristics, and the product terms of anisotropy index and high-frequency energy, thus forming an expanded feature vector containing the original features and their second-order combined terms. This expanded feature vector is then input into a multi-output regression structure, which maps a unified feature input to three physical quantity outputs: uniaxial compressive strength of rock, joint development index, and internal friction angle. In the initial stage, the model can be calibrated offline using existing geological survey data or historical tunneling section samples, and the initial regression coefficients are solved using ordinary least squares, giving the model reasonable initial estimation capabilities. During tunneling operations, the model is updated online using a recursive least squares algorithm. When the system acquires new observation data, it compares the current predicted output with the reference value, calculates the error, and corrects the regression coefficients according to the recursive least squares update rule. A forgetting factor is introduced into the model, giving more weight to recent data in parameter updates, allowing the model to gradually adapt to lithological changes while avoiding excessive influence from early data. The entire update process does not require saving all historical samples; it only performs recursive calculations based on the current input and parameters from the previous time step, meeting the requirements for real-time online operation. Finally, the model outputs a set of rock strata mechanical property parameters for the currently cut area, including uniaxial compressive strength, joint development degree, and internal friction angle, and performs physical rationality checks and smoothing processes to form a stable online identification result for rock strata hardness. This identification result is synchronously updated to the material property space in the digital twin, providing a reliable real-time input basis for subsequent adaptive setting of impedance parameters and collaborative optimization control.

[0047] Furthermore, with the goals of improving overall tunneling efficiency and tooling load balance, a multi-objective optimization solution is performed on the desired stiffness, desired damping, and online identification results of rock strata hardness to obtain impedance synergistic control parameters, including:

[0048] Based on the online identification results of rock layer hardness, lithology identification is performed to obtain lithological mechanical characteristics, including the uniaxial compressive strength, elastic modulus, and internal friction angle of the rock. Initial impedance coordination control parameters are determined based on the lithological mechanical characteristics, desired stiffness, and desired damping. With the goal of improving overall tunneling efficiency and tool load balance, the initial impedance coordination control parameters are optimized and adjusted to obtain the impedance coordination control parameters.

[0049] Optionally, after obtaining the online rock hardness identification result, the system uses this result to extract lithological parameters of the current cut area based on the force field gradient and anisotropy index. During this process, the system analyzes the force field gradient and its anisotropy index to assess the fracturing and bedding characteristics of the rock mass. Then, combining the high-frequency energy proportion, vibration peak factor, and dominant frequency in the vibration data, the system infers the uniaxial compressive strength of the rock layer. Finally, the system maps the UCS, AI, and vibration characteristics to lithological classification parameters to obtain a set of lithological mechanical characteristics, including the uniaxial compressive strength, elastic modulus, and internal friction angle of the rock. Subsequently, based on the lithological mechanical characteristics and desired stiffness and damping, the system matches the current lithological mechanical characteristics, desired stiffness, and desired damping with preset standard samples using cosine similarity. The system selects the set of standard samples with the highest similarity and defines the impedance co-control parameters of this set of standard samples as the initial impedance co-control parameters, including the initial pressure setpoint and the initial proportional valve opening. Next, at the beginning of each control cycle, the initial impedance coordinated control parameters at the end of the previous cycle are read, along with the latest lithological mechanical characteristics and the current operating status. Small-amplitude disturbances are then made around the initial impedance coordinated control parameters within the allowable range to form several candidate schemes. For example, the pressure setpoint and valve opening are finely adjusted in a stepwise manner within the allowable range of equipment response. When generating candidates, the basic constraints are met simultaneously, namely, the pressure does not exceed the hydraulic upper limit, the valve opening does not exceed the limit, and the parameter changes do not exceed the allowable rate of change of the system.

[0050] Subsequently, for each set of candidate parameters, the system performs rapid simulation and pre-running of the next prediction window in the digital twin. During the simulation, the explicit force field combines the propulsion state under the candidate parameters to predict the achievable propulsion speed trend, cutter load level and fluctuation degree, whether a certain machine's load is significantly higher, and whether it induces excessive vibration or overload risk, thereby generating corresponding expected performance results, including efficiency performance, load distribution status, and risk level. Then, based on the prediction results output by the twin, propulsion state data, cutter load distribution data, and vibration response data within the future prediction time window are extracted to construct corresponding performance evaluation indicators. In terms of efficiency evaluation, the overall propulsion capability of the tunneling machine cluster within the prediction time window is the core indicator, comprehensively considering the average propulsion speed of each tunneling machine, propulsion continuity, and the number of active speed reductions caused by predicted overload or impact. If the prediction result corresponding to a certain candidate parameter shows a higher average propulsion speed, smaller speed fluctuation, and fewer propulsion interruptions, its efficiency evaluation level is judged to be better. In terms of load balancing evaluation, the evaluation is based on the load distribution of each tunneling machine's cutters within the predicted time window. The focus is on analyzing the degree of load difference between machines, the peak load level of a single machine, and the stability of the load time series fluctuations. If the prediction result for a candidate parameter shows a small load difference between machines, no significant increase in the maximum load of a single machine, and a smooth load change trend, then its balancing performance is considered superior. Simultaneously, a safety constraint screening mechanism is introduced before comprehensive ranking. That is, the hydraulic pressure value, peak cutter load, vibration amplitude, and spatiotemporal interference risk index predicted in the twin simulation for each candidate parameter are compared item by item. If any index exceeds a preset safety threshold or shows a significant increasing trend in interference risk, the candidate parameter is directly determined as an infeasible solution and removed from the candidate set. For the remaining solutions, the system performs comprehensive ranking based on the principle of efficiency priority while also considering balance. For example, it prioritizes candidate parameters with the smallest load difference under the premise of significantly improved efficiency; if the efficiency improvement is similar, it selects candidate parameters with more balanced load and lower vibration.

[0051] Finally, the system sends the top-ranked candidate parameter as the impedance co-control parameter for this cycle, executing it for only one control cycle or one short window. When the next control cycle arrives, the latest sensor data and twin update results are reread, candidates are generated again, twin calculations are performed again, and the selection is repeated, thus achieving continuous rolling optimization. This allows for rapid selection of more suitable impedance parameters and pressure distribution schemes in the next cycle even if rock hardness, joint development, or friction conditions suddenly change, avoiding efficiency degradation or excessive wear on machine tools caused by long-term use of fixed parameters.

[0052] Furthermore, the impedance coordination control parameters include the pressure setpoints and proportional valve openings of each tunneling machine's hydraulic system.

[0053] Optionally, impedance coordination control parameters include the pressure setpoints of each tunneling machine's hydraulic system and the proportional valve opening control quantity. The pressure setpoint refers to the target pressure parameter of the main pump or distribution circuit in the hydraulic system, used to determine the thrust or torque level that the hydraulic cylinder or hydraulic motor can output per unit time. By adjusting the pressure setpoint, the output force of the cutterhead or propulsion mechanism can be directly changed, thereby affecting the tunneling machine's cutting capability and propulsion intensity. When the rock hardness is high or the predicted stress is large, the pressure setpoint can be appropriately increased to ensure sufficient cutting capability; when the predicted load tends to be balanced or there is an overload risk, the pressure setpoint should be appropriately decreased to avoid impact and energy waste. The proportional valve opening refers to the control input quantity of the hydraulic proportional control valve, used to adjust the flow rate or pressure response rate of the hydraulic oil. The size of the proportional valve opening directly affects the movement speed, response sensitivity, and dynamic characteristics of the system of the hydraulic actuator. By finely adjusting the proportional valve opening, dynamic control of the propulsion speed change rate, torque build-up process, and damping characteristics can be achieved. For example, when vibrations are significant or the rock strata are uneven, the damping effect can be enhanced by reducing the amplitude of the proportional valve opening change, making the system response smoother. When it is necessary to improve propulsion efficiency, the opening can be appropriately increased to accelerate the flow response. By jointly controlling the pressure and valve opening, the hydraulic output force, response speed, and damping characteristics can be comprehensively adjusted, thereby improving the overall tunneling efficiency, suppressing vibration impact, and achieving balanced tool load.

[0054] The tunneling machine cluster is optimized for collaborative operation based on the impedance collaborative control parameters and the collaborative path planning instructions.

[0055] In one embodiment, after obtaining the impedance coordination control parameters, the system writes the pressure setpoints corresponding to each tunneling machine into the controller of the hydraulic main pump or distribution valve, causing each actuator cylinder or hydraulic motor to operate within the target pressure range, thereby forming the expected output thrust and torque levels. Simultaneously, the proportional valve opening control quantity is sent to the corresponding proportional valve drive unit to adjust the hydraulic flow and response speed, ensuring that the propulsion action and cutterhead rotation process have matched dynamic impedance characteristics. During this process, the control parameters for each tunneling machine are not uniformly fixed, but rather differentiated based on optimization results, allowing different tunneling machines to bear different levels of cutting load within the same working area, achieving reasonable load distribution. In actual execution, the system continuously collects real-time propulsion speed, cutter load, and vibration response data for each tunneling machine and compares them with the prediction results from the digital twin. If an excessively high load or abnormal vibration is detected in a particular device, its pressure setpoint or proportional valve opening is fine-tuned for timely compensation. Simultaneously, combined with collaborative path planning instructions, the system coordinates the propulsion sequence or local propulsion rhythm, enabling multiple machines to cooperate in an orderly manner in both spatial and temporal dimensions, avoiding mutual interference or local overload. Through the above-mentioned execution and feedback adjustment process, the comprehensive optimization of hydraulic output force distribution, propulsion rhythm control and spatial coordination is achieved, enabling the tunneling machine cluster to maintain a stable propulsion state under different rock strata conditions, improving overall tunneling efficiency, reducing the risk of single machine overload, and achieving dynamic balance of cutter load.

[0056] In summary, the embodiments of this application have at least the following technical effects:

[0057] First, multi-source sensor data from a tunneling machine cluster is collected in real time using multiple sensors, and a digital twin is constructed based on a neural Gaussian force field. Next, the multi-machine operating status is dynamically analyzed within the digital twin, extracting the propulsion speed characteristic set and cutter load characteristic set of the tunneling machine cluster. Combined with force prediction, the future force field distribution is obtained, and a spatiotemporal conflict interference risk analysis is performed to adjust the path and generate collaborative path planning instructions. Then, an adaptive impedance control algorithm is used to adjust the hydraulic system pressure of each tunneling machine in the cluster, obtaining impedance collaborative control parameters. These parameters are dynamically adjusted based on the online rock hardness identification results output from the explicit force field representation in the digital twin. Finally, the collaborative operation of the tunneling machine cluster is optimized based on the impedance collaborative control parameters and the collaborative path planning instructions. This invention addresses the technical problems of existing tunneling machine clusters lacking real-time perception and prediction capabilities of rock strata physical properties, leading to lagging path planning, unreasonable hydraulic torque distribution, and uneven tool wear. It achieves online identification of rock strata hardness and prediction of future stress based on digital twins and neural Gaussian force fields, improving the accuracy of multi-machine collaborative path planning and the adaptive capability of impedance control, thereby increasing tunneling efficiency, enhancing operational safety, and balancing tool wear.

[0058] Example 2, based on the same inventive concept as the real-time optimization method for collaborative tunneling machine operation based on digital twins in the previous examples, such as... Figure 2 As shown, this application provides a real-time optimization system for collaborative operation of tunneling machines based on digital twins, wherein the system includes:

[0059] The digital twin construction module 11 collects multi-source sensor data of the tunneling machine cluster in real time through multi-source sensors and constructs a digital twin based on a neural Gaussian force field. The path planning module 12 dynamically analyzes the multi-machine operation status in the digital twin, extracts the propulsion speed feature set and the tool load feature set of the tunneling machine cluster, obtains the future force field distribution by combining force prediction, performs spatiotemporal conflict interference risk analysis, adjusts the path, and generates collaborative path planning instructions. The impedance control module 13 uses an adaptive impedance control algorithm to adjust the hydraulic system pressure of each tunneling machine in the tunneling machine cluster to obtain impedance collaborative control parameters. The impedance collaborative control parameters are dynamically adjusted according to the online identification results of rock hardness output by the explicit force field representation in the digital twin. The operation optimization module 14 optimizes the collaborative operation of the tunneling machine cluster according to the impedance collaborative control parameters and the collaborative path planning instructions.

[0060] Furthermore, the twin construction module 11 is used to perform the following methods:

[0061] The digital twin includes an object-centric explicit force field representation used to describe the physical interaction mechanism between the rock strata and the cutting tool.

[0062] Furthermore, the twin construction module 11 is used to perform the following methods:

[0063] The geometric texture images of the rock wall and the pose images of the tunnel boring machine (TBM) cluster are captured using a multi-view RGB-D camera in a multi-source sensor suite to obtain a set of rock wall geometric texture images and a set of TBM pose images. A fiber optic strain sensor in the multi-source sensor suite senses the cutterhead bearing to obtain cutter load data. A triaxial accelerometer in the multi-source sensor suite senses the root of the cutting teeth to obtain rock vibration data. The multi-source sensor data is then integrated with the rock wall geometric texture image set, the TBM pose image set, the cutter load data, and the rock vibration data to obtain the multi-source sensor data. This multi-source sensor data is then co-encoded, and combined with a neural Gaussian force field and an architecture based on neural operators, the explicit force field representation of the TBM cluster is determined, thus constructing the digital twin.

[0064] Furthermore, the path planning module 12 is used to perform the following methods:

[0065] The tunneling machine cluster is traversed through the digital twin to perform streaming computation and obtain a set of propulsion speed characteristics; the time-frequency domain characteristics of the tool load in the digital twin are analyzed to obtain a set of tool load characteristics; force prediction is performed based on the explicit force field representation to obtain the future force field distribution; real-time operation path planning for the tunneling machine cluster is obtained; spatiotemporal conflict interference risk analysis is performed based on the set of propulsion speed characteristics, the set of tool load characteristics, and the future force field distribution to obtain the tunneling machine cluster interference risk analysis results; and real-time collaborative path planning adjustment is executed in conjunction with the tunneling machine cluster interference risk analysis results to obtain collaborative path planning instructions.

[0066] Furthermore, the path planning module 12 is used to perform the following methods:

[0067] Obtain the set of tunnel boring machine pose images from the multi-source sensor data, and extract the set of tunnel boring machine pose state features; based on the explicit force field representation and the set of tunnel boring machine pose state features, predict the force field distribution to obtain the future force field distribution.

[0068] Furthermore, the path planning module 12 is used to perform the following methods:

[0069] Based on the real-time operation path planning, the motion trajectory is projected into the digital twin to construct a spatiotemporal occupancy volume; using the propulsion speed feature set, the tool load feature set, and the future force field distribution as constraints, and combining the spatiotemporal occupancy volume, a continuous collision detection algorithm is used to perform spatiotemporal conflict interference risk analysis to obtain the tunneling machine cluster interference risk analysis results.

[0070] Furthermore, the job optimization module 14 is used to perform the following methods:

[0071] Based on the explicit force field representation and the rock stratum vibration data from the multi-source sensing data, the online identification result of rock stratum hardness is obtained; the desired stiffness and desired damping are determined based on the online identification result of rock stratum hardness; with the goal of improving the overall tunneling efficiency and tool load balance, multi-objective optimization is performed on the desired stiffness, desired damping, and the online identification result of rock stratum hardness to obtain impedance collaborative control parameters.

[0072] Furthermore, the job optimization module 14 is used to perform the following methods:

[0073] Based on the online identification results of rock layer hardness, lithology identification is performed to obtain lithological mechanical characteristics, including the uniaxial compressive strength, elastic modulus, and internal friction angle of the rock. Initial impedance coordination control parameters are determined based on the lithological mechanical characteristics, desired stiffness, and desired damping. With the goal of improving overall tunneling efficiency and tool load balance, the initial impedance coordination control parameters are optimized and adjusted to obtain the impedance coordination control parameters.

[0074] Furthermore, the job optimization module 14 is used to perform the following methods:

[0075] The impedance coordination control parameters include the pressure setpoints and proportional valve openings of each tunneling machine's hydraulic system.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A real-time optimization method for collaborative operation of tunneling machines based on digital twins, characterized in that, The method includes: Multi-source sensor data of the tunneling machine cluster are collected in real time by multi-source sensors, and a digital twin is constructed based on the neural Gaussian force field; The operation status of multiple machines is dynamically analyzed in the digital twin, the propulsion speed feature set and the tool load feature set of the tunneling machine cluster are extracted, the future force field distribution is obtained by combining the force prediction, and the spatiotemporal conflict interference risk analysis is performed to adjust the path and generate collaborative path planning instructions. An adaptive impedance control algorithm is used to adjust the hydraulic system pressure of each tunneling machine in the tunneling machine cluster to obtain impedance collaborative control parameters. The impedance collaborative control parameters are dynamically adjusted based on the online identification results of rock layer hardness output by the explicit force field representation in the digital twin. The tunneling machine cluster is optimized for collaborative operation based on the impedance collaborative control parameters and the collaborative path planning instructions.

2. The real-time optimization method for collaborative operation of tunneling machines based on digital twins as described in claim 1, characterized in that, The digital twin includes an object-centric explicit force field representation used to describe the physical interaction mechanism between the rock strata and the cutting tool.

3. The real-time optimization method for collaborative operation of tunneling machines based on digital twins as described in claim 1, characterized in that, Multi-source sensor data from a tunneling machine cluster is collected in real time using multiple sensors, and a digital twin is constructed based on a neural Gaussian force field, including: By using a multi-view RGB-D camera from a multi-source sensor to capture geometric texture images of the rock wall and pose images of the tunnel boring machine cluster, a set of geometric texture images of the rock wall and a set of pose images of the tunnel boring machine are obtained. The tool bearing is sensed using a fiber optic strain sensor among multi-source sensors to obtain tool load data; The root of the cutting tooth is sensed by a triaxial accelerometer among the multi-source sensors to obtain rock vibration data; By integrating the rock wall geometric texture image set, the tunnel boring machine pose image set, the cutter load data, and the rock stratum vibration data, the multi-source sensor data is obtained. The multi-source sensor data is co-encoded, and combined with the neural Gaussian force field and the architecture built based on neural operators, the explicit force field representation of the tunneling machine cluster is determined, and the digital twin is constructed.

4. The real-time optimization method for collaborative operation of tunneling machines based on digital twins as described in claim 3, characterized in that, The system dynamically analyzes the multi-machine operation status within the digital twin, extracts the propulsion speed characteristic set and tool load characteristic set of the tunneling machine cluster, combines force prediction to obtain the future force field distribution, performs spatiotemporal conflict interference risk analysis, adjusts the path, and generates collaborative path planning instructions, including: The tunneling machine cluster is traversed through the digital twin to perform streaming computation and obtain a set of propulsion speed features; Analyze the time-frequency domain characteristics of the tool load in the digital twin to obtain a set of tool load characteristics; Force prediction is performed based on the explicit force field representation to obtain the future force field distribution; The real-time operation path planning of the tunneling machine cluster is obtained, and a spatiotemporal conflict interference risk analysis is performed based on the propulsion speed feature set, the tool load feature set, and the future force field distribution to obtain the interference risk analysis results of the tunneling machine cluster. Based on the results of the interference risk analysis of the tunneling machine cluster, real-time collaborative path planning adjustments are performed to obtain collaborative path planning instructions.

5. The real-time optimization method for collaborative operation of tunneling machines based on digital twins as described in claim 4, characterized in that, Based on the explicit force field representation, force prediction is performed to obtain the future force field distribution, including: Obtain the set of tunnel boring machine pose images from the multi-source sensor data, and extract the set of tunnel boring machine pose state features; Based on the explicit force field representation and the set of features of the tunnel boring machine's pose state, the force field distribution is predicted to obtain the future force field distribution.

6. The real-time optimization method for collaborative operation of tunneling machines based on digital twins as described in claim 4, characterized in that, The real-time operation path planning of the tunneling machine cluster is obtained. Based on the propulsion speed characteristic set, the tool load characteristic set, and the future force field distribution, a spatiotemporal conflict interference risk analysis is performed to obtain the interference risk analysis results of the tunneling machine cluster, including: Based on the real-time operation path planning, the motion trajectory is projected into the digital twin to construct a spatiotemporal occupancy volume; Using the propulsion speed feature set, the tool load feature set, and the future force field distribution as constraints, and combining the spatiotemporal occupancy volume, a continuous collision detection algorithm is used to perform spatiotemporal conflict interference risk analysis, and the results of the tunneling machine cluster interference risk analysis are obtained.

7. The real-time optimization method for collaborative operation of tunneling machines based on digital twins as described in claim 1, characterized in that, An adaptive impedance control algorithm is used to adjust the hydraulic system pressure of each tunneling machine in the tunneling machine cluster to obtain impedance coordination control parameters, including: Based on the explicit force field representation and the rock stratum vibration data in the multi-source sensing data, the online identification result of rock stratum hardness is obtained; Determine the desired stiffness and desired damping based on the online rock layer hardness identification results; With the goal of improving overall tunneling efficiency and tool load balance, multi-objective optimization is performed on the desired stiffness, desired damping, and the online identification results of rock layer hardness to obtain impedance collaborative control parameters.

8. The real-time optimization method for collaborative operation of tunneling machines based on digital twins as described in claim 7, characterized in that, With the goals of improving overall tunneling efficiency and tooling load balance, a multi-objective optimization solution is performed on the desired stiffness, desired damping, and online rock hardness identification results to obtain impedance synergistic control parameters, including: Based on the online identification results of rock layer hardness, lithology identification is performed to obtain lithological mechanical characteristics, wherein the lithological mechanical characteristics include the uniaxial compressive strength, elastic modulus and internal friction angle of the rock; Based on the aforementioned lithological mechanical characteristics, desired stiffness, and desired damping, determine the initial impedance coordinated control parameters; With the goal of improving overall tunneling efficiency and tool load balance, the initial impedance coordination control parameters are optimized and adjusted to obtain the impedance coordination control parameters.

9. The real-time optimization method for collaborative operation of tunneling machines based on digital twins as described in claim 8, characterized in that, The impedance coordination control parameters include the pressure setpoints and proportional valve openings of each tunneling machine's hydraulic system.

10. A real-time optimization system for collaborative operation of tunneling machines based on digital twins, characterized in that, For implementing the real-time optimization method for collaborative tunneling machine operation based on digital twins as described in any one of claims 1-9, the system comprises: Digital twin construction module: Real-time acquisition of multi-source sensor data from the tunneling machine cluster through multi-source sensors, and construction of digital twins based on neural Gaussian force fields; Path planning module: Dynamically analyzes the multi-machine operation status in the digital twin, extracts the propulsion speed feature set and tool load feature set of the tunneling machine cluster, obtains the future force field distribution by combining force prediction, performs spatiotemporal conflict interference risk analysis, adjusts the path, and generates collaborative path planning instructions; Impedance control module: The adaptive impedance control algorithm is used to adjust the hydraulic system pressure of each tunneling machine in the tunneling machine cluster to obtain impedance collaborative control parameters. The impedance collaborative control parameters are dynamically adjusted according to the online identification results of rock layer hardness output by the explicit force field representation in the digital twin. Operation optimization module: Optimizes the collaborative operation of the tunneling machine cluster based on the impedance collaborative control parameters and the collaborative path planning instructions.

Images