Tunnel lining trolley adaptive path planning method and system based on digital twinning

Abstract

The invention relates to the technical field of tunnel construction, and discloses a tunnel lining trolley adaptive path planning method and system based on digital twinning. According to the method, a virtual mapping model of a tunnel structure is established according to real-time environment data collected by various sensors on a lining trolley; analyzing passable space characteristics of the tunnel lining trolley in the virtual mapping model; obtaining current pose information, preset path planning and dynamic operation parameters of the tunnel lining trolley, and calculating a space compatibility index of a path and a tunnel environment; when the spatial compatibility index deviates from the security standard, activating a path re-planning mechanism, and generating a plurality of alternative path sets; for each alternative path set, measuring collaborative operation overhead caused by adjustment of a tunnel lining trolley control unit; fusing the collaborative operation overhead and geometric attributes of the alternative path set, and screening an optimal alternative path; and updating the path planning in the digital twin system according to the optimal alternative path.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel construction, in particular to a tunnel lining trolley self-adaptive path planning method and system based on digital twinning. BACKGROUND

[0002] The current tunnel lining trolley path planning mainly adopts a control mode based on a preset trajectory and a fixed safety interval. The existing technology has limited perception ability for the tunnel environment, and relies on offline measurement data or a simplified geometric model, which is difficult to adapt to the dynamic changes in the construction process. There is a lack of real-time quantitative evaluation mechanism for the matching degree of the actual running state of the trolley and the tunnel environment in the path planning process. When there are environmental changes or path deviations, the re-planning strategy often only considers the geometric feasibility, ignoring the execution cost and coordination complexity of the control system. The existing method needs to solve the problems of inaccurate environment perception, poor path adaptability and low control efficiency.

[0003] Traditional path planning methods have obvious shortcomings in environmental modeling accuracy and dynamic adaptability. The application of digital twinning technology is mostly limited to visual display, and has not been deeply integrated into the path planning decision-making link. The use of environmental data is relatively single, and the characteristic information of the passable space has not been fully tapped. The path evaluation standard is too simple, mainly considering geometric constraints, and lacks comprehensive consideration of control execution cost. When generating alternative solutions in the re-planning process, there is no systematic evaluation of the influence of each solution on the coordinated operation of the trolley's multiple control units, which may result in a theoretically feasible planning path but low execution efficiency. The information transmission between the monitoring platform and the planning system lags behind, making it difficult to adjust the effect in a timely manner. SUMMARY

[0004] The purpose of the present application is to provide a tunnel lining trolley self-adaptive path planning method and system based on digital twinning to solve the problems raised in the background.

[0005] To achieve the above purpose, the present application provides a tunnel lining trolley self-adaptive path planning method and system based on digital twinning, which comprises:

[0006] Integrating real-time environmental data collected by various sensors on the tunnel lining trolley through a digital twinning system, a virtual mapping model of the tunnel structure is established; In the virtual mapping model, the passable space characteristics of the tunnel lining trolley are analyzed; Combining the passable space characteristics, the current pose information of the tunnel lining trolley, the preset path planning and the dynamic running parameters are taken to calculate the spatial compatibility index of the path and the tunnel environment; When the spatial compatibility index deviates from the safety standard, activate the path re-planning mechanism, and generate a plurality of alternative path sets based on the passable space characteristics; For each set of alternative paths, measure the collaborative operation overhead caused by the adjustment of the tunnel lining trolley control unit; Fusing the collaborative operation overhead and the geometric properties of the set of alternative paths, screening the optimal alternative path; According to the optimal alternative path, update the path planning in the digital twin system, and generate a path adjustment record to transmit to the monitoring platform.

[0007] Preferably, the plurality of sensors includes a laser radar, a visual sensor, and an inertial measurement unit; The real-time environmental data includes three-dimensional point cloud data of the tunnel wall surface, image texture information, and motion acceleration data; When establishing a virtual mapping model of the tunnel structure, the three-dimensional point cloud data is registered and fused, the image texture information is combined to enhance the model details, and the motion acceleration data is used to correct the dynamic error of the model.

[0008] Preferably, when analyzing the passable space features of the tunnel lining trolley, the contour boundary, surface concave-convex degree, and obstacle occupation area of the tunnel inner wall are extracted from the virtual mapping model; The current pose information of the tunnel lining trolley includes global coordinates and attitude angles, and the dynamic running parameters include travel speed, hydraulic outrigger state, and steering angle; When calculating the spatial compatibility index, the minimum gap distance between the tunnel lining trolley shape projection and the tunnel inner wall contour boundary is compared.

[0009] Preferably, when activating the path re-planning mechanism, based on the obstacle occupation area in the passable space features, the key nodes that the tunnel lining trolley needs to avoid are located; Divide the safety channel on the preset path planning, and the safety channel is composed of conflict-free path segments between consecutive avoidance key nodes; a path search algorithm is used to generate a set of alternative paths for each conflict-free path segment.

[0010] Preferably, when measuring the collaborative operation overhead caused by the adjustment of the tunnel lining trolley control unit, the dynamic running parameters and the current pose information of the tunnel lining trolley are obtained; Simulate the control instruction sequence and execution time required for the tunnel lining trolley to switch from the current running state to the preset state, the control instruction sequence is called from the historical operation library, and the execution time is obtained through digital twin system simulation.

[0011] Preferably, when fusing the collaborative operation overhead and the geometric properties of the set of alternative paths, the geometric properties are determined by calculating the curvature mean and length variance of each path segment in the set of alternative paths; When screening the optimal alternative path, the set of alternative paths with low collaborative operation overhead and smooth geometric properties is preferentially selected.

[0012] Preferably, after selecting the optimal alternative path, it is verified whether the hydraulic outrigger status and steering angle of the tunnel lining trolley need to be adjusted to adapt to the optimal alternative path. When updating the path planning in the digital twin system, the optimal alternative path is written into the path database, overwriting the original path data.

[0013] Preferably, when generating the path adjustment record, the record content includes the path change timestamp, detailed parameters of the optimal alternative path, and historical values ​​of the spatial compatibility index; Path adjustment records are uploaded to the monitoring platform in real time, and the monitoring platform indexes and stores the records.

[0014] Preferably, the method further includes periodically calibrating the consistency between the virtual mapping model and the real tunnel environment; The calibration process involves comparing real-time data from multiple sensors with predictions from a virtual mapping model to adjust the model's resolution and dynamic parameters.

[0015] Preferably, the present invention also includes a digital twin-based adaptive path planning system for tunnel lining trolleys, comprising a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the aforementioned digital twin-based adaptive path planning method for tunnel lining trolleys.

[0016] Compared with the prior art, the beneficial effects of the present invention are: In a virtual model, the passable space characteristics of the trolley are analyzed to accurately identify the boundaries and morphological features of the safe passage area. Combining the trolley's real-time pose, preset path, and operating parameters, a dynamic compatibility index between the path and the environment is calculated. This index quantitatively evaluates the safety and feasibility of the current path in a real-time environment, providing data for path adjustment. When the compatibility index exceeds the safe range, multiple candidate path schemes are generated based on the passable space characteristics. Each candidate path meets basic geometric passage requirements, providing a basis for subsequent optimization selection.

[0017] For each candidate path, the collaborative operation overhead required for each control unit of the trolley to execute that path is quantitatively evaluated. Operational overhead includes parameters such as the motion coordination complexity of each actuator, energy consumption cost, and timing requirements. A multi-objective optimization algorithm is used to integrate the collaborative operation overhead with the geometric properties of the path. Geometric properties include features such as path length, curvature, and smoothness, while operational overhead reflects the execution difficulty of the path. A comprehensive evaluation function is established to screen out alternative paths that achieve the optimal balance between geometric performance and operational efficiency. The optimal path ensures traffic safety while minimizing the execution burden on the control system.

[0018] The path planning in the digital twin system is updated based on the optimal path, achieving real-time synchronization between the virtual and physical spaces. Detailed path adjustment records are generated and transmitted to the monitoring platform, including information such as the reasons for adjustments, scheme comparisons, and expected effects. The monitoring platform can view the path adjustment status and execution effects in real time, forming a complete decision-making closed loop. This adaptive planning method based on digital twins significantly improves the accuracy of path planning and the efficiency of control execution. Through dynamic compatibility assessment and multi-objective optimization screening, it ensures that the path scheme meets both safety requirements and high execution efficiency. Real-time data-driven and closed-loop optimization mechanisms enable the system to quickly adapt to changes in the construction environment, improving construction quality and efficiency. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the working principle of the adaptive path planning method and system for tunnel lining trolley based on digital twins as described in this invention. Figure 2 A flowchart for establishing a virtual mapping model and processing sensor data; Figure 3 A flowchart for analyzing the characteristics of passable space and calculating spatial compatibility indicators; Figure 4 A comparison diagram of the curvature changes of the tunnel lining trolley along different paths; Figure 5 The time series analysis diagram of the calibration error of the tunnel lining trolley and the model resolution is shown. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1This invention provides a method and system for adaptive path planning of a tunnel lining trolley based on digital twins. The method includes: establishing a virtual mapping model of the tunnel structure based on real-time environmental data collected by multiple sensors on the tunnel lining trolley using a digital twin system. This model serves as a digital copy of the real tunnel and can accurately reflect the tunnel's geometry and dynamic changes. In the virtual mapping model, the traversable space characteristics of the tunnel lining trolley are analyzed, including the tunnel's inner wall contour and obstacle distribution. These features are extracted and quantified into spatial parameters using algorithms. Combining the traversable space characteristics, the current pose information of the tunnel lining trolley, such as global coordinates and attitude angles, preset path planning, such as a pre-set travel route, and dynamic operating parameters, such as travel speed and hydraulic outrigger status, are obtained. A spatial compatibility index between the path and the tunnel environment is calculated. This index quantifies the safety level by comparing the minimum clearance distance between the trolley's shape and the tunnel boundary. When the spatial compatibility index deviates from the safety standard, it indicates a collision or interference risk on the current path. The system then activates a path replanning mechanism, generating multiple sets of alternative paths based on the traversable space characteristics. Each set contains multiple candidate path segments. For each set of alternative paths, the coordination overhead caused by adjustments to the tunnel lining trolley control unit is measured. This overhead includes the switching cost and execution time of the control command sequence. The coordination overhead is combined with the geometric properties of the alternative path set, such as path curvature and length variance, to screen for the optimal alternative path, prioritizing paths with low overhead and geometric smoothness. The path planning in the digital twin system is updated based on the optimal alternative path, the new path is written to the database, and path adjustment records, including timestamps and parameter changes, are generated and transmitted to the monitoring platform for real-time tracking and auditing.

[0022] Example 1: See Figure 2This paper presents an adaptive path planning method for tunnel lining trolleys based on digital twins, integrating multiple sensors and establishing a virtual mapping model. These sensors include LiDAR, vision sensors, and inertial measurement units (IMUs). These sensors are fixedly installed at key locations on the tunnel lining trolley, such as the roof and sidewalls. The LiDAR emits pulsed laser light and receives reflected signals to collect 3D point cloud data of the tunnel wall. The vision sensors use high-resolution cameras to capture image texture information. The IMUs incorporate accelerometers and gyroscopes to record motion acceleration data. In practical implementation, real-time environmental data includes 3D point cloud data of the tunnel wall, image texture information, and motion acceleration data. The 3D point cloud data stores the 3D coordinates and reflection intensity of each point in the form of a point set. The image texture information includes an RGB color matrix, and the motion acceleration data covers linear acceleration and angular velocity values. This data is transmitted in real-time to the data acquisition module of the digital twin system via wired or wireless networks. When establishing a virtual mapping model of the tunnel structure, the 3D point cloud data is registered and fused. The registration operation uses an iterative nearest-point algorithm to align multiple frames of point clouds to a unified coordinate system. The fusion process reduces data volume while maintaining shape integrity through voxel meshing. Image texture information is combined to enhance model details, and texture mapping technology is used to attach image pixels to the 3D mesh surface. Motion acceleration data is used to correct dynamic errors in the model, and displacement changes are derived through integral acceleration and synchronized with point cloud timestamps to compensate for motion drift. The digital twin system runs on an embedded computing platform, and model data is stored in a distributed database, supporting real-time updates and queries.

[0023] In practical implementation, the LiDAR scanning frequency is set to 10Hz to ensure real-time data transmission, the exposure time of the vision sensor is adaptively adjusted to adapt to changes in illumination within the tunnel, and the sampling rate of the inertial measurement unit is set to 100Hz to capture high-frequency motion. Registration and fusion of the 3D point cloud data involve preprocessing steps, such as using statistical filters to remove outliers. After registration, the point cloud is used to generate a triangular mesh model through a surface reconstruction algorithm. When enhancing model details with image texture information, distortion correction and color balancing are performed on the image to ensure accurate alignment of texture and geometry. Correction of motion acceleration data uses Kalman filtering to smooth noise and compares it with the point cloud registration results to calculate the error offset. The virtual mapping model is built in sync with the sensor data stream, typically updated every 100 milliseconds. The model resolution is dynamically adjusted according to the tunnel complexity, increasing point cloud density in high-detail areas. In some embodiments, the LiDAR employs a rotating design to cover a 360-degree field of view, the vision sensor is equipped with infrared illumination for low-light environments, and the inertial measurement unit is calibrated to an initial stationary state. The data stream is strictly synchronized via timestamps to avoid model distortion caused by timing misalignment.

[0024] Optionally, machine learning algorithms can be introduced into the establishment of the virtual mapping model to optimize registration accuracy, such as using feature matching to improve point cloud alignment speed. The digital twin system periodically verifies model consistency by adjusting parameters by comparing real-time sensor data with model predictions. In specific implementation, the registration of 3D point cloud data adopts a multi-scale strategy, first coarse registration and then fine registration to reduce computational load, while preserving the original features of the point cloud during fusion to avoid over-smoothing. The image texture enhancement process includes texture coordinate generation and mapping weight calculation to ensure natural transitions in details. Motion acceleration data correction integrates the trolley kinematics model to improve the accuracy of dynamic error compensation. It is understandable that the quality of the virtual mapping model directly depends on the accuracy of the sensor data; therefore, regular sensor maintenance and calibration are crucial. The model output format is a standard 3D file such as OBJ or PLY, facilitating parsing by the subsequent path planning module.

[0025] In some embodiments, LiDAR point cloud data is transmitted via network protocols, visual sensor images are compressed and encoded to reduce bandwidth usage, and inertial measurement unit data is directly read via serial port. When establishing the virtual mapping model, the registration algorithm can use point feature histogram descriptors to improve robustness, and the fusion operation uses an octree structure to manage point cloud spatial partitions. Image texture enhancement employs a multi-resolution texture pyramid to adapt to the detail requirements of different viewing distances. Motion acceleration data correction introduces a temperature compensation coefficient to offset sensor drift. The digital twin system provides a model editing interface, allowing manual adjustment of geometric imperfections. Optionally, model data can be backed up to cloud storage for off-site disaster recovery. The entire implementation process is automated, requiring no manual intervention, ensuring that the virtual mapping model always reflects the real tunnel condition.

[0026] In practical implementation, the LiDAR point cloud data contains millions of points. The digital twin system uses GPU acceleration for registration and fusion calculations, while visual sensor images are decoded using the OpenCV library. Inertial measurement unit (IMU) data is filtered and stored in a cache. When building the virtual mapping model, the registration error threshold is set at the millimeter level. After fusion, the model undergoes hole filling and smoothing. Image texture enhancement combined with photometric stereo technology estimates surface normals, improving the realism of details. Motion acceleration data correction and trolley control feedback loops dynamically adjust the model's attitude. It's understandable that building the virtual mapping model is fundamental to path planning, requiring high reliability and low latency. The digital twin system records the model's version history, supporting rollback to any point in time.

[0027] In practical implementation, the sensor data acquisition module features anti-interference design, such as shielding electromagnetic noise to ensure data purity. When establishing the virtual mapping model, an open-source algorithm for registering and fusing 3D point cloud data is used, combined with image texture information to enhance model details and employs shader programming for real-time rendering. When correcting dynamic errors in the model using motion acceleration data, a polynomial fitting offset for the error model is established. The digital twin system monitors the deviation between the model and the real environment, triggering remodeling when limits are exceeded. In some embodiments, reflectivity information is added to LiDAR point cloud data to assist material identification, feature extraction from visual sensor images is used for model alignment, and inertial measurement unit data is fused with GPS signals to improve positioning accuracy. Optionally, the virtual mapping model can be exported in a lightweight format for viewing on mobile devices.

[0028] Example 2: See Figure 3 In specific implementation, when analyzing the passable space characteristics of the tunnel lining trolley, the contour boundary, surface unevenness, and obstacle-occupied area of ​​the tunnel inner wall are extracted from the virtual mapping model. The contour boundary of the tunnel inner wall is generated by processing the three-dimensional point cloud data using the alpha-shape algorithm. This algorithm constructs polygonal boundaries by defining radius parameters to accurately describe the shape of the tunnel cross section. The surface unevenness is quantified by calculating the normal vector of each point in the point cloud and statistically analyzing the local curvature changes. The obstacle-occupied area is identified using a density-based spatial clustering algorithm, marking connected regions in the point cloud with a density higher than a threshold as obstacles. The current pose information of the tunnel lining trolley includes global coordinates and attitude angles. Global coordinates are calculated by fusing GPS signals and UWBS data. Attitude angles are converted to Euler angles using quaternion data provided by the inertial measurement unit. Dynamic operating parameters include travel speed, hydraulic outrigger status, and steering angle. Travel speed is calculated using the encoder pulse frequency of the tunnel lining trolley's drive motor. Hydraulic outrigger status is determined by reading values ​​from hydraulic cylinder displacement sensors to indicate whether they are in a retracted, extended, or floating state. Steering angle is directly measured using the angle sensor of the steering mechanism. When calculating spatial compatibility indicators, the minimum gap distance between the tunnel lining trolley's external projection and the tunnel's inner wall contour boundary is compared. The external projection is a simplified polygon obtained by projecting the tunnel lining trolley's 3D model along the current attitude angle onto a 2D horizontal plane. The tunnel's inner wall contour boundary is a polygon extracted from slices of a virtual mapping model.

[0029] In practical implementation, the contour boundary extraction process involves equal-height cross-section processing of the virtual mapping model. Edge detection operators are applied at each cross-section height to obtain continuous boundary lines. Surface convexity is calculated using principal component analysis to estimate the local surface normal direction of the point cloud and calculate the normal variance as an convexity index. Clustering of obstacle-occupied areas uses a density-based clustering method with noise, distinguishing obstacles from noise points by setting neighborhood radii and minimum point thresholds. Current pose information is acquired using an extended Kalman filter algorithm that fuses multi-source sensor data. While the Global Positioning System (GPS) provides absolute position reference, its signal is easily blocked within the tunnel. An ultra-wideband positioning system (UWBS) deploys base stations inside the tunnel to provide higher-precision relative positioning. Inertial measurement unit (IMU) data is used to compensate for short-term drift in the positioning system. Dynamic operating parameters are periodically read from each subsystem of the tunnel lining trolley via the controller area network (CLAN). Travel speed values ​​are low-pass filtered to eliminate bounce caused by tire slippage. Hydraulic outrigger status is encoded as discrete enumerated values ​​such as "fully retracted," "working," and "fully extended." Steering angles are converted into deflection angles relative to the longitudinal axis of the tunnel lining trolley. The calculation cycle of the spatial compatibility index is consistent with the sensor data update frequency, typically once every 100 milliseconds. The calculated minimum gap distance is compared in real time with a preset safety threshold (e.g., 200 mm). When the calculated value is lower than the safety threshold for several consecutive cycles, a path replanning process is triggered. It can be understood that the accuracy of the passable space feature resolution directly depends on the quality of the virtual mapping model, while the accuracy of the current pose information determines the reliability of the spatial compatibility index calculation.

[0030] In some embodiments, contour boundary extraction can incorporate active contour modeling methods to better adapt boundary lines to the irregular shapes of tunnel walls. Surface roughness quantification can combine gray-level co-occurrence matrix statistical texture features to enhance the description of surface characteristics. Obstacle-occupied areas can be identified by fusing semantic segmentation results from visual sensors to distinguish obstacle types. The calculation of current pose information can supplement inertial measurement unit integration with wheeled odometer data to reduce position drift when satellite signals are lost. The acquisition of dynamic operating parameters can incorporate vibration sensor data to evaluate the smoothness of tunnel lining trolley operation. The calculation of spatial compatibility indices can consider the dynamic envelope of the tunnel lining trolley, converting factors such as travel speed and steering angle into a safety margin threshold for dynamically adjusting the minimum clearance distance. Optionally, the spatial compatibility index can be extended to a multi-dimensional vector, including not only the minimum clearance distance but also the minimum distance between key components of the tunnel lining trolley and the tunnel roof.

[0031] When calculating the spatial compatibility index, the dynamic dimensional changes of the tunnel lining trolley need to be considered in the external projection. For example, the overall outline of the tunnel lining trolley will change when the hydraulic outriggers are in different extension states. Therefore, before generating the external projection, the geometric parameters of the 3D model of the tunnel lining trolley need to be adjusted according to the real-time reading of the hydraulic outrigger status to generate accurate projected polygons for the corresponding states. The tunnel inner wall outline boundary also needs to be dynamically sliced ​​according to the current height position of the tunnel lining trolley to ensure that the comparison is performed at the correct spatial level. The calculation of the minimum clearance distance is implemented using an efficient collision detection library to meet real-time requirements. It can be understood that the spatial compatibility index is the core quantitative basis for path planning safety, and its computational efficiency and accuracy have a decisive impact on the performance of the tunnel lining trolley adaptive path planning system. In some embodiments, the calculation of the spatial compatibility index can be parallelized, distributing the comparison task of the tunnel lining trolley external projection and the tunnel inner wall outline boundary to multiple computing cores simultaneously to improve the system response speed. Optionally, in order to cope with the centrifugal effect generated when the tunnel lining trolley travels on curves, a dynamic compensation factor can be introduced into the calculation of the minimum clearance distance, and the safety margin can be appropriately increased according to the travel speed and the curvature of the curve.

[0032] Example 3: In specific implementation, the path replanning mechanism is activated based on the obstacle-occupied area in the passable space to locate the key nodes that the tunnel lining trolley needs to avoid. The obstacle-occupied area is extracted from the virtual mapping model using a 3D point cloud clustering algorithm. Key nodes are defined as the centroid or most convex point of the obstacle's geometric region. These nodes constitute the constraint point set in the path planning. The key nodes are located using a spatial grid indexing method, discretizing the tunnel environment into cubic voxel units. Each voxel unit is marked as an obstacle or free state according to the point cloud density. The coordinates of the key nodes are selected from the set of center points of the obstacle voxel units and redundant points are removed by distance filtering. A safety passage is divided on the preset path plan. The safety passage consists of conflict-free path segments between consecutively avoiding key nodes. Conflict-free path segments must satisfy the condition that the passage width is greater than the sum of the tunnel lining trolley's external width and the dynamic safety margin. The safety passage division is based on topological map construction, using key nodes as vertices and connecting edges through visibility detection to form a passage network. Conflict-free path segments are extracted from the passage network using a graph search algorithm. A path search algorithm is used to generate an alternative path set for each conflict-free path segment. The path search algorithm selected is A... The algorithm, or fast random tree algorithm, takes as input the coordinates of the start and end points of the current path segment, a binary map of the passable space features, and the kinematic constraints of the tunnel lining trolley. The output is multiple parametric path curves. Generating the set of alternative paths requires optimizing the geometric properties and safety of the paths, for example, by reducing curvature extrema through path smoothing algorithms.

[0033] In practical implementation, the processing of obstacle-occupied areas uses a clustering algorithm based on Euclidean distance. Point cloud data is first downsampled using voxels to reduce computation, then adjacent point cloud clusters are merged using a region growing algorithm. The circumcenter of each point cloud cluster serves as the initial candidate location for key nodes. Key node localization incorporates a motion prediction module to estimate the future position of dynamic obstacles and generate a dynamic key node sequence. The safe passage partitioning process employs a conservative expansion strategy, expanding the obstacle-occupied area along the normal direction by a certain distance to form a safe boundary. Conflict-free path segments must be completely outside the safe boundary. When the path search algorithm generates an alternative path set for each conflict-free path segment, A... The algorithm's heuristic function is designed as the Euclidean distance from the current point to the destination. The fast random tree algorithm constructs a path tree through random sampling and nearest neighbor expansion. Each path in the alternative path set is fitted with a B-spline curve to ensure second-order continuity and differentiability. It can be understood that the real-time performance of the path replanning mechanism depends on the optimal allocation of computational resources. Key node localization and safety passage partitioning can be processed in parallel to improve efficiency. In some embodiments, key node localization can fuse multi-frame sensor data and improve position accuracy through Kalman filtering; safety passage partitioning introduces a time dimension to establish a spatiotemporal corridor to handle moving obstacles; and the path search algorithm combines a machine learning model to predict the optimal path type.

[0034] In practical implementation, the activation conditions of the path replanning mechanism are monitored by the decision module of the digital twin system. The replanning process is triggered when the spatial compatibility index falls below a threshold for several consecutive periods. The key node localization algorithm runs on a dedicated computing unit, using an octree data structure to accelerate spatial queries. The localization results include the node's 3D coordinates, confidence level, and timestamp information. Safe passage segmentation is based on a visibility graph constructed from key nodes. Connectivity between nodes is detected through ray casting, and conflict-free path segment generation uses a weighted shortest path algorithm to optimize passage quality. After the path search algorithm generates a set of alternative paths, each path needs to be comprehensively evaluated to support subsequent selection. The evaluation function needs to comprehensively consider multiple objective factors such as path length, smoothness, and safety. For example, the path evaluation function can be designed in the following dimensionless form:

[0035] in: Indicates the path evaluation score. , , This represents the weight coefficient of each item. This indicates the actual length of the current candidate path. A baseline value representing the path length. This represents the absolute value of the maximum curvature along the path. This indicates the maximum permissible curvature limit of the tunnel lining trolley. This indicates the safe distance threshold set by the system. This represents the minimum distance from the path to the nearest obstacle.

[0036] It is understandable that the effectiveness of the path replanning mechanism depends on the accuracy of environmental modeling; errors in key node positioning or deviations in safety passage delineation may lead to path planning failure. In specific implementations, the update frequency of obstacle-occupied areas is synchronized with the LiDAR scanning frame rate. Key node positioning results are handled for transient occlusion using a multi-hypothesis tracking algorithm. After safety passage delineation, collision detection verification is required to ensure passage effectiveness. Parameter configurations of the path search algorithm, such as step size and sampling interval, affect path quality and need to be calibrated based on the control accuracy of the tunnel lining trolley. In some embodiments, key node positioning can be combined with a deep learning segmentation model to improve the ability to identify irregularly shaped obstacles. Safety passage delineation uses adaptive mesh refinement to improve resolution in complex areas. The path search algorithm integrates a trajectory optimization module to directly generate feasible control commands. After the alternative path set is generated, it is stored in a shared memory pool for subsequent use by the collaborative operation overhead measurement module.

[0037] Example 4: In specific implementation, when measuring the collaborative operation overhead caused by the adjustment of the tunnel lining trolley control unit, it is necessary to obtain the dynamic operating parameters and current pose information of the tunnel lining trolley. The dynamic operating parameters include travel speed, hydraulic outrigger status and steering angle. These parameters are read in real time through the vehicle bus network. The current pose information includes global coordinates and attitude angles, which are calculated by fusing GPS / IMU data. The core of overhead measurement is the sequence of control commands and execution time required for the tunnel lining trolley to switch from the current operating state to the preset state. The control command sequence is called from the historical operation library, which is a relational database that stores past successful control records. The record fields include timestamp, control command code, state parameters before and after execution, and actual time consumption. The simulation process constructs a high-fidelity dynamic model of the tunnel lining trolley in the digital twin system. This model includes mass, inertia tensor, joint constraints, and hydraulic system response characteristics. The entire process of tracking the target path from the current state to the steady state is simulated by numerical integration method. The execution time is recorded by the simulation clock. The cooperative operation overhead is finally quantified into a comprehensive scalar value, which takes into account factors such as command switching frequency, energy consumption, and time delay.

[0038] In practical implementation, the invocation of control command sequences adopts a scene-matching-based retrieval strategy. The dynamic operating parameters and current pose information of the current tunnel lining trolley are compared with records in the historical operation database to calculate similarity. The most similar historical scenes are matched, and their control command sequences are extracted. These command sequences contain a series of control quantities at discrete time points, such as throttle opening, braking pressure, hydraulic cylinder displacement, and steering angle. The simulation of execution time uses the Runge-Kutta method to solve the multibody dynamics equations of the tunnel lining trolley. These equations consider tire-ground friction, hydraulic actuator delays, and structural flexibility. The calculation model for cooperative operation overhead can be expressed as a weighted sum of various cost functions, such as time cost, energy cost, and control effort cost. When integrating the geometric properties of the cooperative operation cost and the alternative path set, the geometric properties are determined by calculating the mean curvature and length variance of each path segment in the alternative path set. The mean curvature reflects the overall curvature of the path and is obtained by calculating the curvature of the path point sequence and then taking the arithmetic mean. The length variance describes the dispersion of the length of each candidate path in the path set and is obtained by calculating the variance after statistically analyzing the total length of each path. When selecting the optimal alternative path, the non-geometric indicator of cooperative operation cost needs to be normalized along with geometric indicators such as the mean curvature and length variance to transform them into dimensionless score values. Then, multi-objective decision-making methods such as weighted scoring or Pareto optimal front analysis are used to prioritize the alternative path set with low cooperative operation cost and smooth geometric properties. "Low" and "smooth" are specifically manifested as small cooperative operation cost score values, small mean curvature, and small length variance.

[0039] It is understandable that the accuracy of measuring collaborative operation overhead is highly dependent on the completeness of the historical operation database and the accuracy of the dynamic model. The historical operation database needs to be continuously updated to cover more working conditions, and the parameters of the dynamic model need to be calibrated through physical trolley testing. The calculation of geometric properties needs to ensure the consistency of path parameterization to avoid errors introduced by different sampling point intervals. Selecting the optimal alternative path is a multi-objective optimization problem, and the setting of weight coefficients needs to be adjusted according to the specific tasks of the tunnel lining trolley and the characteristics of the tunnel environment. For example, in emergency repair and rescue tasks, more emphasis may be placed on low collaborative operation overhead to shorten time, while in normal lining operations, more emphasis may be placed on smooth geometric properties to ensure pouring quality. Refer to Table 1, which shows a simplified path selection evaluation table to illustrate how to quantitatively evaluate different candidate paths.

[0040] Table 1: Comprehensive Evaluation Table of Alternative Path Sets

[0041] In practical implementation, the collaborative operation costs in the table have been normalized, while the mean curvature and length variance are the original calculated values. The comprehensive score is calculated through weighted combination (for example, assuming the weights are: cost 0.5, curvature 0.3, length variance 0.2, then the comprehensive score of P1 = 0.5 × 0.85 + 0.3 × 0.15 + 0.2 × 2.1 = 0.72; this is just an example, and the actual weights need to be set based on engineering experience). In actual digital twin system software implementations, such evaluation tables are dynamically generated and updated in memory as the basis for path decision-making. Optionally, the evaluation process can introduce fuzzy logic to handle the uncertainty of the indicators, or use the analytic hierarchy process (AHP) to determine the weights of each indicator. The screening results are not absolutely unique; the system can provide top-K optimal paths for the operator to make the final choice.

[0042] In some embodiments, the stationarity of the control instruction sequence is considered when measuring the cost of cooperative operation, and abrupt changes in instructions are penalized, for example, by adding a cost term for the rate of change of the control quantity to the cost calculation model. The simulation execution time takes into account the communication delay from the sensor to the controller and the controller's computation cycle, making the simulation closer to the response characteristics of the real system. When calculating geometric properties, the mean curvature can be calculated using a weighted average method, assigning higher weights to points with greater curvature in the path to better reflect the impact of sharp bends. The length variance can be normalized based on the standard deviation of the path set length to eliminate the influence of magnitude. Optionally, the optimal path selection module can be designed as a configurable plug-in, supporting flexible switching between different multi-objective decision-making algorithms, such as ant colony optimization or simulated annealing, to adapt to different optimization needs. The digital twin system records all intermediate data and final decision results in each path selection process, forming a decision log for subsequent analysis and algorithm optimization. The entire implementation process embodies a model-driven decision-making approach. By using high-fidelity simulation, the control and adjustment costs that are difficult to measure directly are quantified and combined with intuitive geometric properties. This allows for the systematic evaluation and selection of the optimal driving path in a virtual space, providing core support for the safe, efficient, and autonomous operation of the tunnel lining trolley.

[0043] See Figure 4 This illustrates how the curvature of five paths, P1 to P5, changes with path distance (in meters). Curvature, a key indicator of path geometry, reflects the degree of path bending, and is measured in meters. -¹. As can be seen from the figure, the curvature of different paths exhibits significant dynamic changes within the path distance range of 0 to 100 meters. The curvature fluctuation characteristics of P1 (blue curve), P2 (orange curve), P3 (green curve), P4 (red curve), and P5 (purple curve) are different, reflecting the diversity of geometric shapes of different paths. This curvature variation data is an important basis for evaluating the geometric properties of paths. Combined with indicators such as collaborative operation overhead, it can be used to screen the optimal path for tunnel lining trolleys, providing quantitative support for their safe and efficient operation in tunnel environments.

[0044] Example 5: In specific implementation, after selecting the optimal alternative path, it is verified whether the tunnel lining trolley needs to adjust the hydraulic outrigger status and steering angle to adapt to the optimal alternative path. The verification process is carried out through the multibody dynamics model of the tunnel lining trolley in the digital twin system. This model contains detailed mechanical structural parameters and kinematic constraints. During verification, the optimal alternative path is discretized into a series of dense path points. The theoretical extension of the hydraulic outriggers is calculated sequentially when the tunnel lining trolley travels along each path point. It is determined whether the theoretical extension is within the physical stroke limit of the hydraulic cylinder. At the same time, the required steering angle is calculated to ensure it is within the maximum steering angle range of the steering servo. For example, when the optimal alternative path includes a sharp bend, the verification module simulates the trajectory of the tunnel lining trolley's center of gravity, calculates the steering angle requirements of the inner and outer tires, and compares them with the maximum design steering angle of the steering system. For the hydraulic outrigger status, if the path passes through a raised area, it is necessary to verify whether the maximum extension length of the outriggers is sufficient to maintain the vehicle's level. The verification process is iterative. If the hydraulic outrigger status or steering angle at any path point fails to meet the requirements, the path will be marked as infeasible. The system will then backtrack and select the alternative path with the second-best overall score for re-verification.

[0045] When updating the path planning in the digital twin system, the optimal alternative path is written to the path database, overwriting the original path data. The path database uses a time-series database structure, and each path record contains a globally unique path identifier, version number, creation timestamp, and path point sequence. The path point sequence stores the 3D coordinates, attitude angles, and corresponding preset speed information of each point in array form. The overwrite operation uses atomic transactions to ensure data consistency. Before writing new path data, the system automatically archives the original path data to a historical version table for rollback when needed. The path planning module of the digital twin system monitors changes to the path database in real time. Once a new optimal alternative path is detected, the path data is immediately loaded and converted into a command sequence that the tunnel lining trolley control unit can recognize. The command sequence is then sent to the underlying controller via the vehicle bus. It is understood that path planning updates must be immediate and reliable; any delay or data error may cause the tunnel lining trolley to execute outdated or unsafe paths.

[0046] When generating route adjustment records, the record content includes a route change timestamp, detailed parameters of the optimal alternative route, and historical values ​​of spatial compatibility indicators. The route change timestamp is accurate to milliseconds and uses Coordinated Universal Time (UTC). Detailed parameters of the optimal alternative route include the route identifier, total length, average curvature, a list of key point coordinates, and deviation statistics from the original route. Historical values ​​of spatial compatibility indicators are stored as a time-series array, recording sampled values ​​of spatial compatibility indicators for a period before the route replanning is triggered, illustrating the necessity of replanning. Route adjustment records are uploaded to the monitoring platform in real time. Upon receiving the records, the monitoring platform first parses and verifies them, then builds an index based on information such as the route change timestamp and tunnel lining trolley number. The index structure typically uses a B+ tree to support efficient range and time-point queries. The storage layer writes the records to a distributed file system or time-series database and backs them up to an off-site disaster recovery center to ensure data security and traceability. The monitoring platform also provides a graphical interface for operators to query, filter, and visualize historical route adjustment records.

[0047] The consistency between the virtual mapping model and the real tunnel environment is periodically calibrated. The calibration process involves comparing real-time data from multiple sensors with the predicted values ​​of the virtual mapping model, adjusting the model's resolution and dynamic parameters. The calibration cycle is configurable, for example, it can be set to be executed automatically every 5 minutes, or triggered immediately when a significant deviation between sensor data and model predictions is detected. The calibration process first selects a series of feature points in the virtual mapping model, such as specific cracks or bolt locations on the tunnel wall, and then identifies the true 3D coordinates of these feature points in the LiDAR point cloud and visual sensor images at the same time. Next, the deviation vector set between the predicted coordinates of the feature points in the virtual mapping model and the true coordinates measured by the sensors is calculated, and the overall translation, rotation, and scaling parameter errors of the model are fitted using algorithms such as least squares. The global pose of the virtual mapping model is adjusted based on these errors. For resolution, if the model is found to be unable to distinguish small obstacles in areas with rich detail, the sampling rate and mesh density of the point cloud in that area are dynamically increased. Dynamic parameters, such as sensor noise model parameters or coefficients of the motion prediction model, are also adaptively updated based on long-term average deviations. Optionally, the calibration process can incorporate a confidence score. When the errors of multiple consecutive calibrations are all below a threshold, the time interval for the next calibration can be appropriately extended to save computational resources. In some embodiments, the calibration process can be remotely triggered by a monitoring platform. When an operator observes a significant discrepancy between the virtual model and the real-time video stream, they can manually initiate a forced calibration command.

[0048] In practical implementation, the generation of path adjustment records is a crucial basis for auditing and fault analysis. The record files are serialized using a JSON structured format and transmitted to the monitoring platform via an encrypted channel. When the monitoring platform indexes and stores the records, it simultaneously updates a cache of the latest records in memory to support real-time updates of the monitoring interface. Regularly calibrating the virtual mapping model is key to maintaining the high fidelity of the digital twin system. The calibration algorithm typically runs in a low-priority background task to avoid interfering with real-time path planning. Optionally, for critical tunnel sections, additional positioning base stations or markers can be deployed to provide a more accurate reference benchmark for calibration. Optionally, the calibration results themselves also generate logs, recording the calibration time, the amount of calibration parameter adjustments, and the improvement in model accuracy after calibration. Another optional approach is that when calibration detects a large deviation that cannot be compensated for by parameter adjustments, the system can issue an alarm, indicating that a large-scale model reconstruction or sensor check may be necessary. In some embodiments, the path verification process can be linked to the calibration process. If a path fails frequently during verification, it may indicate a deviation in the underlying virtual mapping model, thereby triggering a targeted local area calibration.

[0049] See Figure 5 This paper demonstrates the trends of calibration error and model resolution over time during the calibration process of a tunnel lining trolley system based on digital twins, and their relationship with the error threshold. In the figure, the blue line represents the calibration error (unit: m), reflecting the degree of deviation between the virtual mapping model and the real tunnel environment; the orange line represents the model resolution (unit: m / point), reflecting the model's ability to depict the details of the tunnel environment; the green dashed line represents the error threshold (0.10 m), a key indicator for determining whether the model needs calibration. From the time series (09:00 to 09:45), the calibration error peaks at 09:10 (approximately 0.15 m), exceeding the error threshold, indicating a significant deviation between the virtual model and the real environment at this point, requiring attention to the calibration effectiveness. Subsequently, the calibration error gradually decreases, dropping to approximately 0.05 m by 09:45, below the threshold, indicating that the calibration operation effectively improved the model's fidelity. The model resolution showed a gradual upward trend, increasing from approximately 0.05m / point initially to approximately 0.02m / point after 09:40. This means that the model's ability to distinguish details of the tunnel environment is continuously improving. This is consistent with the mechanism of periodically calibrating the virtual mapping model to adjust the resolution in the project. It reflects the process of optimizing the model accuracy through calibration of the digital twin system, providing high-fidelity virtual environment support for the adaptive path planning of the tunnel lining trolley.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive path planning method for tunnel lining trolleys based on digital twins, characterized in that, The method includes: By integrating real-time environmental data collected by multiple sensors on the tunnel lining trolley through a digital twin system, a virtual mapping model of the tunnel structure is established. In the virtual mapping model, the passable space characteristics of the tunnel lining trolley are analyzed; Based on the characteristics of the passable space, the current position information of the tunnel lining trolley, the preset path planning and dynamic operating parameters are taken to calculate the spatial compatibility index between the path and the tunnel environment. When the spatial compatibility index deviates from the safety standard, the path replanning mechanism is activated to generate a set of multiple alternative paths based on the characteristics of the passable space. For each set of alternative paths, measure the collaborative operation overhead caused by adjustments to the tunnel lining trolley control unit; By integrating the overhead of cooperative operations with the geometric properties of the set of alternative paths, the optimal alternative path is selected; The path planning in the digital twin system is updated based on the optimal alternative path, and path adjustment records are generated and transmitted to the monitoring platform.

2. The adaptive path planning method for tunnel lining trolley based on digital twin as described in claim 1, characterized in that, The various sensors include lidar, vision sensors, and inertial measurement units; The real-time environmental data includes three-dimensional point cloud data of the tunnel wall, image texture information, and motion acceleration data; When establishing a virtual mapping model of the tunnel structure, the three-dimensional point cloud data is registered and fused, the model details are enhanced by combining image texture information, and the dynamic error of the model is corrected by using motion acceleration data.

3. The adaptive path planning method for tunnel lining trolley based on digital twin as described in claim 2, characterized in that, When analyzing the passable space characteristics of the tunnel lining trolley, the contour boundary, surface unevenness, and obstacle-occupied area of ​​the tunnel inner wall are extracted from the virtual mapping model. The current position and pose information of the tunnel lining trolley includes global coordinates and attitude angles, and the dynamic operating parameters include travel speed, hydraulic outrigger status and steering angle. When calculating the spatial compatibility index, the minimum gap distance between the projection of the tunnel lining trolley and the contour boundary of the tunnel inner wall is compared.

4. The adaptive path planning method for tunnel lining trolley based on digital twin as described in claim 3, characterized in that, When the path replanning mechanism is activated, the key nodes that the tunnel lining trolley needs to avoid are located based on the obstacle-occupied areas in the passable space characteristics. A safety passage is defined in the preset path planning. The safety passage consists of a non-conflicting path segment that continuously avoids key nodes. A path search algorithm is used to generate a set of alternative paths for each conflict-free path segment.

5. The adaptive path planning method for tunnel lining trolley based on digital twin as described in claim 4, characterized in that, When measuring the collaborative operation overhead caused by the adjustment of the tunnel lining trolley control unit, the dynamic operating parameters and current pose information of the tunnel lining trolley are obtained. The simulation program simulates the control command sequence and execution time required for the tunnel lining trolley to switch from its current operating state to a preset state. The control command sequence is retrieved from the historical operation library, and the execution time is obtained through simulation using a digital twin system.

6. The adaptive path planning method for tunnel lining trolley based on digital twin as described in claim 5, characterized in that, When integrating the collaborative operation overhead with the geometric properties of the alternative path set, the geometric properties are determined by calculating the mean curvature and length variance of each path segment in the alternative path set. When selecting the optimal alternative path, priority should be given to the set of alternative paths with low cooperative operation overhead and smooth geometric properties.

7. The adaptive path planning method for tunnel lining trolley based on digital twin as described in claim 6, characterized in that, After selecting the optimal alternative path, it is verified whether the hydraulic outrigger status and steering angle of the tunnel lining trolley need to be adjusted to adapt to the optimal alternative path. When updating the path planning in the digital twin system, the optimal alternative path is written into the path database, overwriting the original path data.

8. The adaptive path planning method for tunnel lining trolley based on digital twin as described in claim 7, characterized in that, When generating the path adjustment record, the record content includes the path change timestamp, detailed parameters of the optimal alternative path, and historical values ​​of spatial compatibility indicators. Path adjustment records are uploaded to the monitoring platform in real time, and the monitoring platform indexes and stores the records.

9. The adaptive path planning method for tunnel lining trolley based on digital twin as described in claim 1, characterized in that, The method also includes periodically calibrating the consistency between the virtual mapping model and the real tunnel environment; The calibration process involves comparing real-time data from multiple sensors with predictions from a virtual mapping model to adjust the model's resolution and dynamic parameters.

10. A digital twin-based adaptive path planning system for tunnel lining trolleys, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the adaptive path planning method for tunnel lining trolley based on digital twin as described in any one of claims 1 to 9.

Images