A method for adaptive in-situ 3D printing of TBM lining

Abstract

This invention discloses an adaptive in-situ 3D printing method for TBM lining, belonging to the field of tunnel construction and lining technology. The method includes: acquiring segmented point cloud data through surrounding rock scanning; using adaptive voxel downsampling and mode-switchable coarse registration combined with ICP fine registration to stitch the point cloud together, generating a continuous inner surface of the surrounding rock; constructing a target lining model based on the inner surface of the surrounding rock, following the principle of filling convex pits and avoiding concave peaks; employing a dual-channel slicing strategy based on the geometric features of the target tunnel, performing axial slicing on the target tunnel model and radial slicing on the lining model, and reversing the slice sequence; generating a printing path through safety zone construction and adaptive path filling of gap width; determining process parameters using a lookup table method; and achieving closed-loop control through real-time monitoring. This invention achieves high-precision adaptive molding of TBM lining, significantly improving construction efficiency, molding quality consistency, and intelligent level.

Description

Technical Field

[0001] This invention belongs to the field of tunnel construction and lining technology, and particularly relates to an adaptive in-situ 3D printing method for TBM lining. Background Technology

[0002] To efficiently construct tunnels capable of withstanding long-term rock pressure, operational loads, and external forces such as earthquakes, hard rock tunnel boring machines (TBMs) have become the most commonly used mechanized excavation equipment. Their construction generally includes excavation, muck removal, support, and tunneling. Lining is a permanent support structure installed along the perimeter of the tunnel to control rock deformation and prevent collapse. It is typically made of reinforced concrete, with the type selected based on the surrounding rock conditions. Lining is divided into primary lining and secondary lining: in TBM construction, primary lining often uses a steel arch frame combined with shotcrete, while secondary lining mainly relies on the assembly of precast concrete lining blocks. While current secondary lining methods can provide stable support, they still fall short in terms of cost, process complexity, and quality control. Before molding, they require component production, transportation, and on-site assembly; the production process includes steel reinforcement cage fabrication, pouring, finishing, curing, and inspection, resulting in long cycles, large land occupation, and high labor costs; transportation is subject to weight and safety constraints, and is prone to damage during loading, unloading, and transshipment, leading to high logistics costs; assembly quality depends on the coordination of personnel and equipment, and hoisting can easily cause swaying and collisions, posing safety hazards. Extrusion-based 3D printed concrete uses a digital model to drive the nozzle, extruding layer by layer without the need for templates, enabling rapid molding of complex curved surfaces and local thickening. Recent research has proposed integrating this technology with TBM secondary lining construction, hoping to replace or partially replace the traditional precast lining "production-transportation-assembly" process through in-situ additive manufacturing, improving construction continuity and intelligence, simplifying procedures, shortening construction time, reducing costs and increasing efficiency, and improving lining quality and construction safety.

[0003] Existing in-situ 3D printing lining processes generally exhibit characteristics of "human experience-driven, pre-defined path execution" at the implementation level: before construction, fixed printing trajectories and process parameters are generated offline based on design standard cross-sections or limited measurement information. During on-site printing, there is a lack of online perception, identification, and dynamic updating mechanisms for the convergence deformation, local undulations, and obstacle distribution of the actual excavation cross-section of the TBM. Key parameters such as nozzle posture, extrusion speed, and printing speed mainly rely on manual adjustment by operators based on experience. At the same time, existing methods often use the design lining identification center as the geometric reference for slice and path generation. However, when there are irregular undulations such as pits and peaks on the actual surrounding rock surface, the identification and fitting of the lining center are prone to deviation, which leads to slice reference drift, interlayer contour misalignment, and path offset, further amplifying printing errors. When there are deviations between the actual geometry of the surrounding rock and the preset 3D model, the printing path is difficult to adaptively correct in a timely manner, which can easily lead to problems such as insufficient adhesion between the outer surface of the lining and the surrounding rock, uneven thickness distribution, or local deviations. Subsequent steps often require repeated patching, trimming, or secondary grouting to eliminate these gaps, significantly increasing construction time and economic costs. Furthermore, it may cause safety risks such as nozzle-surrounding rock collisions due to path mismatch with the environment. In summary, current in-situ 3D printed linings have not yet established an integrated technical system driven by the actual geometry of the surrounding rock, encompassing "acquisition of actual surrounding rock geometry—target lining modeling—target lining slicing—printing path filling—process parameter decision-making—process monitoring closed-loop control." Therefore, it is difficult to fully unleash the potential advantages of in-situ additive manufacturing in overall forming and geometric adaptation. Summary of the Invention

[0004] To address the above problems, this invention proposes an adaptive in-situ 3D printing method for TBM lining.

[0005] The technical solution adopted in this invention is as follows: An adaptive in-situ 3D printing method for TBM lining includes the following steps: (1) Establish a coordinate system benchmark consistent with the tunnel axis in the target tunnel, calibrate the vision module and plan the scanning station spacing, walking route, field of view coverage and overlap rate of each station; (2) Scan the surrounding rock surface segment by segment according to the predetermined station location and obtain the point cloud data of the zone, and the point clouds of adjacent zones overlap. (3) Perform data cleaning and adaptive voxel downsampling on the segmented point cloud, and align the point cloud to a unified coordinate system by using a mode-switchable coarse registration algorithm combined with ICP fine registration, and output the stitched surrounding rock point cloud. (4) Extract the continuous inner surface of the surrounding rock from the spliced ​​surrounding rock point cloud as the bonding reference, and generate the outer surface of the lining by following the shape complementary principle of filling the pits and avoiding the peaks. Then, construct a printable lining model according to the design thickness. (5) Based on the geometric features of the target tunnel, a dual-channel slicing strategy is adopted: axial slicing is performed on the target tunnel model along the tunnel axis, and the position of the lining model in the target tunnel is located by axial slicing; radial slicing is performed on the lining model along the radial direction of the lining to generate a layered closed slice profile, and the slice profile of the radial slice is reversed in the whole according to the layer sequence. (6) Unfold the slice outline onto a two-dimensional plane, construct a safe zone and identify the four sides, and adaptively generate a zigzag filling path based on the relationship between the gap width and the track width. After being clipped and bridged by the safe zone, the path is projected back onto the three-dimensional coordinate system to generate a continuous print path point series. (7) Based on the run width and slice thickness, consult the pre-calibrated empirical parameter table and match the process parameters including nozzle diameter, extrusion speed and printing speed; (8) Integrate the printing path points, process parameters and equipment control instructions into a task package, and send it to the printing equipment after coordinate system transformation and motion planning to start lining printing; monitor the printing process in real time and adjust the process parameters to achieve closed-loop control of molding quality.

[0006] Furthermore, the rock mass point cloud splicing process in step (3) includes: (3.1) Read the source point cloud data and target point cloud data of the surrounding rock of the tunnel. The source point cloud is a block point cloud that needs to be aligned, and the target point cloud is a fixed reference point cloud. The point cloud coordinates are normalized from millimeter level to meter level. (3.2) Remove outliers with non-finite values ​​and coordinate anomalies from the source point cloud and the target point cloud, adaptively estimate the voxel size based on the bounding box size and the set voxel upper limit, perform voxel mesh downsampling, and control the overall number of primes. (3.3) The coarse registration matrix is ​​obtained through a mode-switching coarse registration method, wherein mode one uses the SAC-IA algorithm and mode two uses the Super4PCS algorithm. (3.4) Using the coarse registration result as input, the transformation matrix is ​​optimized by the ICP fine registration algorithm to obtain the fine registration matrix; (3.5) Multiply the fine registration matrix with the coarse registration matrix to obtain the global total transformation matrix, realize the alignment and stitching of the block point cloud, and output the stitched surrounding rock point cloud.

[0007] Further, step (5) includes: (5.1) Input the dual STL model of the target tunnel model and the target lining model, extract the vertex of the model to generate a unique vertex set, and perform deduplication and rounding operations. (5.2) Obtain the approximate center and radius of the target tunnel at the end of the target tunnel. Translate and rotate the dual STL model to the end center. Using the target tunnel model as the reference, normalize the target lining model according to the radius r_core to obtain the length of the target tunnel after removing the influence of irregular lining. Transform each vertex of the lining triangular mesh from the Cartesian coordinate system to the cylindrical coordinate system (φ, k, r) to form a set of triangular facets represented by φ-kr triples. (5.3) Processing is performed only on lining area points whose radial coordinates are greater than the target tunnel radius. The expansion joint is adaptively calculated based on the overall angle distribution, and the points in the triangular mesh whose vertex angles are greater than the expansion joint are processed as a whole. (5.4) Dual-channel slicing strategy: Perform axial slicing along the tunnel axis on the target tunnel model to locate the position of the lining model in the target tunnel axis through axial slicing; perform radial slicing along the lining radial direction on the lining model to generate a layered closed slice profile. (5.5) Perform a layer sequence reversal on the slice outline of the radial slice, so that the slice order is changed from the direction of increasing radius to the direction of decreasing radius.

[0008] Furthermore, the formula for calculating the development seam is toX = max(φ) - min(φ), where φ is the angle.

[0009] Furthermore, the process of performing axial slicing along the tunnel axis on the target tunnel model includes: In the Cartesian coordinate system, based on the target tunnel length and slice thickness, a z-direction slice sequence is generated by slicing z-isosurfaces. For each triangular facet in the target tunnel model, its lower and upper bounds in the z-direction are pre-calculated. For a given slice layer, only triangular facets belonging to the range between the lower and upper bounds are retained as a candidate triangular facet set. The plane z = z slice Intersections with each candidate triangular facet are calculated, and intersection points are generated on the three edges to form a set of line segments. Closed loops are constructed through coordinate deduplication, connected component partitioning, and depth-first traversal, with NaN values ​​inserted between adjacent loops as separators; where z slice This represents the height of the current slice layer along the Z-axis in the Cartesian coordinate system.

[0010] Furthermore, the process of performing radial slicing along the radial direction of the lining model includes: In a cylindrical coordinate system, based on the target tunnel radius and slice thickness, an r-direction slice sequence is generated by slicing isosurfaces in the r direction. For each triangular facet in the lining model, its lower and upper bounds in the r direction are pre-calculated. For a given slice layer r... slice Only triangular faces belonging to the range from the lower bound to the upper bound are retained as the candidate triangular face set; using the cylindrical surface r = r sliceIntersecting with each candidate triangular facet one by one, generating a set of line segments and constructing a closed loop, and finally transforming x = r slice *cos(φ), y = r slice *sin(φ) and z = k transform the sliced ​​results from cylindrical coordinates back to Cartesian coordinates; where r slice This represents the radius of the cylindrical surface corresponding to the current slice layer.

[0011] Further, step (6) includes: (6.1) Read the lining slice results and the width setting, transform the three-dimensional slice contour points of each layer from the Cartesian coordinate system to the cylindrical coordinate system, obtain the cylindrical coordinates (r, φ, k) of each point, and merge the Cartesian coordinate and cylindrical coordinate points to form an N×6 sequence, using NaN as the segment boundary to maintain the path topology within the layer; (6.2) Unfold the outline of each slice from the cylindrical surface to a two-dimensional plane, and set the x-coordinate on the two-dimensional plane to... plane and the vertical coordinate y plane Represented as x plane =φ*r、y plane =k, generate a uniformly dense planar polygon by resampling the arc length at equal intervals, and construct an axis-aligned circumscribed rectangle so that the long or short side of the rectangle is aligned with the axis of the target tunnel. (6.3) Perform full wrap detection on polygons at the same level to construct a safe zone; (6.4) Based on a preset angle threshold, detect the included angle of the vertices of the circumscribed rectangle after shrinkage, and identify the two opposing long sides and their gap width; (6.5) Within the safe zone, based on the relationship between the gap width between the two identified opposing long sides and the track width, three path filling scenarios are adaptively selected: when the gap width is not less than twice the track width, multiple parallel zigzag filling lines are generated with the long side as the baseline; when the gap width is between the track width and twice the track width, two long sides are retained as filling lines; when the gap width is less than the track width, only one long side is retained as a filling line. (6.6) Cut the intersection of the fill line and the safe area to remove out-of-range line segments, then bridge multiple paths in the same layer using a zigzag strategy, connect them through linear interpolation and cut them again, and finally form a continuous path in the layer using NaN as the separator. (6.7) Project the two-dimensional in-layer path back to the cylindrical coordinate system and transform it back to the Cartesian coordinate system to generate a three-dimensional printing path point list containing multi-dimensional parameters.

[0012] Furthermore, the process of constructing the safe zone in step (6.3) is as follows: Perform a full wrap detection on all expanded polygons in the same layer. Polygons that are completely wrapped by other polygons are marked as inner layers, and the rest are outer layers. Define the safe zone by shrinking the outer polygons inward by half the width of the track and expanding the inner polygons outward by half the width of the track.

[0013] Furthermore, in step (6.6), the zigzag strategy for bridging multiple paths at the same level specifically refers to: Each time, starting from the end of the current line segment, the endpoint of the next line segment closest to it is calculated based on the spatial distance and selected as the starting endpoint. This endpoint is then selected as the connection target, and a smooth transition bridging path is generated using the equal-spacing arc length linear interpolation method. During the interpolation process, interpolation points are evenly distributed based on the arc length.

[0014] Further, step (8) includes: (8.1) Integrate the printing path points, process parameters and equipment control instructions into a task package, and transform the path from the design coordinate system to the robot base coordinate system and perform online or offline motion planning, including acceleration and deceleration control, attitude limiting and collision or soft limit checks; send the data to the robot controller and stepper motor driver through the serial port protocol, and set pause, resume, emergency stop and breakpoint resume printing mechanism. (8.2) During printing, the extrusion pressure, pump speed or valve opening, actual robot speed and trajectory, forming layer thickness and line width are collected in real time and compared with the target values; when the layer thickness, width or position deviation exceeds the threshold, the extrusion speed or printing speed is adjusted in conjunction with the empirical parameter table or process window.

[0015] Furthermore, during the lining printing process, extrusion pressure, printing speed and trajectory, forming layer thickness and line width data are collected in real time. The collected data are compared with the determined process parameters and the expected geometric target of the printing path. When the deviation exceeds the threshold, the extrusion speed or printing speed is adaptively adjusted, or a local path micro-correction and re-printing for the current printing task is triggered, thereby realizing closed-loop control of TBM lining in-situ 3D printing.

[0016] The beneficial effects of this invention are: (1) The TBM lining adaptive in-situ 3D printing method proposed in this invention achieves a breakthrough in tunnel secondary lining construction technology by constructing an integrated process driven by the real geometry of the surrounding rock. Its core advantage lies in the combination of surrounding rock environment point cloud stitching, lining slicing based on the target tunnel, and printing path filling generation to form a complete adaptive printing method of surrounding rock scanning, point cloud stitching, target lining modeling, slicing, path planning, parameter decision-making, and closed-loop control. This method effectively solves the problem in the prior art that the real geometry of the surrounding rock deviates from the preset three-dimensional model due to the convergence deformation of the actual excavation section of the TBM, local undulations, and obstacle distribution, and the difficulty in timely adaptive correction of the printing path, significantly improving the fit and thickness consistency between the lining and the surrounding rock.

[0017] (2) The present invention is based on an adaptive voxel downsampling mode switchable coarse registration strategy, which can quickly obtain accurate data of the inner surface of the surrounding rock under large-scale block point cloud conditions, providing a reliable benchmark for the "concave-convex" of the lining; the dual-channel slicing strategy is adopted to anchor the benchmark to the geometric features of the target tunnel, effectively avoiding the center identification deviation caused by local concavity and convexity, and ensuring the stability of the slicing results.

[0018] (3) The present invention utilizes a path generation strategy of constructing a safe zone, identifying four sides and filling the path with a zigzag pattern for three gap conditions to achieve quantitative filling of pits and active avoidance of peaks, providing a directly executable path point sequence for in-situ adaptive 3D printing of TBM lining, significantly reducing the risk of empty travel and collision with printing equipment, and improving printing efficiency and consistency of molding quality.

[0019] (4) This invention uses a lookup table method to reason about process parameters and a closed-loop control of process monitoring to enable the extrusion speed and printing speed to be automatically optimized and adjusted according to real-time working conditions. This ensures that the speed is automatically reduced and stabilized in the geometric under-extrusion zone and automatically increased and improved in the over-extrusion zone. When necessary, it can trigger local re-printing or path correction, thereby effectively suppressing line width fluctuations and interlayer defects, and achieving consistent output of lining forming quality.

[0020] In summary, this invention significantly reduces the reliance on manual experience in in-situ 3D printing while effectively ensuring lining fit, thickness uniformity, and construction safety. It provides key technical support and application guarantee for efficient, adaptive, and intelligent closed-loop construction operations of in-situ additive manufacturing of TBM tunnel linings. Attached Figure Description

[0021] Figure 1 This is a flowchart of an adaptive in-situ 3D printing method for TBM lining.

[0022] Figure 2 This is a flowchart of point cloud splicing for the surrounding rock environment.

[0023] Figure 3It is source cloud data of the surrounding rock of the tunnel.

[0024] Figure 4 It is target point cloud data of the surrounding rock of the tunnel.

[0025] Figure 5 It is the result of point cloud splicing of the surrounding rock environment.

[0026] Figure 6 This is a flowchart of the lining slicing process based on the target tunnel.

[0027] Figure 7 It is a model of the target tunnel and its lining when there are protrusions on the surface of the surrounding rock during excavation.

[0028] Figure 8 It is a model of the target tunnel and its lining when there are pits on the surface of the surrounding rock during excavation.

[0029] Figure 9 It is the result of a three-dimensional slice of the lining when there are protrusions on the surface of the excavated surrounding rock.

[0030] Figure 10 This is the result of a three-dimensional slice of the lining when there are pits on the surface of the excavated surrounding rock.

[0031] Figure 11 It prints a flowchart by filling in the path.

[0032] Figure 12 It is the structure and path filling result of the safe zone when there are protrusions on the surface of the excavated surrounding rock.

[0033] Figure 13 It is the structure of the safe zone and the result of path filling when there are pits on the surface of the excavated surrounding rock.

[0034] Figure 14 It is the result of printing the path filling layer by layer when there are protrusions on the surface of the excavated surrounding rock.

[0035] Figure 15 It is the result of printing the path filling layer by layer when there are pits on the surface of the excavated surrounding rock.

[0036] Figure 16 It is the two-dimensional projection result of the lining printing path when there are protrusions on the surface of the excavated surrounding rock.

[0037] Figure 17 It is the two-dimensional projection result of the lining printing path when there are pits on the surface of the excavated surrounding rock.

[0038] Figure 18 It is the 3D printing path for lining when there are protrusions on the surface of the excavated surrounding rock.

[0039] Figure 19 It is the 3D printing path for lining when there are pits on the surface of the excavated surrounding rock. Detailed Implementation

[0040] The present invention will be further described and illustrated below with reference to specific embodiments. The embodiments described are merely examples of the content of this disclosure and do not limit the scope of the invention. The technical features of each embodiment in the present invention can be combined accordingly, provided that there is no mutual conflict.

[0041] This invention proposes an adaptive in-situ 3D printing method for TBM lining, applicable to various TBM excavation tunnel rock undulations and convergence deformation conditions. It focuses on the core function of secondary tunnel lining construction: achieving in-situ adaptive forming of the lining under the constraints of real surrounding rock geometry. Its core lies in maximizing the automation and intelligence of the entire process—from acquiring the real geometry of the surrounding rock, modeling the target lining, slicing the target lining, filling the printing path, making process parameter decisions, to closed-loop control of process monitoring—reducing the accumulation of errors caused by reliance on manual experience and preset models in traditional in-situ printing.

[0042] This invention mainly comprises the following core components: surrounding rock environment point cloud stitching, lining slicing based on the target tunnel, and printing path filling generation. Surrounding rock environment point cloud stitching consists of steps such as data reading and writing and scale unification, outlier removal and adaptive voxel downsampling, coarse registration (mode switchable), ICP fine registration, and result fusion. It is primarily responsible for quickly and robustly aligning and stitching the segmented surrounding rock environment point clouds to a unified coordinate system, outputting a continuous and reliable inner surface of the surrounding rock as the "fitting reference" for the outer surface of the target lining. Lining slicing based on the target tunnel consists of steps such as model loading and preprocessing, coordinate reference and centering, joint setting and angle translation, dual-channel slicing strategy, and sequence reversal. It is primarily responsible for avoiding center estimation deviations caused by local pits / peaks in the surrounding rock, anchoring the slicing reference to the geometric features of the target tunnel, thereby achieving rapid and accurate slicing of the lining at the section or sector level, and providing reliable layered closed contour data for printing path filling generation. The printing path filling generation consists of several steps, including data preprocessing and coordinate transformation, surface unfolding and polygon modeling, enclosed area determination and safe zone construction, quadrilateral recognition, path filling for three gap scenarios, safe zone trimming and zigzag bridging, cylinder return projection, and 3D path assembly. It is primarily responsible for rapidly generating continuous 3D printing paths for cross-section or sector-level linings, enabling quantitative filling of surrounding rock depressions and active avoidance of peaks, thus providing directly executable path points for TBM in-situ adaptive 3D printing. Furthermore, in the TBM lining adaptive in-situ 3D printing process, the above three core steps are systematically linked: on-site preparation and calibration, surrounding rock scanning and data acquisition, target lining model construction, lookup table-based reasoning for process parameters, upper-computer task packaging and distribution, and process monitoring and closed-loop control. This effectively ensures the real-time adaptive capability, construction efficiency, and consistent forming quality of the TBM tunnel lining in-situ 3D printing.

[0043] like Figure 1As shown, the present invention proposes an adaptive in-situ 3D printing method for TBM lining, comprising the following steps: S1. On-site preparation and calibration: After entering the site, the vision module is calibrated for internal and external parameters and distortion correction is performed. The measurement dimensions and resolution are determined (e.g., display to the millimeter level). Several stable positioning markers are set up in the tunnel or obvious natural features are selected to establish a benchmark framework that is consistent with the tunnel axis. At the same time, the spacing between scanning stations, the walking route, the field of view coverage and overlap rate of each station are planned. Dust, vibration and supplementary lighting conditions are assessed in advance. Finally, an executable calibration document, coordinate system definition and scanning operation plan are formed.

[0044] S2, Surrounding Rock Scanning and Data Acquisition: The surrounding rock surface is scanned segment by segment according to the predetermined station location and zoning order. The original point cloud is acquired at each station and sufficient overlap is ensured between adjacent stations to facilitate subsequent registration. During the acquisition process, the point density, missing areas, and voids caused by reflection / shadow are checked in real time. Key structures (near peaks and pits) are supplemented or scanned from multiple angles. Finally, the original point cloud packets of multiple segments in each zone are output.

[0045] S3, Surrounding Rock Environment Point Cloud Stitching: For the source and target point clouds to be stitched, first perform basic data cleaning (remove isolated noise points, floating points, and other outliers), then perform voxel downsampling to unify density and reduce computational load; use coarse registration to obtain the initial rigid body pose, then use fine registration with an iterative nearest-point strategy to minimize errors as much as possible, and finally output the fused point cloud and the total transformation matrix in the same coordinate system. Repeating this coarse and fine registration operation completes the stitching of the zoned surrounding rock environment point clouds.

[0046] S4, Target Lining Model Construction: Based on the surrounding rock environment point cloud already stitched together in a unified coordinate system, the effective area of ​​the tunnel inner surface is first extracted and non-target points are cleaned up. Necessary interpolation / resampling is performed on sparse areas of the point cloud to ensure surface continuity and reliability. Subsequently, the inner surface point cloud is reconstructed into a reference triangular mesh or continuous surface, which serves as the fitting reference for the outer surface of the lining. The principle of complementary shape and thickness constraint is followed when modeling the target lining: where there are depressions in the surrounding rock, the outer surface of the target lining protrudes accordingly to fill the depressions; where there are peaks in the surrounding rock, the outer surface of the target lining protrudes accordingly to avoid the peaks, thus ensuring that the fit between the lining and the surrounding rock meets the design requirements. Based on this, the inner surface of the lining is generated according to the design thickness, resulting in a printable 3D model of the target lining. The model can be generated and corrected using any stable STP / mesh format, and finally exported uniformly to STL format for slicing and path planning.

[0047] S5, Lining Slicing Based on the Target Tunnel: Read the 3D STL model of the target tunnel and target lining, establish a slicing coordinate framework consistent with the tunnel geometry, and expand the STL model to a parameter domain that is easy to process as needed; then slice layer by layer according to the set layer height (i.e., the thickness of one slice), and select only the triangular facets that may intersect with the layer (i.e., the triangular facets that will be cut by the cutting plane and generate effective intersection segments under the current slice thickness) for intersection calculation to obtain the contour line segments of the layer; after deduplicating and sorting the endpoints of the line segments, automatically connect them into several closed loops, including the outer contour and the inner hole, and unify the direction and order of the loops; finally, reverse the overall outline of each layer of lining slices according to the layer sequence and save the output (with NaN segmentation markers between loops), providing stable slicing result input for subsequent printing path filling generation.

[0048] S6, Printing Path Fill Generation: Unfold the outline of each layer of lining slices onto a 2D plane, then use the bounding rectangle of the outline shape as the filling base shape, and construct a safety zone (outer polygon shrinks inward, inner polygon expands outward) to ensure that the printing path maintains a safe distance from the boundary; identify the relationship between the long and short sides of the bounding rectangle after shrinking based on the angle relationship, and generate a zigzag internal fill according to the actual width of the outline or simply follow along the edge to ensure that a printable path can be formed even in narrow areas; trim the generated printing path into the safety zone, and bridge multiple paths of the same layer into continuous lines according to the principle of proximity to reduce idle travel, and then trim the safety zone again; finally, project the 2D path back into the 3D tunnel coordinates, output and save it in layer order, and provide a directly usable path point sequence for subsequent robot execution.

[0049] S7, Look-up table method for reasoning and decision-making on process parameters: After obtaining the layered contour and printing path of the target lining, the system sets and matches the empirical parameter table or process window established in advance through calibration tests based on the runner width and layer thickness, and simultaneously determines parameters such as nozzle diameter, extrusion speed, and printing speed; if the local geometry falls near the boundary of the process window (e.g., too narrow, abrupt curvature change, or sudden increase in thickness), the system will trigger a conservative strategy, such as prioritizing forming stability and safety clearance, and providing an adjustable range for manual confirmation. Table 1 shows the empirical parameter table established in advance through calibration tests.

[0050] Table 1. Empirical parameter table established in advance through calibration experiments. S8, upper computer task packaging and distribution: integrates hierarchical path point list, process parameters, and equipment control instructions into a task package, while transforming the path from the design coordinate system to the robot base coordinate system and performing online / offline planning (acceleration / deceleration, attitude limiting, collision / soft limit checks); distributes the data to the robot controller and stepper motor driver via serial port protocol, sets pause / resume, emergency stop, and breakpoint resume mechanisms to ensure reliable execution of on-site 3D printing lining.

[0051] S9, process monitoring and closed-loop control: During printing, extrusion pressure, pump speed / valve opening, robot actual speed and trajectory, forming layer thickness and line width are collected in real time and compared with target values; when the layer thickness / width / position deviation exceeds the threshold, the extrusion speed or printing speed is adjusted in conjunction with the empirical parameter table or process window, and local re-printing or path micro-correction is triggered when necessary to achieve closed-loop control of TBM lining in-situ 3D printing.

[0052] In one specific embodiment of the present invention, the flowchart for splicing point clouds of the surrounding rock environment is as follows: Figure 2 As shown, the specific process and technical solution for point cloud stitching of the surrounding rock environment are as follows: (1) Data reading and writing and scale unification Read the source point cloud data and target point cloud data (PLY) of the tunnel surrounding rock, respectively as follows: Figure 3 and Figure 4 As shown, the source point cloud represents a section of tunnel surrounding rock that needs to be aligned, and the target point cloud represents another section of tunnel surrounding rock that serves as a fixed reference. When stitching together the tunnel surrounding rock point clouds obtained from the two stations, the point cloud scanned first or selected as the reference is designated as the target point cloud, and the point cloud scanned later and required to be aligned with it is designated as the source point cloud. Both are in PLY format, and their respective bounding box ranges are output. The point cloud coordinates are normalized from millimeters to meters (i.e., for x, y, z coordinates, execute: x,y,z←0.001*(x,y,z)) to ensure numerical stability and consistency with subsequent parameters (meter level).

[0053] (2) Outlier removal and adaptive voxel downsampling Outliers with non-finite values ​​and abnormal coordinates in both the source and target point clouds are removed to form a clean point set. Then, based on the bounding box size and a set voxel upper limit, the voxel size is adaptively estimated and the voxel grid is downsampled to control the overall number of voxels (adaptively increasing the voxel count from the initial value until the voxel grid number falls below a threshold or reaches the set voxel upper limit), avoiding computational explosion in subsequent feature and matching stages. This process records the downsampling ratio and voxel size.

[0054] (3) Coarse registration (mode can be switched) Mode 1: SAC-IA (PCL) SAC-IA coarse registration can be understood as first calculating a local shape fingerprint for each point, and then using random trials to find the most suitable alignment pose. The specific process is as follows: First, estimate the normal vector of each point in the source / target point clouds after adaptive voxel downsampling. Within a certain neighborhood radius, count the relative positions and normal vector changes of the surrounding neighboring points of each point, and mark this information as the local shape fingerprint of that point. Based on the idea of ​​the random sampling consensus algorithm, repeatedly and randomly extract a small number of points from the source point cloud, use their local shape fingerprints to match the most similar points in the target point cloud, temporarily assuming these pairs of points correspond, and solve a set of rigid body transformations (rotation + translation); after transforming the source point cloud as a whole, count the number of points that can match the target point cloud within a reasonable distance. If the proportion is high enough, it means that this set of transformations conforms to both the local shape and the overall geometric relationship, and it is used as the transformation matrix coarse_tf(T) of the coarse registration result. coarse Finally, this set of matrices is applied to the original high-resolution source point cloud to provide the initial pose for the more detailed registration (ICP) process.

[0055] Mode 2: Super4PCS (OpenGR) Super4PCS coarse registration can be understood as traversing and matching four geometrically similar points in the source and target point clouds. The four-point correspondence with the maximum overlap and the solved rigid transformation are the final coarse registration pose. It is a global search method based entirely on geometric relationships. Several key parameters are set: geometric error, maximum running time, number of sampling points, and the approximate overlap ratio of the two point clouds. Four points are randomly selected multiple times in the target point cloud to form a small skeleton. Using only the geometric relationships between these four points, such as the side length ratio and included angle, which do not change with the rigid body transformation, another set of four similar points in shape is searched in the source point cloud. For each matching of such a four-point correspondence, a set of candidate rigid body transformations is solved, transforming the source point cloud as a whole into the target coordinate system. Then, the number of source points that overlap with the target points within the geometric error range is counted. The algorithm continuously tries different combinations of four points and correspondences within a given time, and finally selects the transformation matrix with the highest overlap as the coarse registration result coarse_tf(T) of Super4PCS. coarse This matrix is ​​then applied to the original source point cloud, providing a globally robust starting point that is insensitive to the initial pose for subsequent fine registration.

[0056] (4) ICP precision registration ICP fine-tuning can be understood as a process of continuously fine-tuning the pose based on existing coarse registration, increasing the overlap between the source and target point clouds. Using the coarsely registered source point cloud as the input source and the full-resolution target point cloud as the input target, a matching allowable distance (e.g., only points within 3mm are considered candidates for the same point), a convergence threshold, and a maximum number of iterations are set. Each iteration then performs the same three steps: ① For each point in the source point cloud, match the nearest point in the target point cloud as the current corresponding point pair, and discard pairs whose distance exceeds the matching allowable distance; ② Based on the nearest point pair, use the least squares method to find a new set of rigid body transformations (rotation + translation) to minimize the overall error of these corresponding point pairs; ③ Apply this set of transformations to the source point cloud to update its position. The process involves repeatedly finding the nearest point, estimating the transformation, and applying the transformation, while monitoring changes in the transformation amount and error. If, after continuous iterations, the change is sufficiently small, the error decreases insignificantly, or the maximum number of iterations is reached, convergence is considered achieved. The resulting transformation matrix is ​​the fine registration result of ICP, icp_tf(T). ICP ).

[0057] (5) Results fusion Multiplying the fine registration matrix and the coarse registration transformation matrix yields the final global total transformation matrix: total_tf = icp_tf · coarse_tf (T total =T ICP ·T coarse This achieves the alignment and stitching of the source point cloud to the target point cloud. The stitching result of the surrounding rock environment point cloud is as follows: Figure 5 As shown, by quickly and robustly aligning and stitching the segmented surrounding rock environment point clouds to a unified coordinate system, it can be used to obtain a subsequent continuous and reliable inner surface of the surrounding rock as a bonding reference for the outer surface of the target lining.

[0058] In one specific embodiment of the present invention, the lining slicing process based on the target tunnel is as follows: Figure 6 As shown, the specific steps include: (1) Model loading and preprocessing The input uses a dual STL model: a target tunnel model and a target lining model placed on the target tunnel. Vertex extraction is performed on the model to generate a unique vertex set (through deduplication and rounding) to reduce noise and suppress floating-point jitter. The target tunnel and lining model when there are convex peaks on the excavated surrounding rock surface is shown below. Figure 7 As shown, the target tunnel and lining model are shown when there are pits on the surface of the excavated surrounding rock. Figure 8 As shown.

[0059] (2) Coordinate reference and centering An approximate center (xc, yc, zc) and radius r_core are obtained at the end of the target tunnel. The STL model is then translated and rotated to the end center, and the heterogeneous model is scaled according to r_core to obtain the target tunnel length h_core after removing the influence of irregular lining. Each vertex of the lining triangular mesh is transformed from Cartesian coordinates (x, y, z) to cylindrical coordinates (φ, k, r), forming a set of triangular facets represented by φ-kr triples. This ensures that subsequent lining slicing is performed within a unified parameter domain, reducing fault and splicing errors.

[0060] (3) Joint setting and angle translation This step only processes points where r > r_core, which are the lining area. To eliminate the breakage caused by the expansion domain spanning ±π, the expansion joint toX is adaptively calculated based on the overall φ distribution (take toX = max(φ) - min(φ)). Points in the triangle vertex where φ > toX are processed as a whole to ensure that the path with a span of 2π is continuous and without jumps in the expansion domain. This strategy is particularly critical in the radial slicing of the lining, as it can seamlessly cross the joint.

[0061] (4) Dual-channel slicing strategy This invention involves axial slicing of the target tunnel and radial slicing of the lining.

[0062] Target tunnel axial slices (Cartesian coordinate system): Based on the target tunnel length h_core and the given slice thickness slice_height, slice the tunnel according to the z-isosurface and generate a z-slice layer sequence z_slices; for each triangular facet in the target tunnel STL model, pre-calculate its lower bound lb and upper bound ub in the z-direction, and for a given slice layer z... slice Only keep z slice Triangular facets ∈ [lb, ub] are used as the candidate triangular facet set for this layer, thus avoiding redundant intersection calculations for all triangular facets globally. For each layer slice, the plane z=z is used. sliceIntersections with candidate triangular faces are calculated one by one, with intersection points and corresponding Boolean masks calculated on the three edges p1p2, p2p3, and p3p1 respectively. Triangular faces truly truncated by the plane are selected, and at most one line segment (consisting of two valid intersection points) is generated for each triangular face, thus obtaining the line segment set for this layer. The start and end coordinates of all line segments are deduplicated and sorted, establishing a mapping from node coordinates to node indices, and constructing an undirected edge table accordingly. Based on this, line segments are treated as edges of a graph, and edges in each connected subgraph are sequentially concatenated into several closed loops through connected component partitioning and depth-first traversal. For each loop, its traversal direction is unified according to the relationship between the start and end points and the coordinate sequence; if the direction does not match the expectation, the entire loop is reversed to ensure consistent loop direction. Finally, all closed loops within the same layer are sequentially concatenated into an array, with NaN values ​​inserted between adjacent loops as separators.

[0063] Lining radial slicing (cylindrical coordinate system): Based on the target tunnel radius r_core and the given slice thickness slice_height, slice according to r isosurface and generate r slice layer sequence r_slices (add a small offset (e.g., 10) to each radius value). -2 (To avoid numerical instability caused by cutting exactly on a vertex or edge); pre-calculate the lower bound lb and upper bound ub in the r direction for each triangular facet in the lining STL model, for a given slice layer r. slice Only keep r slice Triangular facets ∈ [lb, ub] are used as the candidate triangular facet set for this layer, thus avoiding redundant intersection calculations for all triangular facets globally. For each layer slice, the cylindrical surface r = r slice Intersections with candidate triangular faces are calculated one by one, with intersection points and corresponding Boolean masks calculated on the three edges p1p2, p2p3, and p3p1 respectively. Triangular faces truly pierced by the cylindrical surface are selected, and at most one line segment (consisting of two valid intersection points) is generated for each triangular face, thus obtaining the line segment set for this layer. The starting and ending coordinates of all line segments are deduplicated and sorted, establishing a mapping from node coordinates to node indices, and constructing an undirected edge table accordingly. Based on this, line segments are treated as edges of a graph, and edges in each connected subgraph are sequentially concatenated into several closed loops through connected component partitioning and depth-first traversal. For each loop, its traversal direction is unified according to the relationship between the start and end points and the coordinate sequence; if the direction does not match the expectation, the entire loop is reversed to ensure consistent loop direction. Finally, all closed loops within the same layer are sequentially concatenated into an array, with NaN inserted between adjacent loops as separators. The x=r operation is performed on each point on the loop. slice *cos(φ), y=r slice The operations *sin(φ) and z=k transform the lining slice result from cylindrical coordinates (φ,k,r) to Cartesian coordinates (x,y,z).

[0064] (5) Stratification reversal After completing the radial slicing of the lining and generating the slice layer sequence r_slices, the slice order needs to be adjusted to achieve the printing path planning from the outside to the inside. Specifically, the original slice order is r1→r2→r3→……→r n This means gradually cutting from the inside to the outside of the tunnel along an increasing radius direction. To accommodate in-situ 3D printing of the TBM lining, the lining slice sequence is synchronously reversed with r_slices, resulting in r_slices. n The slices are output in the order of r3→r2→r1. The 3D slice result for the lining when there are protrusions on the surface of the excavated surrounding rock is as follows: Figure 9 As shown, the three-dimensional slice results of the lining when there are pits on the surface of the excavated surrounding rock are as follows: Figure 10 As shown in the two images, the layered outline arrangement effect from the outside to the inside after inversion is clearly demonstrated.

[0065] In one specific embodiment of the present invention, the print path filling generation process is as follows: Figure 11 As shown, the specific steps include: (1) Data preprocessing and coordinate transformation Read the lining slice results and set the width. The slice results are the 3D path points (x, y, z) of each layer. Convert the point cloud in Cartesian coordinates to cylindrical coordinates to obtain the cylindrical coordinates of each point. The conversion formula is r = sqrt(x...). 2 +y 2 ), φ=arctan(y / x), k=z, and merge the six-dimensional parameters x, y, z, φ, r, k into an N×6 sequence; use NaN as the segment boundary to maintain the topological structure of multiple paths within the layer.

[0066] (2) Surface unfolding and polygon modeling For each slice, perform a cylindrical unfolding to a two-dimensional plane, where the x and y coordinates of the two-dimensional plane are x and y, respectively. plane =φ*r,y plane =k; The unfolded contour is converted into a planar polygon with uniform point density by resampling with equal-interval arc lengths to remove nearest-neighbor duplicate points and improve robustness; Subsequently, an axis-aligned circumscribed rectangle is constructed for each unfolded polygon so that the long or short side of the rectangle is aligned with the target tunnel axis.

[0067] (3) Envelope determination and safe zone construction Perform a complete wrapping detection on all polygons in the same layer, mark the polygons completely surrounded by other polygons as inner layers. The inner layer polygons do not participate in the filling of this layer and only serve as inner layer no-entry zones for the delineation of the safety zone. The safety zone is constructed through the following operations: shrink the entire outer layer polygon inward by width / 2, and at the same time expand the inner layer polygon outward by width / 2. The area obtained by taking the difference between the two is the safety zone, ensuring that a safety margin of ≥ width / 2 is reserved from the generated path to the inner and outer boundaries, effectively avoiding equipment collisions and ensuring the integrity of the printing contour.

[0068] (4)Four-sided recognition Shrink the circumscribed rectangle of the outer layer polygon by width / 2, and based on a preset angle threshold (≈20°), detect the included angle between adjacent sides at each vertex of the shrunk circumscribed rectangle. If the included angle is greater than the angle threshold, determine that the point is a sharp corner point. Separate the points into upper and lower groups according to the y plane coordinate, and then select the four corner points (p1 / p2 / p3 / p4) according to the x plane coordinate, and rearrange the starting point of the contour. Output the index pairs (l1, w1, l2, w2) of the four sides, where l1 / l2 are the two long sides and w1 / w2 are the two short sides, and extract the two opposite long sides (original side / ending side) and the minimum distance sep between them.

[0069] (5)Fill the path in three gap cases according to the relationship between the gap width and the track width Wide gap (sep≥2*width): Retain the long sides l1 and l2 and offset them equidistantly and parallelly with the long sides as the baseline, intersect with the outer layer polygon to generate multiple parallel filling lines. The starting and ending ends of each filling line are shrunk by width / 2 to avoid sticking to the edge, and form a zigzag path by alternately flipping the direction of the filling line to reduce subsequent connection jumps.

[0070] Medium gap (width≤sep<2*width): Only retain the long sides l1 and l2 and intersect with the outer layer polygon as filling lines. The starting and ending ends of each filling line are shrunk by width / 2 to avoid sticking to the edge, reducing printing defects caused by over-density and near the boundary.

[0071] Narrow gap (sep<width): After performing a safety shrinkage on the outer layer polygon, only intersect with the outer layer polygon along the side l1 or l2 to generate a process-feasible safe side walking route as the filling line. The starting and ending ends of the filling line are shrunk by width / 2 to avoid sticking to the edge, ensuring that a feasible path can still be formed in a narrow area and avoiding来回 in a narrow slit.

[0072] (6)Safety zone clipping and zigzag bridging All fill lines are intersected with the safety zone and trimmed, removing segments exceeding the safety margin. Then, a zigzag strategy is used to bridge multiple parallel fill lines in the same layer: each time, the direction of the endpoint closer to the end of the previous segment is selected, and the connection is made using linear interpolation (equal-spacing arc lengths); the safety zone is trimmed again to ensure that the bridging segments also meet the safety margin; finally, NaN is used as the segment end separator to form a continuous and safe intra-layer path. The safety zone structure and path filling results when there are peaks on the excavated surrounding rock surface are as follows: Figure 12 As shown, the safety zone structure and path filling results when there are pits on the surface of the excavated surrounding rock are as follows: Figure 13 As shown, when there are protrusions on the surface of the excavated surrounding rock, the result of the lining layer-by-layer printing path filling is as follows: Figure 14 As shown, when there are pits on the surface of the excavated surrounding rock, the result of the lining layer-by-layer printing path filling is as follows. Figure 15 As shown.

[0073] (7) Cylindrical return projection and three-dimensional path assembly Project the planar path back onto the cylindrical coordinate system (φ, r, k) according to each slice layer. Given the value of r for each slice layer, calculate φ = x. plane / r,y plane =k, further transformed back to the Cartesian coordinate system (x,y,z) (x=r*cos(φ), y=r*sin(φ), z=k), finally merged into a three-dimensional combined path containing x, y, z, φ, r, and k, providing directly callable point data for printing execution. The two-dimensional projection result of the lining printing path when there are convex peaks on the excavated surrounding rock surface is as follows: Figure 16 As shown, the two-dimensional projection result of the lining printing path when there are pits on the surface of the excavated surrounding rock is as follows: Figure 17 As shown, the 3D printing path for the lining is as follows when there are protrusions on the surface of the excavated surrounding rock. Figure 18 As shown, the 3D printing path for the lining is as follows when there are pits on the surface of the excavated surrounding rock. Figure 19 As shown.

[0074] This invention constructs an integrated TBM lining in-situ adaptive 3D printing process driven by the real geometry of the surrounding rock by sequentially linking three core links: surrounding rock environment point cloud stitching, target tunnel lining slicing, and printing path filling generation with other steps such as on-site preparation and calibration, surrounding rock scanning and data acquisition, target lining model construction, table lookup method reasoning and decision-making of process parameters, upper computer task packaging and distribution, and process monitoring and closed-loop control. It realizes rapid connection from "surrounding rock scanning - point cloud stitching - target lining modeling - target lining slicing - printing path planning - parameter decision-making - closed-loop control", which meets the needs of achieving in-situ adaptive forming of lining under the constraints of real surrounding rock geometry, reducing manual dependence and improving the intelligence of in-situ 3D printing of lining.

[0075] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. An adaptive in-situ 3D printing method for TBM lining, characterized in that, Includes the following steps: (1) Establish a coordinate system benchmark consistent with the tunnel axis in the target tunnel, calibrate the vision module and plan the scanning station spacing, walking route, field of view coverage and overlap rate of each station; (2) Scan the surrounding rock surface segment by segment according to the predetermined station location and obtain the point cloud data of the zone, and the point clouds of adjacent zones overlap. (3) Perform data cleaning and adaptive voxel downsampling on the segmented point cloud, and align the point cloud to a unified coordinate system by using a mode-switchable coarse registration algorithm combined with ICP fine registration, and output the stitched surrounding rock point cloud. (4) Extract the continuous inner surface of the surrounding rock from the spliced ​​surrounding rock point cloud as the bonding reference, and generate the outer surface of the lining by following the shape complementary principle of filling the pits and avoiding the peaks. Then, construct a printable lining model according to the design thickness. (5) Based on the geometric features of the target tunnel, a dual-channel slicing strategy is adopted: axial slicing is performed on the target tunnel model along the tunnel axis, and the position of the lining model in the target tunnel is located by axial slicing; Perform radial slicing along the radial direction of the lining model to generate layered closed slice profiles, and then reverse the slice profiles of the radial slices in the whole according to the layer sequence. (6) Unfold the slice outline onto a two-dimensional plane, construct a safe zone and identify the four sides, and adaptively generate a zigzag filling path based on the relationship between the gap width and the track width. After being clipped and bridged by the safe zone, the path is projected back onto the three-dimensional coordinate system to generate a continuous print path point series. (7) Based on the run width and slice thickness, consult the pre-calibrated empirical parameter table and match the process parameters including nozzle diameter, extrusion speed and printing speed; (8) The printing path points, process parameters and equipment control instructions are integrated into a task package, which is then sent to the printing equipment after coordinate system transformation and motion planning to start lining printing; The printing process is monitored in real time and process parameters are adjusted to achieve closed-loop control of molding quality.

2. The adaptive in-situ 3D printing method for TBM lining according to claim 1, characterized in that, The rock mass point cloud splicing process in step (3) includes: (3.1) Read the source point cloud data and target point cloud data of the surrounding rock of the tunnel. The source point cloud is a block point cloud that needs to be aligned, and the target point cloud is a fixed reference point cloud. The point cloud coordinates are normalized from millimeter level to meter level. (3.2) Remove outliers with non-finite values ​​and coordinate anomalies from the source point cloud and the target point cloud, adaptively estimate the voxel size based on the bounding box size and the set voxel upper limit, perform voxel mesh downsampling, and control the overall number of primes. (3.3) The coarse registration matrix is ​​obtained through a mode-switching coarse registration method, wherein mode one uses the SAC-IA algorithm and mode two uses the Super4PCS algorithm. (3.4) Using the coarse registration result as input, the transformation matrix is ​​optimized by the ICP fine registration algorithm to obtain the fine registration matrix; (3.5) Multiply the fine registration matrix with the coarse registration matrix to obtain the global total transformation matrix, realize the alignment and stitching of the block point cloud, and output the stitched surrounding rock point cloud.

3. The adaptive in-situ 3D printing method for TBM lining according to claim 1, characterized in that, Step (5) includes: (5.1) Input the dual STL model of the target tunnel model and the target lining model, extract the vertex of the model to generate a unique vertex set, and perform deduplication and rounding operations. (5.2) Obtain the approximate center and radius of the target tunnel at the end of the target tunnel. Translate and rotate the dual STL model to the end center. Using the target tunnel model as the reference, normalize the target lining model according to the radius r_core to obtain the length of the target tunnel after removing the influence of irregular lining. Transform each vertex of the lining triangular mesh from the Cartesian coordinate system to the cylindrical coordinate system (φ, k, r) to form a set of triangular facets represented by φ-kr triples. (5.3) Processing is performed only on lining area points whose radial coordinates are greater than the target tunnel radius. The expansion joint is adaptively calculated based on the overall angle distribution, and the points in the triangular mesh whose vertex angles are greater than the expansion joint are processed as a whole. (5.4) Dual-channel slicing strategy: Perform axial slicing along the tunnel axis on the target tunnel model to locate the position of the lining model in the target tunnel axis through axial slicing; perform radial slicing along the lining radial direction on the lining model to generate a layered closed slice profile. (5.5) Perform a layer sequence reversal on the slice outline of the radial slice, so that the slice order is changed from the direction of increasing radius to the direction of decreasing radius.

4. The adaptive in-situ 3D printing method for TBM lining according to claim 3, characterized in that, The formula for calculating the development seam is toX = max(φ) - min(φ), where φ is the angle.

5. The adaptive in-situ 3D printing method for TBM lining according to claim 3, characterized in that, The process of performing axial slicing along the tunnel axis on the target tunnel model includes: In the Cartesian coordinate system, based on the target tunnel length and slice thickness, a z-direction slice sequence is generated by slicing z-isosurfaces. For each triangular facet in the target tunnel model, its lower and upper bounds in the z-direction are pre-calculated. For a given slice layer, only triangular facets belonging to the range between the lower and upper bounds are retained as a candidate triangular facet set. The plane z = z slice Intersections with each candidate triangular facet are calculated, and intersection points are generated on the three edges to form a set of line segments. Closed loops are constructed through coordinate deduplication, connected component partitioning, and depth-first traversal, with NaN values ​​inserted between adjacent loops as separators; where z slice This represents the height of the current slice layer along the Z-axis in the Cartesian coordinate system.

6. The adaptive in-situ 3D printing method for TBM lining according to claim 3, characterized in that, The process of performing radial slicing along the radial direction of the lining model includes: In a cylindrical coordinate system, based on the target tunnel radius and slice thickness, an r-direction slice sequence is generated by slicing isosurfaces in the r direction. For each triangular facet in the lining model, its lower and upper bounds in the r direction are pre-calculated. For a given slice layer r... slice Only triangular faces belonging to the range from the lower bound to the upper bound are retained as the candidate triangular face set; using the cylindrical surface r = r slice Intersecting with each candidate triangular facet one by one, a set of line segments is generated and a closed loop is constructed. Finally, the coordinate transformation x = r is performed. slice *cos(φ), y = r slice *sin(φ) and z = k transform the sliced ​​results from cylindrical coordinates back to Cartesian coordinates; where r slice This represents the radius of the cylindrical surface corresponding to the current slice layer.

7. The adaptive in-situ 3D printing method for TBM lining according to claim 1, characterized in that, Step (6) includes: (6.1) Read the lining slice results and the width setting, transform the three-dimensional slice contour points of each layer from the Cartesian coordinate system to the cylindrical coordinate system, obtain the cylindrical coordinates (r, φ, k) of each point, and merge the Cartesian coordinate and cylindrical coordinate points to form an N×6 sequence, using NaN as the segment boundary to maintain the path topology within the layer; (6.2) Unfold the outline of each slice from the cylindrical surface to a two-dimensional plane, and set the x-coordinate on the two-dimensional plane to... plane and the vertical coordinate y plane Represented as x plane =φ*r、y plane =k, generate a uniformly dense planar polygon by resampling the arc length at equal intervals, and construct an axis-aligned circumscribed rectangle so that the long or short side of the rectangle is aligned with the axis of the target tunnel. (6.3) Perform full wrap detection on polygons at the same level to construct a safe zone; (6.4) Based on a preset angle threshold, detect the included angle of the vertices of the circumscribed rectangle after shrinkage, and identify the two opposing long sides and their gap width; (6.5) Within the safe zone, based on the relationship between the gap width between the two identified opposing long sides and the track width, three path filling scenarios are adaptively selected: when the gap width is not less than twice the track width, multiple parallel zigzag filling lines are generated with the long side as the baseline; when the gap width is between the track width and twice the track width, two long sides are retained as filling lines; when the gap width is less than the track width, only one long side is retained as a filling line. (6.6) Cut the intersection of the fill line and the safe area to remove out-of-range line segments, then bridge multiple paths in the same layer using a zigzag strategy, connect them through linear interpolation and cut them again, and finally form a continuous path in the layer using NaN as the separator. (6.7) Project the two-dimensional in-layer path back to the cylindrical coordinate system and transform it back to the Cartesian coordinate system to generate a three-dimensional printing path point list containing multi-dimensional parameters.

8. The adaptive in-situ 3D printing method for TBM lining according to claim 7, characterized in that, The process of constructing the safe zone in step (6.3) is as follows: Perform a full wrap detection on all expanded polygons in the same layer. Polygons that are completely wrapped by other polygons are marked as inner layers, and the rest are outer layers. Define the safe zone by shrinking the outer polygons inward by half the width of the track and expanding the inner polygons outward by half the width of the track.

9. The adaptive in-situ 3D printing method for TBM lining according to claim 7, characterized in that, In step (6.6), the zigzag strategy for bridging multiple paths at the same level specifically refers to: Each time, starting from the end of the current line segment, the endpoint of the next line segment closest to it is calculated based on the spatial distance and selected as the starting endpoint. This endpoint is then selected as the connection target, and a smooth transition bridging path is generated using the equal-spacing arc length linear interpolation method. During the interpolation process, interpolation points are evenly distributed based on the arc length.

10. The adaptive in-situ 3D printing method for TBM lining according to claim 1, characterized in that, Step (8) includes: (8.1) Integrate the printing path points, process parameters and equipment control instructions into a task package, and transform the path from the design coordinate system to the robot base coordinate system and perform online or offline motion planning, including acceleration and deceleration control, attitude limiting and collision or soft limit checks; send the data to the robot controller and stepper motor driver through the serial port protocol, and set pause, resume, emergency stop and breakpoint resume printing mechanism. (8.2) During printing, the extrusion pressure, pump speed or valve opening, actual robot speed and trajectory, forming layer thickness and line width are collected in real time and compared with the target values; when the layer thickness, width or position deviation exceeds the threshold, the extrusion speed or printing speed is adjusted in conjunction with the empirical parameter table or process window.

Images