A machine vision-based intelligent installation precision control method, system, and computer-readable storage medium for tunnel steel arch frames.

Abstract

This application relates to a machine vision-based intelligent installation accuracy control method, system, and computer-readable storage medium for tunnel steel arch frames. The method includes: synchronously acquiring visual images, 3D point clouds, and motion state data of the steel arch frame and performing spatiotemporal alignment; compensating for vibration noise in the point cloud to generate a denoised 3D point cloud; identifying the coordinates of the arch apex and left and right arch foot points, constructing a topological skeleton, outputting a structured feature set, and registering and comparing it with a preset design model to calculate the deviation; selecting a control mode based on the deviation and generating a hydraulic adjustment command set; constructing a hysteresis compensation model to correct command transmission delay; controlling distributed hydraulic jacks to perform displacement actions according to the corrected commands, monitoring displacement and pressure feedback data in real time and triggering a new round of data acquisition, calculating the pose deviation, and performing iterative control until the deviation is less than or equal to a threshold, at which point a steel arch frame locking signal is output. This application improves the installation efficiency and accuracy stability of tunnel steel arch frames.

Description

Technical Field

[0001] This application relates to the field of tunnel engineering construction technology, and in particular to a method and system for intelligent installation accuracy control of tunnel steel arch frames based on machine vision. Background Technology

[0002] With the continuous development of infrastructure construction in my country, long tunnel projects in transportation, water conservancy, and other fields are increasing. The safety, quality, and efficiency of tunnel construction are crucial, and the installation accuracy of the steel arch frame—a core component of the initial support—directly affects the structural stability and long-term safety of the tunnel. Traditional steel arch frame installation mainly relies on manual measurement using equipment such as total stations, followed by rough positioning and adjustment using manual machinery. This method is not only labor-intensive and inefficient, but also susceptible to measurement errors, human judgment biases, and environmental interference, making it difficult to guarantee consistent installation accuracy and potentially creating safety hazards for subsequent construction. In recent years, with the development of Building Information Modeling (BIM), industrial robots, and sensor technology, the mechanization and informatization levels of tunnel construction have improved. For example, methods such as using robotic arms for assisted handling and preliminary positioning based on single-type sensors (such as laser ranging) aim to reduce manpower and improve operational standardization. However, these improvements mostly focus on handling or single-point measurement, and a systematic automated solution has not yet been developed for the core challenge of "high-precision real-time control" during the final and most critical stage of installation.

[0003] Currently, in the practice of achieving intelligent installation and precise control of steel arch frames, the harsh environment of tunnel construction sites, with its complex interferences such as dust, vibration, and changes in light intensity, and the fact that measurement processes often involve discrete point sampling, make it difficult to quickly obtain complete three-dimensional spatial contour information of the steel arch frame. This results in insufficient judgment of the overall bending and torsional posture of the arch frame. Furthermore, in dynamic construction environments, especially near the excavation face or when heavy equipment is running, the continuous vibrations generated by construction machinery are transmitted to the measuring equipment and the steel arch frame itself, causing significant dynamic noise in the measurement data. This distorts the measurement results based on static assumptions, making it impossible to achieve the millimeter-level high-precision positioning requirements. In addition, even if the measurement is relatively accurate, the inherent response lag and valve control delay of the hydraulic system during the process from acquiring the deviation to driving the hydraulic actuator for adjustment prevent the execution of control commands from being precisely synchronized with the real-time changing posture state. This results in actual actions always being slower than expected and with distorted amplitude. This systemic delay causes oscillations or overshoots in the closed-loop control loop based on real-time perception, making it difficult to achieve stable and accurate posture correction.

[0004] Therefore, in the complex environment of tunnel construction sites, how to obtain real and stable three-dimensional geometric information of steel arch frames and, based on this, quickly and accurately adjust the position and posture of the steel arch frames is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] To enable rapid and precise position control of tunnel steel arch frames, this application provides a machine vision-based intelligent installation accuracy control method and system for tunnel steel arch frames.

[0006] Firstly, this application provides a machine vision-based intelligent installation accuracy control method for tunnel steel arch frames, employing the following technical solution:

[0007] A machine vision-based intelligent installation accuracy control method for tunnel steel arch frames, the installation accuracy control method comprising:

[0008] Multi-source sensors are synchronously triggered by a precise time protocol to collect visual image data, raw 3D point cloud data, and motion state data of the steel arch frame. A spatiotemporal mapping relationship is established based on the calibration parameters of the multi-source sensors, and a spatiotemporally aligned multimodal dataset is output.

[0009] Based on the motion state data, vibration noise compensation is performed on the original three-dimensional point cloud data to generate a denoised three-dimensional point cloud.

[0010] Based on the denoised 3D point cloud and the visual image data, the coordinates of the top vertex, left arch foot, and right arch foot of the steel arch frame are identified and extracted, the topological skeleton of the steel arch frame is constructed, and a structured feature set is output.

[0011] The structured feature set is registered and compared with the preset steel arch frame design model to calculate the pose deviation between the actual pose and the design pose of the steel arch frame.

[0012] The control mode is selected based on the magnitude of the pose deviation, and a corresponding hydraulic adjustment command set is generated.

[0013] A hysteresis compensation model is constructed based on the historical response data of the hydraulic system. The transmission delay of the hydraulic adjustment command set is corrected to generate a corrected hydraulic adjustment command set.

[0014] The distributed hydraulic jacks are controlled to perform displacement actions according to the modified hydraulic adjustment instruction set, and the actual displacement and pressure feedback data of the distributed hydraulic jacks are monitored and output in real time.

[0015] Based on the actual displacement and pressure feedback data, a new round of steel arch frame data acquisition is triggered;

[0016] Based on the newly collected steel arch frame data, the current pose deviation of the steel arch frame is recalculated. If the current pose deviation is greater than the set deviation threshold, the iteration adjustment is repeated until the error value is less than or equal to the set deviation threshold, at which point the steel arch frame locking signal is output.

[0017] By adopting the above technical solutions, the spatiotemporal fusion of multiple sensor sources overcomes the limitations of a single sensing mode in the complex environment of a tunnel, ensuring the authenticity and reliability of the original data. Motion information is used to compensate for point cloud vibration noise, and robust features are extracted by combining visual and geometric information, achieving high-precision and robust analysis of the steel arch frame's pose. A hierarchical intelligent control strategy based on the magnitude of the deviation is adopted, balancing the speed and stability of the adjustment process with the accuracy of the final positioning. An innovative hydraulic hysteresis compensation model is introduced, effectively overcoming the inherent delay of the actuator and improving command tracking accuracy. Finally, through a closed-loop iterative mechanism that triggers a new round of sensing based on execution feedback, the system possesses the ability to continuously self-optimize and verify until it reaches stringent installation accuracy standards.

[0018] Secondly, this application provides a machine vision-based intelligent installation accuracy control system for tunnel steel arch frames, employing the following technical solution:

[0019] A machine vision-based intelligent installation accuracy control system for tunnel steel arch frames, comprising:

[0020] The multi-source data acquisition module is used to synchronously trigger multi-source sensors through a precise time protocol to acquire visual image data, raw 3D point cloud data and motion state data of the steel arch frame, establish a spatiotemporal mapping relationship based on the calibration parameters of the multi-source sensors, and output a spatiotemporally aligned multimodal dataset.

[0021] The point cloud vibration suppression module is used to perform vibration noise compensation on the original three-dimensional point cloud data based on the motion state data to generate a denoised three-dimensional point cloud.

[0022] The multimodal feature extraction module is used to identify and extract the coordinates of the top vertex, left arch foot, and right arch foot of the steel arch frame based on the denoised 3D point cloud and the visual image data, construct the topological skeleton of the steel arch frame, and output a structured feature set.

[0023] The pose deviation calculation module is used to register and compare the structured feature set with the preset steel arch frame design model, and calculate the pose deviation between the actual pose and the design pose of the steel arch frame.

[0024] The decision control module is used to select the control mode according to the magnitude of the pose deviation and generate the corresponding hydraulic adjustment instruction set.

[0025] The hysteresis compensation module is used to construct a hysteresis compensation model based on the historical response data of the hydraulic system, correct the transmission delay of the hydraulic adjustment command set, and generate a corrected hydraulic adjustment command set.

[0026] The collaborative hydraulic execution module is used to control the distributed hydraulic jacks to perform displacement actions according to the modified hydraulic adjustment instruction set, and to monitor and output the actual displacement and pressure feedback data of the distributed hydraulic jacks in real time.

[0027] The iterative verification module is used to recalculate the current pose deviation of the steel arch frame based on the new round of collected steel arch frame data. If the current pose deviation is greater than the set deviation threshold, it returns to iterative adjustment until the error value is less than or equal to the set deviation threshold, at which point a steel arch frame locking signal is output.

[0028] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0029] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect.

[0030] In summary, this application includes at least one of the following beneficial technical effects: Through spatiotemporal synchronous fusion of multi-source sensors and intelligent vibration noise compensation, it achieves automatic control of tunnel steel arch frame installation with millimeter-level precision. Its uniqueness lies in constructing a real-time perception system based on "vision-point cloud-motion" multimodal data fusion and a hydraulic system hysteresis compensation mechanism, breaking through the technical bottlenecks of low efficiency and poor accuracy in traditional manual measurement and adjustment. This technical solution can adapt to vibration interference in the complex environment of tunnels and achieve precise correction of the steel arch frame's posture through iterative optimization of the control algorithm. It not only improves the installation accuracy to several times that of traditional methods but also significantly reduces construction safety risks and labor costs, providing an innovative solution for intelligent construction of tunnel engineering. Attached Figure Description

[0031] Figure 1 This is a first flowchart illustrating the intelligent installation accuracy control method for tunnel steel arch frames according to one embodiment of this application;

[0032] Figure 2 This is a schematic diagram of the second process of the intelligent installation accuracy control method for tunnel steel arch frames according to one embodiment of this application;

[0033] Figure 3 This is a schematic diagram of the third process of the intelligent installation accuracy control method for tunnel steel arch frames according to one embodiment of this application;

[0034] Figure 4 This is a schematic diagram of the fourth process of the intelligent installation accuracy control method for tunnel steel arch frames according to one embodiment of this application;

[0035] Figure 5 This is a schematic diagram of the fifth process of the intelligent installation accuracy control method for tunnel steel arch frames according to one embodiment of this application;

[0036] Figure 6 This is a schematic diagram of the sixth process of the intelligent installation accuracy control method for tunnel steel arch frames according to one embodiment of this application;

[0037] Figure 7 This is a schematic diagram of the seventh process of the intelligent installation accuracy control method for tunnel steel arch frames according to one embodiment of this application;

[0038] Figure 8 This is a schematic diagram of the eighth process of the intelligent installation accuracy control method for tunnel steel arch frames according to one embodiment of this application. Detailed Implementation

[0039] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-8 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0040] This application discloses a method for controlling the intelligent installation accuracy of tunnel steel arch frames based on machine vision.

[0041] Reference Figure 1 A machine vision-based method for intelligent installation accuracy control of tunnel steel arch frames, the specific method includes:

[0042] Step S101: Multi-source sensors are synchronously triggered through a precise time protocol to collect visual image data, raw 3D point cloud data and motion state data of the steel arch frame respectively. A spatiotemporal mapping relationship is established based on the calibration parameters of the multi-source sensors, and a spatiotemporally aligned multimodal dataset is output.

[0043] The core logic of this step lies in establishing a unified time base through a Precision Time Protocol (PTP) to address the timestamp drift problem caused by independent clock sources in multi-sensor data. In the complex electromagnetic and vibration environment of tunnel construction, the sampling periods and transmission delays of laser scanners, industrial cameras, and inertial measurement units (IMUs) vary. If only simple timestamp alignment is performed, microsecond-level errors will be amplified into centimeter-level spatial deviations during subsequent data fusion.

[0044] In this embodiment, the PTP protocol performs high-precision clock synchronization between master and slave devices via the network, ensuring that at the same absolute moment of "triggered acquisition," the RGB-D image captured by the panoramic camera, the raw 3D point cloud acquired by the laser scanner, and the angular velocity / acceleration data recorded by the IMU describe the strictly corresponding physical state of the steel arch at the same instant. This lays the foundation for time consistency in subsequent spatiotemporal alignment and is the primary prerequisite for ensuring the measurement accuracy of the entire system.

[0045] Subsequently, the multi-source data, which are synchronized in time but exist in their respective sensor coordinate systems, are uniformly transformed into a common world coordinate system to achieve spatial alignment. The principle behind this is to fuse laser point clouds, visual image pixels, and IMU data into a single multimodal dataset with a unified spatiotemporal label using pre-calibrated rigid body transformation matrices (i.e., rotation matrices and translation vectors) between the sensors. For example, a point obtained by laser scanning can be precisely correlated with the depth (D) information of the corresponding image pixel and the motion state of the steel arch at that moment (e.g., whether there is instantaneous angular displacement due to mechanical vibration) through coordinate transformation. This process solves the "data silo" problem, enabling subsequent processing to be based on a complete data entity containing geometric, textural, and dynamic information, rather than fragmented signal streams.

[0046] Step S102: Based on the motion state data, vibration noise compensation is performed on the original three-dimensional point cloud data to generate a denoised three-dimensional point cloud.

[0047] The continuous vibrations caused by heavy machinery operations within the tunnel can lead to high-frequency, micro-amplitude jitter in the laser scanner itself, resulting in "ghosting" or blurring of the acquired 3D point cloud and severely reducing the accuracy of feature point localization. This step addresses this by using real-time angular velocity data acquired by the IMU as a vibration representation. A vibration compensation model (whose core is an integral correction term based on kinematic principles) is used to invert and estimate the undesired displacement of the scanning beam caused by platform vibration. This model converts the angular motion sensed by the IMU into a coordinate correction (ΔP) for each point in the point cloud. Essentially, this performs "dynamic image stabilization" at the data level, stripping away errors caused by rigid body motion from the contaminated point cloud data to generate a denoised 3D point cloud that more closely approximates the true static form of the steel arch, thus creating conditions for subsequent high-precision feature extraction.

[0048] Step S103: Based on the denoised 3D point cloud and visual image data, identify and extract the coordinates of the top vertex, left arch foot point and right arch foot point of the steel arch frame, construct the topological skeleton of the steel arch frame, and output the structured feature set;

[0049] The principle behind this step lies in fusing the precise geometric information of the denoised 3D point cloud with the rich texture and edge information of the RGB-D visual image to perform collaborative target recognition and key point localization. Using point clouds alone may struggle to accurately segment the arch outline in complex backgrounds or under occlusion; while using images alone lacks absolute depth information. By combining the two, image recognition algorithms can initially locate the pixel regions of the "arch apex" and "left / right arch feet," then guide them to the corresponding point cloud clusters. Utilizing the 3D coordinate calculation capabilities of the point cloud, the precise 3D spatial coordinates (X, Y, Z) of these key feature points can be calculated.

[0050] Subsequently, using these feature points as nodes and the spatial connection vectors between them (such as the vector from the arch crown to the left arch foot) as edges, a topological skeleton describing the core geometry of the steel arch frame is constructed. This structured feature set represents the complete spatial pose of the steel arch frame using a very simple mathematical model (a few key points and vectors), greatly compressing the amount of data and directly serving subsequent pose comparison.

[0051] Step S104: Register and compare the structured feature set with the preset steel arch frame design model, and calculate the pose deviation between the actual pose and the design pose of the steel arch frame.

[0052] The logic of this step involves registering and comparing the abstract model of the steel arch frame (structured feature set) measured on-site with the ideal digital design model (pre-set steel arch frame BIM design model). The registration process typically employs Iterative Closest Point (ICP) or its variants. The principle is to find an optimal spatial transformation (rotation and translation) that minimizes the overall distance between the measured feature point set and the corresponding point set in the design model. The calculated transformation parameters intuitively express which axis the current steel arch frame needs to rotate around by how many angles and translate in which direction to coincide with the design pose; this amount of rotation and translation is the pose deviation.

[0053] Step S105: Select the control mode according to the magnitude of the pose deviation and generate the corresponding hydraulic adjustment command set;

[0054] This step embodies the system's intelligent decision-making logic, which adaptively selects the optimal control strategy based on the severity (magnitude) of the deviation, rather than adopting a single control mode.

[0055] In some embodiments, for large deviations (e.g., >30mm), the system is in the "coarse adjustment" stage, where the primary goal is to quickly reduce the error. Therefore, open-loop or high-gain fast control commands are generated to pursue efficiency. For medium deviations (e.g., 10-30mm), the system enters the "fine adjustment" stage, employing classic PID position closed-loop control. Commands are dynamically adjusted based on the real-time error ratio, integral, and derivative to achieve smooth convergence. For small deviations (≤10mm), the system is in the "fine-tuning" stage. At this stage, simple PID control may get stuck in local optima or cause oscillations. Therefore, the Particle Swarm Optimization (PSO) algorithm is invoked to perform a swarm intelligence search within the possible fine-tuning command space to find the optimal command combination that minimizes the global final pose error. This three-level progressive strategy balances adjustment speed, stability, and final accuracy.

[0056] Step S106: Construct a hysteresis compensation model based on the historical response data of the hydraulic system, correct the transmission delay of the hydraulic adjustment command set, and generate the corrected hydraulic adjustment command set.

[0057] Hydraulic systems inherently suffer from transmission delays and response lags, primarily due to the current-to-flow conversion of electro-hydraulic proportional valves, the compressibility of hydraulic fluid, and the time consumed in transmitting pipeline pressure. If the calculated ideal adjustment command is directly sent to the hydraulic system, the execution result will always be delayed and distorted.

[0058] In this embodiment, a feedforward compensation mechanism is introduced, utilizing historical data of the hydraulic system response to construct a hysteresis compensation model (such as a Smith predictor) capable of predicting its dynamic behavior. This model simulates the delay characteristics of the hydraulic system within the digital controller. Before a new adjustment command is issued, its execution process and delay effect are "pre-rehearsed" within the model, and the original command is pre-corrected accordingly. The corrected command is then sent to the physical system, which can offset most of the hysteresis effect, enabling the jack's displacement action to track the controller's requirements more promptly and accurately. This is a crucial step in achieving high-precision execution.

[0059] Step S107: Control the distributed hydraulic jacks to perform displacement actions according to the modified hydraulic adjustment instruction set, and monitor and output the actual displacement and pressure feedback data of the distributed hydraulic jacks in real time.

[0060] Among them, the distributed electro-hydraulic proportional valve, as the core component of the actuator, allows for independent and precise flow and pressure control of multiple hydraulic jacks, thereby enabling multi-point coordinated and differentiated lifting or translation adjustment of the steel arch frame.

[0061] In this embodiment, by receiving a hysteresis-compensated hydraulic adjustment command (typically a target displacement or speed), the proportional valve controls the flow and direction of hydraulic oil by adjusting the valve core opening, thereby driving the jack piston to generate a precise displacement stroke. Simultaneously, a real-time monitoring system continuously collects the actual displacement and oil pressure feedback data of each jack using high-precision displacement and pressure sensors. This closed-loop monitoring mechanism not only provides real-time feedback for the control algorithm but also detects abnormal situations during execution, such as overload and jamming, ensuring the safety and reliability of the adjustment process. The high-precision acquisition of feedback data provides an accurate performance evaluation basis for subsequent iterative verification.

[0062] Step S108: Based on the actual displacement and pressure feedback data, trigger a new round of steel arch frame data acquisition;

[0063] Once the hydraulic system completes a round of adjustments and outputs stable actual displacement and pressure feedback data, the system determines that the spatial state of the steel arch has changed. At this point, this feedback data automatically serves as a trigger signal, waking up and initiating the first round of synchronous data acquisition from the multi-source sensors. This avoids the potential gaps or redundant acquisitions that may occur with fixed-cycle acquisition, ensuring that each iteration of control is based on the latest and most accurate state information of the steel arch, thus forming a coherent, event-driven automated sequence for the entire control process.

[0064] Step S109: Based on the new round of collected steel arch frame data, recalculate the current pose deviation of the steel arch frame. If the current pose deviation is greater than the set deviation threshold, return to iterative adjustment until the error value is less than or equal to the set deviation threshold and output the steel arch frame locking signal.

[0065] Specifically, based on the data obtained from the latest round of acquisition and processing, the current pose error value of the steel arch is calculated and compared with a preset threshold representing installation accuracy (e.g., 3mm). This is a strict judgment point: if the error value is greater than the threshold, it indicates that the installation standard has not been met, and the system automatically returns to iterative adjustment, that is, it goes through the process of feature extraction, deviation calculation, instruction generation and compensation execution again for further fine-tuning; if the error value is less than or equal to the threshold, the installation accuracy is deemed qualified, and the system outputs a steel arch locking signal, allowing final physical fixing (such as welding or bolt tightening). This closed-loop verification and iterative mechanism based on actual measurement results enforces that the final output result must meet the preset accuracy requirements from the process perspective, giving the system the ability to self-correct and continuously approach the target.

[0066] In the above embodiments, the spatiotemporal fusion of multiple sensor sources overcomes the limitations of a single sensing mode in the complex environment of a tunnel, ensuring the authenticity and reliability of the original data. Motion information is used to compensate for point cloud vibration noise, and robust features are extracted by combining visual and geometric information, achieving high-precision and robust analysis of the steel arch frame's pose. A hierarchical intelligent control strategy based on the magnitude of deviation is adopted, balancing the speed and stability of the adjustment process with the accuracy of the final positioning. An innovative hydraulic hysteresis compensation model is introduced, effectively overcoming the inherent delay of the actuator and improving command tracking accuracy. Finally, through a closed-loop iterative mechanism that triggers a new round of sensing based on execution feedback, the system possesses the ability to continuously self-optimize and verify until it reaches strict installation accuracy standards. This technical solution transforms the traditional steel arch frame installation process, which relies on manual experience, is labor-intensive, and has difficulty guaranteeing accuracy, into an automated, intelligent, and high-precision industrial control process, significantly improving the quality, efficiency, and safety of tunnel construction.

[0067] Reference Figure 2As one implementation of step S102, the step of performing vibration noise compensation on the original three-dimensional point cloud data based on motion state data to generate a denoised three-dimensional point cloud includes:

[0068] Step S201: Extract the three-axis angular velocity components and timestamp sequence from the motion state data to generate an angular velocity-time relationship matrix;

[0069] The underlying logic of this step lies in performing preliminary structural decoupling and formatting processing on the raw, multi-dimensional motion state data collected by the inertial measurement unit (IMU). The raw IMU data is a continuous data stream containing various types of information such as acceleration, angular velocity, and magnetic field, and the sampling rates of each sensor may differ.

[0070] In this embodiment, the three-axis angular velocity components directly related to the device's angular motion (i.e., rotation), along with their precise time stamps, are extracted and organized into an angular velocity-time relationship matrix. Each row of this matrix represents a specific sampling moment, and its columns correspond to the angular velocity values ​​of the device around its three coordinate axes (X, Y, Z) at that moment. The essence of constructing this matrix is ​​to provide structured, time-axis-clear input data for the subsequent compensation model. Without this structured matrix, the subsequent vibration compensation model cannot accurately correlate specific angular velocity values ​​with the instantaneous point cloud sampling, thus failing to accurately invert the distortion morphology caused by vibration on the point cloud. This is not only a preliminary step in data processing but also establishes a bridge between "motion" and "point cloud distortion."

[0071] Step S202: Calculate the point cloud displacement compensation based on the tunnel environment damping coefficient and the angular velocity-time relationship matrix;

[0072] The logical principle behind this step is to convert the angular motion sensed by the sensor, combined with the characteristics of the engineering environment, into a compensation vector that can correct the point cloud coordinates. The angular velocity generated by the equipment during vibration will cause unexpected, high-frequency, minute changes in the direction and position of the laser beam at the instant of acquiring the 3D point cloud. This causes points that should be stationary to appear displaced or to produce "ghosting" in the point cloud.

[0073] In this embodiment of the application, the formula for calculating the point cloud displacement compensation is as follows:

[0074] ;

[0075] In the above formula, k is the tunnel vibration damping coefficient (range 0.25-0.75), S(ω) is the antisymmetric matrix of angular velocity, and P0 is the initial coordinate of the point cloud.

[0076] Specifically, the instantaneous angular velocity of the device over time can be obtained through the angular velocity-time relationship matrix during the point cloud scanning cycle. This angular velocity is then converted into an antisymmetric matrix and calculated with the initial coordinates of the point cloud (which can be considered an ideal vibration-free scanning vector), and integrated over the time dimension. The physical meaning of this calculation is the cumulative displacement effect caused by the continuous rotation of the device, resulting in the laser scanning point shifting from the ideal position to the actual sampling position. The introduced tunnel environment damping coefficient is a key engineering empirical parameter. It quantifies the vibration attenuation characteristics inside the tunnel, determined by the surrounding rock properties and support structure, and is used to correct the pure kinematic model, making it more closely resemble the actual physical scenario. This damping coefficient makes the compensation model no longer an ideal laboratory model, but rather adaptable to the vibration response differences in specific engineering environments (such as soft rock tunnels and hard rock tunnels), enhancing the targeting and accuracy of the compensation.

[0077] Step S203: Apply the point cloud displacement compensation amount to the original three-dimensional point cloud data to generate a vibration-compensated intermediate point cloud.

[0078] The logical principle behind this step is to perform a reverse geometric transformation, aiming to initially recover the true geometry of the target object from the contaminated measurement data. The point cloud displacement compensation describes how much the coordinates of each point in the point cloud are offset relative to its true position due to equipment angular vibration. Subtracting (or adding inversely) the corresponding point cloud displacement compensation from the coordinates of each point in the original 3D point cloud data is equivalent to mathematically canceling out the geometric distortion caused by the equipment's rotational vibration on the measurement results. This operation is similar to the deconvolution process used to "deblur" a blurred image in image processing, but its object is three-dimensional spatial coordinates.

[0079] After this step, the output vibration-compensated intermediate point cloud has initially eliminated the rotational distortion effect caused by angular velocity. The overall shape and topology of the point cloud are closer to the true static form of the steel arch than the original point cloud. However, the point cloud at this time may still contain the overall offset caused by linear vibration (translational vibration).

[0080] Step S204: Extract the triaxial acceleration components from the motion state data and calculate the linear vibration offset of the equipment;

[0081] The logical principle behind this step is to handle a different type of vibration than rotational vibration—linear vibration, which refers to the translational motion of the equipment in space, including forward / backward, left / right, and up / down movements. Its core is also motion estimation using IMU sensor data. The triaxial acceleration components extracted from the motion data record the instantaneous linear acceleration changes of the equipment along each coordinate axis.

[0082] In one embodiment of this application, the specific formula for calculating the linear vibration offset of the device is as follows:

[0083] ;

[0084] In the above formula, a x a y a z These are the three-axis acceleration components.

[0085] Specifically, according to the principles of physics, integrating acceleration once yields velocity, and integrating it twice yields displacement. Therefore, by integrating separate, time-synchronized acceleration data, the linear vibration offset of the device itself within the point cloud scanning time window can be calculated. This offset is a translation vector describing the distance the scanning device (along with its laser scanner) translates in three-dimensional space. Unlike point cloud distortion caused by angular velocity, linear vibration causes the entire point cloud model to translate as a rigid whole in the spatial coordinate system. If this compensation is ignored, even if rotational vibration is perfectly corrected, the position of the entire steel arch point cloud model will still deviate from its true position, which is unacceptable for scenarios requiring millimeter-level installation accuracy. Therefore, the offset calculated in this step is a key parameter for ultimately correcting the global positional accuracy of the point cloud.

[0086] Step S205: The linear vibration offset is superimposed onto the vibration compensation intermediate point cloud to output a denoised 3D point cloud.

[0087] Specifically, its core is to use a rigid body transformation model to synthesize the two compensation effects calculated separately, and finally output a high-fidelity point cloud that corrects both internal geometric distortion and global position deviation.

[0088] Specifically, firstly, the coordinates of each point in the vibration compensation intermediate point cloud (with rotational distortion eliminated) are confirmed with the calculated rotational compensation details (already included in the intermediate point cloud, but more precisely expressed by the rotation matrix R). Then, the calculated linear vibration offset, representing the overall translation of the equipment, is applied as a unified translation vector to all points in this point cloud. This process is typically mathematically expressed as... It accurately describes the entire process by which a point cloud, from a measurement state disturbed by vibration, undergoes one rotational transformation (compensating for rotational vibration) and one translational transformation (compensating for translational vibration), to recover its true position and orientation. Where P... comp The compensated point cloud coordinates, i.e., the final coordinates of each point in the denoised 3D point cloud, represent the true spatial position of the steel arch; R is the equipment attitude rotation matrix, obtained by converting quaternions from the motion state data; P rawThe original 3D point cloud data is the coordinates of the measured point cloud affected by vibration. It is the unprocessed data directly collected by the laser scanner. ΔP is the point cloud displacement compensation amount, which is the point cloud coordinate offset caused by the angular vibration of the equipment. ΔL is the linear vibration offset amount, which is the overall point cloud translation caused by the linear vibration (translational vibration) of the equipment.

[0089] Ultimately, the output denoised 3D point cloud, with its internal point relative positions and overall absolute position, has largely eliminated the measurement errors introduced by equipment vibration, providing a high-quality and highly reliable 3D data foundation for subsequent feature recognition and pose comparison.

[0090] In the above implementation, the composite motion state data is first structurally separated, and rotational and translational vibrations characterized by angular velocity and acceleration are processed separately. Combined with the environmental damping coefficient unique to tunnel engineering, a compensation model that more closely resembles the actual physical process is constructed. Its core logic lies in using the inertial measurement unit's precise perception of its own motion state to inversely calculate the geometric impact of this motion on the laser scanning results, and then "counteracting" this impact by performing a reverse rigid body transformation on the point cloud. This technical solution proactively corrects the data from its "origins," rather than through traditional post-processing filtering. This allows for accurate reconstruction of the target object's true three-dimensional structure against a strong vibration background, improving the recognition accuracy of key feature points of the subsequent steel arch frame (such as the arch crown and arch feet), and providing a solid and reliable spatial perception foundation for the control and decision-making of the entire intelligent installation system.

[0091] Reference Figure 3 As one implementation of step S103, the steps of identifying and extracting the coordinates of the top vertex, left arch foot, and right arch foot of the steel arch frame based on denoised 3D point cloud and visual image data, constructing the topological skeleton of the steel arch frame, and outputting a structured feature set include:

[0092] Step S301: Perform spatial meshing on the denoised 3D point cloud and extract the outline edge point set of the steel arch frame;

[0093] The underlying logic of this step lies in structuring and initial feature screening of high-density 3D point cloud data, providing an efficient data foundation for subsequent accurate identification. While vibration-compensated denoised 3D point clouds eliminate motion artifacts, they remain an unordered collection of tens or even millions of discrete points. Directly searching for specific features within these points is inefficient and susceptible to interference from internal points. Spatial mesh generation is a data management method that discretizes 3D spatial rules into a series of small volumetric units (voxels). Its core logic is to discretize continuous spatial coordinates and treat point clouds falling within the same voxel as a whole.

[0094] Next, the distribution characteristics of the point cloud are analyzed within each grid. For example, by calculating the dispersion or distribution entropy of the point cloud normal vector, it can be determined whether the region is a smooth surface interior or a contour edge with drastic changes in normal direction. The edges of the steel arch, i.e., the boundaries of its steel section, are regions where the point cloud density and normal direction change abruptly. By extracting points within these grids with significant characteristics, a set of contour edge points representing the spatial contour of the steel arch is obtained. This step essentially delineates the target from the scattered panoramic point cloud, greatly compressing the amount of data and focusing on the edge information that best characterizes the object's geometry, providing a clear and concise input for subsequent precise positioning.

[0095] Step S302: Perform pixel-space coordinate mapping between the visual image data and the contour edge point set to identify candidate points in the arch foot region;

[0096] The core logic of this step lies in achieving cross-modal fusion of three-dimensional geometric information and two-dimensional texture information, utilizing the rich semantics of visual images to assist and verify the geometric judgment of three-dimensional point clouds. The contour edge point set provides the possible spatial range of the arch, but it is difficult to distinguish the arch from the edges of other similar shapes in the background (such as temporary supports) based solely on the point cloud.

[0097] At this point, by introducing visual image data and utilizing pre-calibrated camera intrinsic and extrinsic parameters (i.e., the camera calibration matrix), a precise projection relationship between each pixel in the image and a point in three-dimensional space can be established. By back-projecting the set of three-dimensional contour edge points onto the two-dimensional image plane, a corresponding mask region depicting the possible edges of the steel arch can be generated on the image. Within this region, image processing algorithms (such as feature point detection, edge detection, or template matching) are used to find image areas that highly match the typical visual features of the "arch foot" (such as specific shapes, weld seam textures, and bolt hole patterns). Then, these candidate arch foot regions identified in the image are used again to back-calculate back into three-dimensional space using the pixel-space coordinate mapping relationship, resulting in a series of candidate arch foot regions located near the arch foot. This step further narrows the search range of the arch foot based on the three-dimensional contour through image semantic information and provides richer verification criteria.

[0098] Step S303: Fit the local surface based on the candidate points in the arch foot area, calculate the curvature extremum points, and output the coordinates of the left arch foot point and the right arch foot point.

[0099] The logical principle behind this step is to precisely pinpoint the sub-millimeter coordinates of the arch foot point from the candidate region based on purely geometric differential properties. While the candidate points in the arch foot region provide an approximate spatial location, the precise location of the arch foot is the point where the curvature of the steel section is at its maximum.

[0100] In this embodiment, to achieve sub-millimeter-level positioning, a moving least squares method can be used to fit a smooth local surface within the neighborhood of a candidate point. The advantage of this method is its ability to reconstruct surfaces with high accuracy from discrete, non-uniformly distributed point clouds, without relying on a fixed mesh. After obtaining the fitted surface, the Gaussian curvature or average curvature of each point on the surface is calculated. Geometrically, curvature characterizes the degree of bending of the surface at that point. For arches constructed of I-beams or U-shaped steel sections, there are obvious curvature extrema at the corners. By setting a reasonable curvature threshold, points with curvature exceeding this threshold (e.g., 0.25 mm) are selected. −1 These points are the sharpest and most prominent feature points of the arch foot in terms of geometry. By clustering and filtering these extreme points, the coordinates of the left and right arch foot points can be output separately.

[0101] Step S304: Search for the extreme points of the arch height in the denoised 3D point cloud, and verify the coordinates of the arch vertex by combining the texture features of the visual image data.

[0102] This step employs a dual logic of "geometric initial screening + texture verification" to uniquely determine the arch apex. First, from the entire denoised 3D point cloud, the global maximum point is directly searched based on the Z coordinate (usually representing height), which is the most likely extreme point of the arch height. This is an efficient and direct geometric localization method. However, in complex tunnel scenes, the highest point may not be the arch (for example, it could be a suspended cable or attachment).

[0103] Therefore, a verification mechanism must be introduced. The logic of verification based on visual image texture features is as follows: obtain the image patch corresponding to the candidate highest point in the image, extract its texture, color, gradient direction histogram, and other features, and compare them with the pre-learned or set standard texture features of the arch vertices (such as specific connecting plate patterns, paint marks, or structural textures). If the matching degree of the texture features exceeds the preset confidence threshold, the geometric highest point is confirmed as the true arch vertex coordinates; otherwise, verification may continue at the second highest point. This step corrects and confirms the purely geometric extreme value judgment through visual semantic information, ensuring the accuracy of key point recognition in complex real-world environments and avoiding misidentification caused by scene interference.

[0104] Step S305: Connect the top apex of the arch, the left arch foot point, and the right arch foot point to construct a topological skeleton of the steel arch frame containing spatial vector relationships, and output a structured feature set containing coordinates and vector topology.

[0105] Specifically, while directly using the precise coordinates of the three key points can represent their positions, pose analysis is complex and highly dependent on the coordinate system. However, constructing spatial vector relationships—for example, calculating vector V1 pointing from the arch apex to the left arch foot, vector V2 pointing to the right arch foot, and the angle between these two vectors—extracts the essential information about the steel arch's morphology. This vector topology includes the arch's span (represented by vector magnitude or distance between arch feet), arch height (related to vector direction), and symmetry (represented by the vector angle).

[0106] It should be noted that these relationships are invariants or covariants under rigid body transformations. That is, regardless of whether the steel arch frame translates or rotates in space, the relative vector relationships (module length, included angle) or their transformation laws between its key points are determined. The output structured feature set not only has a very small data volume, but also extracts the original and redundant point cloud and image information into a highly abstract and digital geometric description that best represents the spatial pose of the steel arch frame. This provides the most direct and effective input for subsequent rapid and robust registration and deviation calculation with the BIM design model.

[0107] In the above implementation, starting from a massive, disordered 3D point cloud, the target contour is quickly focused through spatial partitioning. Then, synchronously acquired visual image information is deeply fused, and the rich semantics of the images are used to accurately locate candidate regions for key points under the constraints of the 3D contour. Subsequently, a precise geometric differential method, moving least squares surface fitting and curvature extremum analysis, is used to extract arch foot feature points from the candidate regions with meticulous accuracy. Simultaneously, a strategy combining preliminary geometric extremum judgment and visual texture verification is employed for the arch apex to ensure the uniqueness and accuracy of key point identification. Finally, the discrete key point coordinates are transformed into spatial vectors and angular relationships describing the overall structure, forming a highly abstract structured feature description.

[0108] Based on this, the technical solution effectively overcomes the inherent defects of low recognition rate and poor anti-interference ability of single sensors (pure point cloud or pure vision) in dim and dusty tunnel environments. Through multiple verification and complementarity of cross-modal information, it realizes high-precision and high-robust automated recognition of key features such as the top apex and foot points of the steel arch frame. The results are expressed as a compact vector topology form most suitable for subsequent pose calculation, laying a crucial perception foundation for the reliable operation of the entire intelligent installation accuracy control system for the steel arch frame.

[0109] Reference Figure 4 As one implementation of step S104, the step of registering and comparing the structured feature set with the preset steel arch frame design model, and calculating the pose deviation between the actual pose and the designed pose of the steel arch frame includes:

[0110] Step S401: Based on the coordinates of the arch apex, left arch foot, and right arch foot in the structured feature set, establish a local coordinate system for the steel arch frame;

[0111] Specifically, for steel arches measured on-site and in arbitrary, unknown spatial orientations, a unique and stable reference frame, namely a local coordinate system, is constructed to represent their own geometric orientation. The structured feature set provides the three-dimensional coordinates of three non-collinear key points: the arch apex and the left and right arch feet. Their configuration in space itself defines a minimum and most stable geometric reference for the steel arch.

[0112] In this embodiment, a typical method for establishing a local coordinate system is as follows: the arch apex is taken as the origin (O); the vector pointing from the arch apex to the left arch foot is taken as a reference axis (e.g., the X-axis); then, using the vector pointing from the arch apex to the right arch foot, another reference axis perpendicular to the X-axis (e.g., the Y-axis) is generated through vector cross product operation; finally, a third axis (Z-axis) perpendicular to both the X-axis and Y-axis is obtained through cross product, thus forming a complete right-hand rule Cartesian coordinate system. The essence of this local coordinate system is to "abstract" and "independently" the steel arch frame from the chaotic, absolute world measurement coordinate system, with its coordinate axes closely bound to the physical structure of the steel arch frame (e.g., the orientation of the arch ribs, the normal to the arch plane). Regardless of whether the steel arch frame is tilted, rotated, or translated when measured in the tunnel, this coordinate system, dynamically established based on its three characteristic points, is the most natural and direct mathematical tool for describing its current posture, laying the foundation for subsequent accurate comparison with the ideal model.

[0113] Step S402: Extract the coordinates of the reference points in the preset steel arch frame design model and construct the design coordinate system;

[0114] The logical principle of this step corresponds to the first step: on the side of the digitized ideal model, a reference frame, namely the design coordinate system, is established that is completely isomorphic and corresponds to the local coordinate system of the steel arch frame measured on site. The preset steel arch frame design model (usually a BIM model) contains the ideal design dimensions and spatial orientation of the steel arch frame. From this, the coordinates of the reference points that completely correspond to the three feature points identified on site (arch apex, left arch foot, and right arch foot) are extracted.

[0115] Specifically, using the exact same geometric rules as in the first step (e.g., using the designed arch crown as the origin and the vector pointing to the designed left arch foot as the X-axis), a design coordinate system is constructed on the ideal model. This design coordinate system represents the theoretically correct position and orientation of the steel arch frame after installation. The purpose of constructing the design coordinate system is to provide a precise and error-free "target" or "template" for the local coordinate system measured on-site. Only when the two coordinate systems are established on the exact same geometric definition rules can subsequent calculations of the spatial transformation relationship between them have a clear and consistent physical meaning and mathematical basis.

[0116] Step S403: Calculate the rigid body transformation matrix from the local coordinate system to the design coordinate system of the steel arch frame;

[0117] This step is the mathematical core of the pose deviation calculation. Its logical principle is to solve for an optimal rigid body transformation matrix, which can transform (i.e. rotate and translate) the steel arch frame measured on site from its own local coordinate system to the ideal design coordinate system with the minimum overall error.

[0118] In this embodiment, the rigid body transformation matrix is ​​a 4x4 homogeneous transformation matrix that encapsulates a three-dimensional rotation matrix and a three-dimensional translation vector. It describes the motion of an object in space through translation and rotation without altering its shape and size. A classic method for solving this matrix is ​​to find a transformation T such that when this transformation is applied to the coordinates of three measured feature points, the sum of the squared distances between the transformed points and the corresponding three reference points in the design model is minimized. This process is mathematically known as point set registration or absolute orientation.

[0119] Specifically, the rigid body transformation matrix T is solved by minimizing the objective function:

[0120] ;

[0121] In the above formula, Pi is the set of coordinates of the arch crown, the left arch foot, and the right arch foot. i 实际 P is the set of coordinates of key points of the steel arch frame obtained from actual measurements. i 设计 Let T be the set of reference point coordinates in the pre-defined steel arch frame design model; T is the rigid body transformation matrix. The calculated transformation matrix T precisely describes "how much to rotate the currently installed steel arch frame around an axis and how much to translate it along a direction in order to adjust it to the design pose." Therefore, this matrix itself implicitly contains the main information of the pose deviation and is a key bridge for quantifying abstract deviations.

[0122] Step S404: Map the spatial vector relationships in the structured feature set to the design coordinate system using a rigid body transformation matrix;

[0123] In this process, the unified transformation relationship obtained in the previous step is used to transfer the spatial vector relationships reflecting the actual shape of the steel arch frame (such as the vectors from the arch top to the two arch feet and their included angles) from its measured local coordinate system to the ideal design coordinate system for unified expression and comparison. Directly using the original vectors defined in their respective coordinate systems for comparison is invalid because their numerical magnitude and direction depend on their respective coordinate systems.

[0124] In this embodiment, by applying the rigid body transformation matrix to the start and end coordinates of these spatial vectors, a new coordinate representation of these vectors in the design coordinate system can be obtained. This process is called coordinate mapping or coordinate transformation. The mapped vectors are in the same design coordinate system as the theoretical vectors directly extracted from the design model. This alignment enables fair and direct comparisons, whether comparing the length (module length) of the vectors to evaluate dimensional errors or displacements, or comparing the angles between the vectors to evaluate angular errors of torsion or tilt.

[0125] Step S405: Compare the coordinates of the mapped arch apex, left arch foot point and right arch foot point with the coordinates of the corresponding points in the preset steel arch frame design model to generate a six-degree-of-freedom pose deviation quantity containing displacement deviation components and angle deviation components.

[0126] Within a unified design coordinate system, the differences between the transformed measured coordinates of key points and the corresponding reference points in the design model are compared point-by-point along each coordinate axis. Specifically, the coordinate differences (Δx, Δy, Δz) of the arch apex in the X, Y, and Z directions are calculated, as are the coordinate differences of the left and right arch foot points. These coordinate differences visually reflect the deviation of the currently installed steel arch frame's key components from their design positions in three orthogonal spatial directions. This point-to-point comparison provides the most basic, linear deviation data, serving as primary information for calculating overall translation and assessing installation accuracy. It decomposes complex spatial pose differences into independently measurable and understandable linear error components in three orthogonal directions.

[0127] Next, the discrete point coordinate deviations and vector relationship deviations are synthesized and refined into a six-degree-of-freedom pose deviation quantity with clear physical meaning that describes the complete spatial pose information of an object. The complete pose of an object in three-dimensional space requires six parameters to describe it: three translational degrees of freedom and three rotational degrees of freedom.

[0128] The displacement deviation is usually calculated by integrating the coordinate differences of the points. For example, the average magnitude of the displacement vectors of the three key points is calculated, or the root mean square value of their displacements is calculated, to obtain a scalar value reflecting the magnitude of the overall translational deviation. The specific formula is as follows:

[0129] ;

[0130] In the above formula, ΔD is the displacement deviation, and Δx, Δy, and Δz represent the coordinate differences in the x, y, and z directions between the actual and designed poses of the key points of the steel arch frame, respectively.

[0131] Angular deviations are obtained by comparing the mapped spatial vector relationships with the vector relationships in the design model. For example, the angle between the measured "arch top - left arch foot" vector and the corresponding vector in the design is calculated; this angle reflects the rotational deviation of the steel arch frame around a certain axis. Similarly, the angles between other vectors or the angles between plane normal vectors can be used to calculate the deviations in other rotational degrees of freedom. The specific calculation formula is as follows:

[0132] ;

[0133] In the above formula, Indicates the angular deviation. This represents the spatial vector of the steel arch frame obtained from actual measurements. This represents the corresponding spatial vector in the design model.

[0134] Ultimately, the generated six-degree-of-freedom pose deviation, in a compact and complete form (e.g., a three-dimensional translation vector and a three-dimensional rotation angle, or a transformation matrix), quantifies the difference between the current installation state and the target design state in all possible motion directions, providing a comprehensive and accurate basis for subsequent intelligent control.

[0135] Step S406: Based on the similarity matching between the mapped spatial vector relationship and the vector topology of the design model, the angular component error in the six-degree-of-freedom pose deviation is corrected to obtain the final pose deviation.

[0136] Vector topology refers to the relative relationships between vectors formed by key points, such as the angles and length ratios between vectors. These are inherent geometric properties of an object and are relatively insensitive to measurement noise and local errors. Similarity matching calculates the degree of agreement between the measured vector topology and the designed vector topology, for example, comparing the difference between the measured angle of a double arch foot and the designed angle. Its core function is to perform cross-validation and error diagnosis: if the rotation angle (angle deviation component) calculated directly from the point coordinates is significantly inconsistent with the angle change calculated independently through vector topology (e.g., exceeding the vector angle tolerance), it indicates that the previous registration solution based on three points may have produced unreliable angle results due to gross errors or local deformations in the measurement of a certain point.

[0137] In this embodiment, when the similarity matching check finds that the inconsistency of the vector topology angle exceeds an acceptable tolerance threshold (e.g., 0.5 degrees), the system determines that the angle deviation component directly calculated from the coordinate points may be distorted due to local errors. In this case, the system does not completely discard the preliminary result, but corrects it based on more stable vector topology information. The correction logic can be to scale the initially calculated angle deviation according to the ratio between the measured vector angle and the designed angle, or it can use the angle change directly calculated from the vector topology to replace the unreliable preliminary angle result.

[0138] The pose deviation corrected in this step has a more reliable angular component because it has passed a secondary verification based on the overall geometric consistency of the object. This effectively suppresses erroneous angle estimations caused by individual feature point recognition errors or local point cloud noise, ensuring that the final adjustment commands output to the hydraulic control system are accurate and reliable in direction. Furthermore, if the deviation does not exceed the tolerance threshold, the original deviation is retained.

[0139] In the above implementation, based on three corresponding feature points of the field measurement and the design model, completely isomorphic local and design coordinate systems are established respectively, laying a rigorous mathematical foundation for all subsequent comparisons. Then, by solving the optimal rigid body transformation matrix, the accurate mapping of actual measurement data to the ideal design space is achieved. This technical solution not only performs direct point coordinate comparison to calculate the basic displacement deviation, but also introduces and utilizes spatial vector topological relationships to independently calculate and verify angular deviations, forming a dual verification mechanism that combines coordinate comparison and vector verification.

[0140] Understandably, when inconsistencies arise between the two exceeding the tolerance range, the system can intelligently correct the initial angle deviation based on more robust vector topology information. This improves the robustness and overall reliability of the pose deviation calculation results in complex tunnel construction sites (where local occlusion, point cloud noise, or recognition errors may exist). The final output of the six-degree-of-freedom pose deviation comprehensively, accurately, and reliably quantifies the installation error, providing data support for subsequent graded and precise control. This fundamentally overcomes the technical shortcomings of traditional methods that rely on single-point measurements or ignore overall geometric consistency, leading to inaccurate control commands.

[0141] Reference Figure 5 As one implementation of step S105, the step of selecting a control mode based on the magnitude of the pose deviation and generating a corresponding hydraulic adjustment command set includes:

[0142] Step S501: Decompose the pose deviation into displacement deviation components and angle deviation components, and calculate the projection component of the displacement deviation component in the axis of the distributed hydraulic jack.

[0143] The pose deviation is typically a composite data volume containing translation and rotation information, such as a transformation matrix or a set of Euler angles and translation vectors. The linear displacement deviation component describes the translational error of the overall centroid or feature point of the steel arch frame along the X, Y, and Z directions in a Cartesian coordinate system; essentially, it describes a positional deviation. The angular deviation component describes the rotational error of the steel arch frame about one or more of its own axes; essentially, it describes a deviation in attitude or orientation.

[0144] Next, the abstract deviation describing the overall posture of the steel arch frame, decoupled from the previous process, is mapped and assigned to physical actuators—hydraulic jacks—with fixed installation positions and a single direction of motion. Each jack's extension rod typically generates thrust or pull along its own axis (i.e., the direction of its piston rod's centerline), a direction determined by its physical installation posture and represented by a unit direction vector. The mathematical essence of calculating the projected components is to project the displacement vector describing the overall translation of the steel arch frame onto the axial unit vector of each jack, thereby obtaining "how much displacement along its own axis the jack needs to contribute to correct this overall translation."

[0145] Step S502: Is the maximum projected component greater than the first threshold? If yes, proceed to step S503; otherwise, proceed to step S504.

[0146] Step S503: Select speed control mode to generate basic displacement command;

[0147] This step implements the first level of decision-making in a hierarchical control strategy, aiming to handle large deviations. Its core control objective is rapid convergence, prioritizing response speed. A first threshold (e.g., 30mm) sets a limit; when the maximum projected component that any jack needs to adjust exceeds this value, the system determines that the current installation deviation is large and is in the "coarse adjustment" stage.

[0148] At this point, if a fine but slow adjustment algorithm (such as PID closed-loop) is used, the convergence time will be very long. Therefore, the system switches to speed control mode, which is essentially an open-loop or high-proportion-gain feedforward control. The generated displacement step command is a single displacement command with a large amplitude, directly calculated based on the current deviation (usually multiplied by a coefficient k1 slightly greater than 1 to overcome static friction and inertia, achieving rapid overshoot approximation). This command does not consider minor fluctuations during execution and directly commands the jack to move a large preset stroke at a relatively fast speed, thereby pulling the deviation into the next finely adjustable range in the shortest possible time. This is the key step in improving efficiency throughout the entire adjustment process.

[0149] Specifically, the displacement step command quantity for the speed control mode is: In the above formula, k1 is a coefficient with a value range of 1.2 to 1.5; ΔD is the projection component.

[0150] Step S504: Is the maximum projected component not greater than the first threshold but greater than the second threshold? If yes, proceed to step S505; otherwise, proceed to step S506.

[0151] Step S505: Select the position closed-loop mode to generate the basic displacement command;

[0152] This step is the second-level decision in the hierarchical control strategy, applicable to the precise positioning stage with moderate deviation. Its core control objective is "smooth convergence," ensuring accuracy while also considering a certain speed. When the maximum projected component falls between the first threshold (e.g., 30mm) and the second threshold (e.g., 10mm), the system enters the "fine-tuning" stage.

[0153] At this point, a classic position closed-loop mode can be adopted, the core of which is the PID control algorithm. The PID controller generates continuous adjustment commands by performing comprehensive calculations of proportional (P), integral (I), and derivative (D) terms based on the error e(t) between the real-time displacement feedback (actual position) of the jack and the commanded target position. The proportional term determines the strength of the current response, the integral term is used to eliminate steady-state error, and the derivative term is used to predict the trend of change and suppress overshoot. The generated PID adjustment command is a dynamically changing control quantity that can drive the jack to smoothly and stably approach the target point, effectively avoiding the repeated oscillations near the target point that may be caused by the "speed control mode," thus achieving positioning with high steady-state accuracy.

[0154] Specifically, the output of the location closed-loop mode is:

[0155] ;

[0156] In the above formula, e(t) is the deviation at time t, Ki is the gain coefficient of the integral term, Kp is the gain coefficient of the proportional term, and Kd is the gain coefficient of the differential term.

[0157] Step S506: Is the maximum projected component not greater than the second threshold? If so, proceed to step S507.

[0158] Step S507: Select the optimization fine-tuning mode to generate the basic displacement command;

[0159] This step is the third and highest level of decision-making in the hierarchical control strategy, specifically designed to handle minute sub-millimeter deviations. Its core control objective is "global optimum," pursuing extreme accuracy under complex constraints. When the maximum projected component is less than or equal to the second threshold (e.g., 10 mm), the system is in the "fine-tuning" stage.

[0160] At this point, simple PID control may fall into local optima or cause high-frequency micro-vibrations due to system nonlinearity and coupling interference between multiple jacks. Therefore, the system activates an optimization fine-tuning mode, invoking a particle swarm optimization algorithm, a swarm intelligence-based optimization algorithm. In this scenario, each "particle" represents a set of displacement combinations for all jacks. The algorithm iteratively searches for the optimal displacement command combination that minimizes the final overall pose error (fitness function) of the steel arch frame by simulating the flight and information sharing of the particle swarm in the solution space. The generation of particle swarm optimization commands is no longer based on simple error feedback, but rather actively seeks a fine-tuning scheme that cleverly coordinates all actuators, overcomes system nonlinearity, and achieves optimal alignment through intelligent search. This breaks through the bottleneck of traditional control methods near the limit of accuracy.

[0161] Step S508: Calculate the displacement compensation vector of the distributed hydraulic jack based on the angle deviation components;

[0162] The logical principle behind this step is to specifically handle the correction of the angular deviation component and translate it into additional action commands for each hydraulic jack. Correcting the rotational (angular) deviation of the steel arch frame requires applying a torque through the jack assembly. According to rigid body kinematics, an angular deviation component rotating about an axis will cause an additional displacement at the point on the steel arch frame that contacts the jack's jacking point. The magnitude and direction of this displacement depend on the position vector of the contact point relative to the center of rotation and the angle of rotation, and can be calculated using the cross product of vectors, thus forming the displacement compensation vector. For each jack, the additional displacement required to compensate for the rotational deviation due to its own support point position needs to be calculated.

[0163] Specifically, the formula for calculating the displacement compensation vector is as follows:

[0164] ;

[0165] In the above formula, Let be the radial vector from the jack to the centroid of the arch frame. This represents the angular deviation component.

[0166] Step S509: Superimpose the basic displacement command with the displacement compensation vector to generate the coordinated displacement amount;

[0167] The formula for calculating the cooperative displacement is as follows:

[0168] In the above formula The basic displacement command for the i-th distributed hydraulic jack is the initial displacement value generated after selecting the control mode based on the magnitude of the position deviation.

[0169] Furthermore, the coordinated displacement must satisfy the following constraints:

[0170] ;

[0171] In the above formula, Let be the unit vector along the axis of the jack. For displacement deviation components, is the average radial vector from the jack assembly to the centroid of the steel arch frame.

[0172] Specifically, in a system with multiple distributed hydraulic jacks, correcting a certain overall translational deviation of a steel arch typically involves various combinations of jack actions. The collaborative allocation step aims to determine an optimal or feasible allocation strategy from both mathematical and physical perspectives. Its core is establishing a set of constraint equations: decomposing the total posture deviation of the steel arch into the displacement contribution requirements of each jack. By solving the equations, with the optimization objective of minimizing total displacement energy consumption, a set of specific collaborative displacement amounts allocated to each jack is obtained. This ensures that the actions of all jacks are coordinated and consistent, and their mechanical effects are accurately synthesized into the required total correction displacement, avoiding excessive internal stress in the steel arch or the generation of new parasitic posture errors due to uncoordinated actions.

[0173] Step S510: Encapsulate the coordinated displacement, execution direction vector and preset execution priority of each hydraulic jack into a hydraulic adjustment instruction set.

[0174] The hydraulic adjustment instruction set contains jack number-displacement key-value pairs. The encapsulation process organizes this information according to a predefined data structure agreed upon with the lower-level hydraulic controller. These jack number-displacement key-value pairs constitute the core mapping relationship of the instructions, ensuring accurate delivery to the designated actuator. The execution direction vector provides the direction cosine of the displacement, which is instructive for some advanced actuators with adjustable directions. The introduction of preset execution priorities allows the system to sort or group the actions of multiple jacks based on displacement magnitude, urgency of adjustment, or logical dependencies. This optimizes system response, avoids excessive instantaneous load, or implements specific adjustment timing logic. Ultimately, the generated hydraulic adjustment instruction set is a complete, unambiguous digital command package that can directly drive hardware actions, achieving clear and reliable data exchange between the intelligent decision-making system and the heavy-duty hydraulic actuator system.

[0175] In the above implementation, the composite pose deviation is first finely decoupled and projected onto the actuator, transforming the macroscopic pose problem into microscopic execution parameters. The core innovation of this technical solution lies in the introduction of a three-level intelligent decision-making mechanism based on the magnitude of the deviation: a rapid control mode is used for large deviations to achieve fast approach; a closed-loop PID controller is used for medium deviations to ensure smooth convergence; and a particle swarm optimization algorithm is employed to overcome the bottleneck of extreme precision for small deviations. Furthermore, through rigorous vector synthesis and allocation algorithms, the actions of multiple distributed hydraulic jacks are precisely coordinated, and the overall translational and rotational deviation corrections are rationally decomposed into the collaborative displacement and compensation vectors of each actuator. Finally, all command parameters are encapsulated into a structured machine instruction set, ensuring precise delivery and orderly execution of control. This technical solution changes the outdated approach of traditional engineering machinery that relies on manual experience, single-mode, and open-loop coarse control, achieving high-precision, collaborative intelligent control of the tunnel steel arch frame installation and adjustment process, improving installation efficiency, accuracy, and automation levels.

[0176] Reference Figure 6 As one implementation of step S106, the steps of constructing a hysteresis compensation model based on historical response data of the hydraulic system, correcting the transmission delay of the hydraulic adjustment command set, and generating a corrected hydraulic adjustment command set include:

[0177] Step S601: Extract the command sending timestamps, actual displacement and pressure feedback data from the historical execution records of the hydraulic system, and align them according to the time sequence to generate a command-response dataset with timestamps;

[0178] The historical execution record of the hydraulic system is a database containing multi-dimensional time-series information. The command sending timestamp accurately records the moment the control signal is issued, serving as the starting reference point for evaluating system latency. The actual displacement is measured by a high-precision displacement sensor, reflecting the true extension and retraction position of the hydraulic cylinder piston rod and directly representing the final output response of the system. The pressure feedback data comes from the oil circuit pressure sensor, which characterizes the complex influence of dynamic processes such as the load state, oil compressibility, and valve throttling within the hydraulic system.

[0179] In this embodiment, since command transmission, displacement sensing, and pressure measurement may be performed independently by different hardware units with microsecond-level differences, directly using the raw data would lead to analysis distortion. Therefore, by utilizing high-precision timestamps, each issued control command is precisely correlated with the subsequent temporally corresponding actual displacement change process triggered by that command and the pressure fluctuation state during the same period. Through interpolation, resampling, or event alignment algorithms, a command-response dataset is generated, in which each record explicitly contains a command and its observed system state after a specific delay.

[0180] Step S602: Perform time window segmentation on the command-response dataset and extract the temporal correlation features between the command displacement and the actual displacement within each time window;

[0181] From the aligned macroscopic data sequence, a sliding time window segmentation method is used to extract local microscopic features that can quantify the system's hysteresis dynamics. The response characteristics of hydraulic systems (such as delay time) are not constant but may drift slowly with factors such as oil temperature, load pressure, and sealing condition. Fixed analysis of the entire historical dataset would mask these time-varying characteristics. Using a sliding time window to segment long-sequence data into a series of continuous short-term subsets allows the analysis to focus on the system's short-term behavior under relatively stable operating conditions. Within each time window, analyzing the temporal relationship between commanded displacement (expected action) and actual displacement (actual action) allows for the extraction of a set of temporal correlation features.

[0182] Specifically, these time-domain correlation characteristics may include: the system's rise time to a step command (the time required for the displacement to increase from 10% to 90% of the rated displacement), which directly reflects the system's speed; the phase lag angle obtained by frequency domain transformation of the command and response signals, which reflects the system's delay characteristics at different frequencies; and the attenuation ratio of the system response to the command amplitude, which is related to the system's stiffness and damping characteristics. These characteristics describe the strength and pattern of the system lag from different dimensions and are key input variables for constructing accurate prediction models.

[0183] Step S603: Construct a prediction model for instruction transmission delay based on time-domain correlation features;

[0184] This process utilizes machine learning or system identification methods to learn and establish a mathematical mapping relationship from historical data that can predict future delays based on current operating conditions—that is, a predictive model. The time-domain correlation features (such as rise time and phase lag) extracted in the previous step within each time window are used as input features, and the average or typical command transmission delay statistically calculated within that time window is used as the output label. The model is then trained using regression algorithms (such as support vector regression, neural networks, or linear regression). This model learns the complex nonlinear relationship between system lag and observable features (which are implicitly driven by factors such as stress and command amplitude).

[0185] For example, the model might learn a pattern such as "when the system's average stress is high and the command is a small step, the delay is usually large." The constructed predictive model is a function or algorithm that can infer an estimate of the transmission delay that the next command will experience based on real-time or recently observed system characteristics (which implicitly contain the current dynamic state of the system).

[0186] Step S604: Input the hydraulic adjustment command set into the prediction model to obtain the transmission delay estimate of each hydraulic adjustment command;

[0187] When a new hydraulic adjustment command (including information such as target displacement) needs to be issued, the system captures the current context information in real time, such as the immediate pressure feedback of the hydraulic system, oil temperature (which can be indirectly reflected in features), and the attributes of the command itself (such as step size). This information is transformed into relevance features consistent with those used during model training (or the original signals that can reflect these states are used directly), and then input into the pre-built prediction model. Based on the patterns learned from historical data, the model performs forward computation and outputs a specific estimate of the command transmission delay for this particular command and the current system state. This estimate is no longer a historical average, but a dynamic, context-aware prediction that forecasts the time it will take from issuing the command to the actuator reaching the required position. This is the direct basis for accurate time compensation.

[0188] Step S605: Correct the sending timestamp of the hydraulic adjustment command based on the command transmission delay estimate to obtain the corrected hydraulic adjustment command set.

[0189] In traditional control, the instruction is issued at the expected execution time t_send. However, due to the inherent transmission delay estimate τ_hat in the system, the actual time when the instruction takes effect and completes the action will be t_send + τ_hat, resulting in a time lag in the control effect. To compensate for this lag, this step corrects the instruction issuance time.

[0190] Specifically, the original planned sending time t_send is subtracted from the model-predicted delay τ_hat, resulting in a new, earlier timestamp t_send' = t_send - τ_hat. The controller issues the command at this earlier time t_send', and after the actual system delay τ_hat, the command takes effect precisely (assuming accurate prediction) at the originally expected t_send time. This operation is performed internally by the digital controller and is transparent to the external setpoint or upper-level planner. It cleverly utilizes this time-dimensional advance action to offset the lag effect caused by system dynamics.

[0191] In the above implementation, the temporal correlation characteristics of historical operating data of the hydraulic system are deeply explored, and an intelligent model capable of accurately predicting command transmission delays is dynamically constructed. This model can proactively estimate the delay based on real-time command characteristics and system pressure status. By actively sending commands in advance, it offsets the inherent lag of the system in the time dimension, thereby transforming the hydraulic control system from a passive, lagging response to an active, quasi-synchronous, and precise execution. This improves the response speed and final positioning accuracy of the steel arch frame installation and adjustment, providing a crucial guarantee for the real-time performance and accuracy of the entire intelligent installation system.

[0192] Reference Figure 7 As a further implementation of the intelligent installation precision control method for tunnel steel arch frames, before the step of outputting the steel arch frame locking signal, the method further includes:

[0193] Step S701: Real-time acquisition of tunnel environmental parameters, including surrounding rock stress distribution data, ambient temperature and humidity values, and equipment vibration spectrum data;

[0194] The purpose of this step is to construct a comprehensive, multi-physics-based environmental perception layer, providing a data foundation for understanding and quantifying the impact of the external environment on installation accuracy. The tunnel construction environment is dynamic and complex. Surrounding rock stress distribution data reflects the redistribution of the force field within the rock mass after excavation. Its inhomogeneity or changes can cause slow creep or abrupt displacement of the initially positioned steel arch frame through initial support or direct action. Changes in ambient temperature and humidity affect the physical properties of the steel components (such as thermal expansion and contraction) and the viscosity of the working medium in the hydraulic system, thus introducing minute dimensional errors or execution deviations. Equipment vibration spectrum data characterizes the structural vibration interference transmitted by other heavy machinery on site (such as excavators and shotcrete machines). This vibration, different from the vibration noise of the measuring and executing equipment itself, is forced vibration acting on the steel arch frame body, which may cause it to deviate from its adjusted position.

[0195] Understandably, the core of synchronously collecting these multi-dimensional parameters through a distributed sensor network is to transform the originally elusive and empirical environmental disturbances into real-time data streams that can be monitored synchronously and analyzed quantitatively. This establishes a traceable and analyzable physical connection between environmental disturbances and the installed objects, which is a prerequisite for achieving intelligent environmental compensation.

[0196] Step S702: Align the current pose deviation with the tunnel environment parameters in the spatiotemporal dimension, establish the mapping relationship between the rate of change of the environment parameters and the pose deviation, and generate the environment-pose coupling matrix.

[0197] The core of this step is to perform data correlation analysis and feature extraction, aiming to discover potential causal or strongly correlated patterns from concurrent data and to model them mathematically. This is achieved by aligning these data within a unified spatiotemporal dimension, confirming that a specific pose error value observed at a certain moment corresponds to the environmental parameter state at the same moment (or a previous moment considering transmission delays).

[0198] Building upon this foundation, the correlation between the rate of change of environmental parameters (such as stress gradient and the rate of change of temperature and humidity) and pose errors is further analyzed. For example, a rapid drop in temperature is more likely to cause uneven shrinkage than low temperature itself. Through correlation analysis and regression analysis, a quantitative mapping relationship can be established. For instance, it was found that the arch foot displacement is linearly proportional to the stress gradient of the surrounding rock in a specific direction. The correlation strength coefficients between all key environmental factors and various pose errors (such as arch crown settlement, arch foot convergence, and overall torsion) are systematically organized into a mathematical structure, namely, the environment-pose coupling matrix. This matrix is ​​essentially a knowledge base or model that condenses the domain knowledge of "under what environmental disturbance characteristics, what type and magnitude of pose deviation does the steel arch tend to produce," serving as a key converter for transforming environmental perception data into specific compensation actions.

[0199] Step S703: Based on environmental deformation cases in the historical construction database, calculate the similarity between the environmental feature vector of the tunnel environmental parameters and the environmental deformation cases, and output the typical deformation pattern identifier with the highest feature matching degree.

[0200] The historical construction database stores environmental parameters, observed deformation patterns of steel arches, and corrective measures recorded in numerous past construction sections (or conditions). The currently collected and feature-extracted environmental parameters (represented as feature vectors) are compared with the feature vectors of historical cases in the database using similarity calculations (e.g., Euclidean distance and cosine similarity). The goal is to find the historical case most similar to the current environmental condition, and the matched typical deformation pattern identifiers (e.g., "inward displacement of the arch foot due to left-side bias" or "slight upward tilting of the arch crown due to a sudden drop in temperature") represent a high-level semantic induction and classification of complex environmental coupling effects. This technical solution avoids starting from scratch with complex physical modeling and solving for every new and complex multi-parameter coupling effect. Instead, it quickly locates the most likely type of deviation by retrieving similar historical experiences, providing clear expectations and directions for compensation, greatly improving the system's decision-making speed and reliability in dealing with complex environments.

[0201] Step S704: Based on the environment-pose coupling matrix, retrieve the reference offset of the corresponding key point of the steel arch frame according to the typical deformation mode identifier, and generate a compensation vector set by combining the feature matching degree weight.

[0202] In this process, based on the deformation pattern identifier obtained from the previous matching step, the baseline offset of key points on the steel arch frame that this pattern typically causes is retrieved from the case library or knowledge model. For example, for the "left-side bias" pattern, the baseline offset might be preset as follows: the left arch foot point shifts into the tunnel by X millimeters, the right arch foot point changes less, and the arch apex shifts to the right by Y millimeters. However, historical typical values ​​cannot be directly applied; they must be dynamically adjusted based on the matching weight between the current environment and historical cases. If the similarity between the current environmental features and historical case A is 90%, and with case B it is 70%, then the final compensation amount should be closer to the baseline value of case A, and interpolated or scaled using weights.

[0203] Understandably, the matching weights act as an adaptive modulator, ensuring that the compensation amount inherits historical experience while smoothly adjusting to subtle differences in the current situation. The final output compensation vector set represents the predicted displacement of each key point caused by environmental disturbances. This is a reverse predictive correction performed before the target pose is installed and adjusted.

[0204] Step S705: Superimpose the compensation vector set onto the current pose deviation, recalculate the target displacement of the hydraulic jack, and generate the hydraulic adjustment command after environmental correction.

[0205] The goal of environmental compensation is to further offset the effects of environmental disturbances that are "expected to occur or continue to occur." This is achieved by superimposing a compensation vector set (feedforward signal) onto the current pose deviation. Specifically, this is equivalent to pre-correcting the final target point of the hydraulic adjustment from the "theoretical design pose" to the position of "theoretical design pose minus the predicted environmental disturbance offset." Thus, when the system executes the command to adjust the steel arch to this "corrected target pose," subsequent environmental disturbances (whose effects have been predicted) will "push" it back to the vicinity of the theoretical design pose. Based on this new, composite pose target, the required target displacement for each hydraulic jack is recalculated, thereby generating the environmentally corrected hydraulic adjustment command. This gives the control system "predictability" and "proactiveness" against known disturbance patterns.

[0206] Step S706: After executing the hydraulic adjustment command after environmental correction, obtain the current pose deviation of the steel arch frame in the new round;

[0207] Step S707: Determine whether the current pose deviation in the new round is greater than the set deviation threshold.

[0208] After executing the environmentally compensated adjustment command, the system needs to pause and start a new round of measurement procedures (such as re-scanning 3D, feature extraction, and pose calculation) to obtain the latest pose error value of the steel arch after the compensation action. This latest actual error value is compared with the system's required installation accuracy threshold. If the error value is less than or equal to the threshold, it indicates that the environmental compensation is effective, the installation accuracy meets the standard, and the locking process can proceed. If the error value still exceeds the threshold, it indicates that the current compensation model (including coupling matrix, case matching, or weight calculation) has errors, the predicted compensation amount is inaccurate, it has failed to completely offset the impact of environmental disturbances, or an unexpected new disturbance mode has appeared. The judgment result of this step is the direct trigger condition for determining whether model self-learning is required.

[0209] If yes, proceed to step S708; otherwise, output a steel arch frame locking signal.

[0210] Step S708: Extract tunnel environment parameters and the current pose deviation for the new round, and update the mapping coefficients of the environment-pose coupling matrix.

[0211] When verification fails, it indicates a deviation between the system's current environment-pose relationship model and the real physical world. In this case, the system captures the current complete operating data (including environmental parameter characteristics and the final pose error data) as a new "failure case" or "correction sample." Using this new data, the mapping coefficients in the environment-pose coupling matrix are updated through optimization algorithms (such as least squares). For example, if the model initially predicted that the arch foot would shift inward by 5mm under a certain stress gradient, but it actually shifted inward by 7mm, the model will automatically increase the correlation coefficient between the stress gradient and the arch foot displacement.

[0212] Understandably, this process is an incremental learning process, enabling the coupling matrix to continuously self-correct and improve as construction progresses, geological conditions change, or seasons shift, resulting in increasingly accurate predictions. Through this continuous updating, the system not only provides more accurate compensation for current steel arch adjustments but, more importantly, accumulates more accurate knowledge for the installation of subsequent steel arch sections, achieving the digital accumulation and intelligent transfer of construction experience.

[0213] In the above implementation, deep environmental disturbances such as surrounding rock stress, temperature, humidity, and vibration are captured in real time, and a dynamic coupling model between these disturbances and the deformation error of the steel arch frame is constructed. Then, using case library matching technology, the most likely deformation mode is quickly identified, and a predictive spatial compensation vector is generated. This compensation amount is fed forward to the main control loop to actively cancel out environmental disturbances. Furthermore, by comparing the actual effect after compensation with the expected target, the error of the compensation model is automatically diagnosed, and its internal parameters are dynamically updated. This enables the system to adapt to the complex and variable geological and environmental conditions of tunnel engineering, thereby improving the robustness and adaptability of the intelligent installation system for tunnel steel arch frames under real and harsh working conditions.

[0214] Reference Figure 8 As a further implementation following step S707, if the current pose deviation in the new round is less than or equal to a set deviation threshold, the method further includes:

[0215] Step S801: Obtain the actual pose data and current pose deviation of the adjacent installed steel arch frames;

[0216] The tunnel support is an integral load-bearing structure composed of a series of steel arch frames connected by longitudinal joints, with close mechanical connections between adjacent arch frames. Obtaining the actual positional data of adjacent installed steel arch frames—that is, reading the final stable spatial coordinates and attitude of previously installed and locked arch frames—represents the spatial configuration of the formed, relatively stable support network. The current positional deviation of the steel arch frame represents its deviation from its theoretical design position.

[0217] Acquiring these two sets of data simultaneously aims to establish the "boundary conditions" for the current adjustment action: the system not only needs to adjust the current arch frame to its own design position, but also must ensure that this adjustment process does not cause excessive interference to adjacent arch frames that are already in place and may be bearing the pressure of the surrounding rock, thereby destroying the overall stability of the entire support system.

[0218] Step S802: Based on the pose data of adjacent installed steel arch frames, establish a mechanical coupling model of the arch frame group and calculate the interaction force between the current steel arch frame and the adjacent steel arch frames.

[0219] Specifically, by applying the principles of structural mechanics, the mechanical transmission relationship between adjacent arch frames is digitally modeled. The adjacent arch frames are regarded as a discretized mechanical system connected by springs, rods or beams. Based on actual pose data, the spatial relative positions of key points of each arch frame (such as arch foot and arch top) can be determined. By establishing a mechanical coupling model (usually a simplified finite element model based on stiffness matrix), this model can describe the interaction forces (including axial force, shear force or bending moment) transmitted to the connection points of adjacent arch frames through longitudinal connectors (such as steel bars and structural steel) when the key point of one arch frame is displaced.

[0220] Next, by solving the model, it is possible to calculate the pushing, pulling, or torsional effects that each step on the adjacent arches might produce along the "proposed adjustment path" from the initial deviation position to the designed orientation of the current steel arch. This calculation transforms the spatial geometric proximity into a quantifiable prediction of mechanical impact, which is the theoretical basis for predicting the impact of adjustment behavior on the surrounding structure.

[0221] Step S803: Based on the interaction force, predict the deformation trend of the current steel arch frame during the hydraulic adjustment process and generate a deformation compensation vector;

[0222] The logical principle behind this step is to consider the elastic deformation effect of the steel arch frame itself under non-ideal constraints. The interaction force for the steel arch frame being adjusted is the additional constraint reaction force at its support (i.e., the connection point with the adjacent arch frame).

[0223] Specifically, when hydraulic jacks push the main body of the steel arch frame, the constraint reaction forces from the supports cause the steel arch frame (as an elastic body) to undergo not only rigid body displacement but also additional elastic deformation. For example, if adjacent arch frames "hold" the arch foot of the current arch frame through connectors, the arch foot will be restricted during hydraulic jacking, potentially causing inward bending in the middle of the arch ring. Predicting the deformation trend involves calculating this undesirable elastic displacement field caused by interaction forces using mechanical models (such as beam bending models).

[0224] Subsequently, in order to counteract this elastic deformation in the final state and make the net deformation of the arch frame conform to the design shape after these coupling effects are removed (i.e., after adjustment is completed and the system is stable), a reverse displacement correction amount of equal magnitude needs to be added in advance to the adjustment command, that is, to generate a deformation compensation vector, which is a feedforward compensation based on mechanical prediction.

[0225] Step S804: The deformation compensation vector is superimposed on the current pose deviation of the steel arch frame to reconstruct the target displacement of the hydraulic adjustment command and obtain the hydraulic adjustment command after collaborative correction.

[0226] The original pose deviation only describes the amount of movement required for the current arch frame as a rigid body, while the deformation compensation vector includes additional adjustments needed to offset the predicted elastic deformation caused by coupling with adjacent arch frames. Superimposing these two vectors generates a new, composite adjustment target. The logic is as follows: the hydraulic system executes commands to drive the steel arch frame to this superimposed new target position. At this position, the steel arch frame is in a "pre-deformed" state: its own elastic deformation caused by coupling is precisely offset by this pre-deformation. When the adjustment is complete, the hydraulic system maintains force, and after the internal forces are redistributed between the connecting parts of adjacent arch frames to reach a new equilibrium, the steel arch frame rebounds from the "pre-deformed" state, and its final stable geometric shape is exactly the theoretical design shape.

[0227] Understandably, based on this new overall adjustment target, the target displacement required for each hydraulic jack is reconstructed, thereby generating a set of coordinated control commands that can both correct its own positional deviation and "digest" the impact on adjacent structures.

[0228] Step S805: Execute the hydraulic adjustment command after collaborative correction and monitor the displacement changes of adjacent arch frames in real time;

[0229] The accuracy of the preceding mechanical predictions and the safety of construction are verified by executing actions and observing actual effects. After the system executes the coordinated and corrected hydraulic commands, the displacement changes of adjacent arch frames must be monitored in real time. Monitoring whether adjacent arch frames have shifted, the direction of the shift, and the magnitude of the shift is direct evidence to determine whether the adjustment has truly achieved "no-interference" or "low-interference" coordination. If the prediction is accurate and the compensation is appropriate, the displacement of adjacent arch frames should be negligible (within the measurement noise range). If observable displacement occurs, it indicates a deviation between the actual mechanical coupling and the model prediction. This step extends the theoretical closed loop into a verification stage that includes real feedback from the physical world.

[0230] Step S806: If the displacement change of adjacent arch frames exceeds the safety threshold, trigger the parameter recalibration of the mechanical coupling model of the arch frame group.

[0231] The preset safety threshold is a small displacement that is permissible in engineering and does not affect the overall stability of the support system. When the monitored displacement change of adjacent arch frames exceeds this threshold, the system determines that the current mechanical coupling model has failed to accurately predict the actual interaction and that there is a model error (such as underestimating the connection stiffness or not considering the participation of the surrounding rock).

[0232] At this point, the system automatically triggers the model parameter recalibration process. Utilizing the measured hydraulic commands and adjacent arch displacement response data from this adjustment, the system identifies or optimizes parameters to correct key parameters in the coupled model (such as the equivalent stiffness coefficient of the connectors). This allows the model to learn from actual construction feedback, continuously improving its prediction accuracy for the mechanical behavior between arches under specific tunnel and connection methods, making the collaborative control strategy increasingly accurate and reliable as construction progresses.

[0233] Step S807: Output the steel arch locking signal that satisfies the cooperative stability condition.

[0234] Specifically, before the system outputs the locking signal, it must perform dual verification: First, the current positional error of the steel arch frame itself must meet the accuracy requirements; second, the adjustment process must not cause any disturbance to the established support system beyond the safe range, meaning the displacement monitoring results of adjacent arch frames are within the allowable threshold, and the entire arch frame group is in a stable new equilibrium state. Only when both conditions are met simultaneously will the system output the final steel arch frame locking signal, allowing welding or bolting for final fixing. This step ensures that the intelligent installation system not only focuses on single-point accuracy but also emphasizes maintaining the overall safety and stability of the tunnel support structure, achieving an intelligent upgrade from individual installation to system maintenance.

[0235] In the above implementation, the adjustment of a single arch frame is transformed from an isolated action into a collaborative operation within the overall support system. By establishing a mechanical coupling model of the arch frame group, the interaction forces during the adjustment process and the resulting elastic deformation trend are predicted and calculated in advance. This generates a predictive deformation compensation vector, which intelligently reconstructs the original adjustment command. This technical solution not only effectively suppresses the disturbance of the adjustment action to adjacent installed structures through a feedforward compensation mechanism, ensuring the overall stability of the support system, but also enables the system to continuously accumulate actual mechanical response data on site and optimize its internal model through real-time displacement monitoring and model parameter recalibration mechanisms after execution. This allows the system to achieve high-precision, low-interference, and self-learning intelligent installation of continuous steel arch frames under complex tunnel geological and structural conditions, fundamentally avoiding the risk of chain instability caused by improper adjustment of a single arch frame, and significantly improving the safety, efficiency, and intelligence level of tunnel construction.

[0236] In this embodiment, steps S701-S708 focus on the influence of external environmental factors on the posture of the steel arch frame. By real-time monitoring of environmental parameters such as surrounding rock stress, temperature, humidity, and vibration, an environment-posture coupling model is established to predict and compensate for deformation caused by environmental changes. Steps S801-S807 mainly focus on the mechanical coupling effect within the steel arch frame system. By analyzing the interaction forces between adjacent arch frames, deformation problems during installation are predicted and compensated, ensuring the overall stability when multiple arch frames work together. The former solves the problem of chain deformation caused by traditional single-unit adjustments and avoids overall instability caused by local adjustments; the latter effectively addresses the interference of environmental changes on construction accuracy and improves the reliability of construction in harsh environments. The combination of these two solutions constitutes a dual-guarantee system for intelligent installation of steel arch frames, suitable for tunnel construction under complex geological conditions.

[0237] In practical applications, environmental parameters are first acquired through a sensor network to predict and compensate for environmental impacts. Then, an environment-pose coupling matrix is ​​established to generate hydraulic adjustment commands after environmental correction. This step is equivalent to "coarse adjustment," eliminating the main impacts of environmental factors on the steel arch frame. Subsequently, based on environmental compensation, the mechanical coupling effect between adjacent arch frames is considered, and a mechanical coupling model of the arch frame group is established for precise coordinated adjustment. This step is equivalent to "fine adjustment," resolving the mutual influence within the multi-arch frame system. The synergistic effect of these two approaches significantly improves the installation accuracy and construction safety of the tunnel support structure, reduces the risk of engineering accidents caused by pose deviations, and provides key technical support for intelligent tunnel construction.

[0238] This application also discloses a machine vision-based intelligent installation accuracy control system for tunnel steel arch frames.

[0239] A machine vision-based intelligent installation precision control system for tunnel steel arch frames specifically includes:

[0240] The multi-source data acquisition module is used to synchronously trigger multi-source sensors through a precise time protocol to acquire visual image data, raw 3D point cloud data and motion state data of the steel arch frame, establish a spatiotemporal mapping relationship based on the calibration parameters of the multi-source sensors, and output a spatiotemporally aligned multimodal dataset.

[0241] The point cloud vibration suppression module is used to compensate for vibration noise in the original 3D point cloud data based on motion state data, and generate a denoised 3D point cloud.

[0242] The multimodal feature extraction module is used to identify and extract the coordinates of the top apex, left arch foot, and right arch foot of the steel arch frame based on denoised 3D point cloud and visual image data, construct the topological skeleton of the steel arch frame, and output a structured feature set.

[0243] The pose deviation calculation module is used to register and compare the structured feature set with the preset steel arch frame design model, and calculate the pose deviation between the actual pose and the design pose of the steel arch frame.

[0244] The decision control module is used to select the control mode based on the magnitude of the pose deviation and generate the corresponding hydraulic adjustment instruction set.

[0245] The hysteresis compensation module is used to build a hysteresis compensation model based on the historical response data of the hydraulic system, correct the transmission delay of the hydraulic adjustment command set, and generate a corrected hydraulic adjustment command set.

[0246] The collaborative hydraulic actuator module is used to control the distributed hydraulic jacks to perform displacement actions according to the modified hydraulic adjustment instruction set, and to monitor and output the actual displacement and pressure feedback data of the distributed hydraulic jacks in real time.

[0247] The iterative verification module is used to recalculate the current pose deviation of the steel arch frame based on the new round of collected steel arch frame data. If the current pose deviation is greater than the set deviation threshold, it returns to iterative adjustment until the error value is less than or equal to the set deviation threshold, at which point a steel arch frame locking signal is output.

[0248] The intelligent installation accuracy control system for tunnel steel arch frames based on machine vision according to the embodiments of this application can realize any of the above methods, and the specific working process of each module in the system can refer to the corresponding process in the above method embodiments.

[0249] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0250] This application also discloses a computer-readable storage medium.

[0251] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the machine vision-based intelligent installation precision control methods for tunnel steel arch frames.

[0252] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0253] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for intelligent installation accuracy control of tunnel steel arch frames based on machine vision, characterized in that, The control method includes: Multi-source sensors are synchronously triggered using a precise time protocol to collect visual image data, raw 3D point cloud data, and motion state data of the steel arch frame. A spatiotemporal mapping relationship is established based on the calibration parameters of the multi-source sensors, and a spatiotemporally aligned multimodal dataset is output. The calibration parameters of the multi-source sensors include the pre-calibrated rigid body transformation matrices between each sensor, namely the rotation matrix and the translation vector. Based on the motion state data, vibration noise compensation is performed on the original three-dimensional point cloud data to generate a denoised three-dimensional point cloud. Based on the denoised 3D point cloud and the visual image data, the coordinates of the top vertex, left arch foot, and right arch foot of the steel arch frame are identified and extracted, the topological skeleton of the steel arch frame is constructed, and a structured feature set is output. The structured feature set is registered and compared with the preset steel arch frame design model to calculate the pose deviation between the actual pose and the design pose of the steel arch frame. The control mode is selected based on the magnitude of the pose deviation, and a corresponding hydraulic adjustment command set is generated. A hysteresis compensation model is constructed based on the historical response data of the hydraulic system. The transmission delay of the hydraulic adjustment command set is corrected to generate a corrected hydraulic adjustment command set. The distributed hydraulic jacks are controlled to perform displacement actions according to the modified hydraulic adjustment instruction set, and the actual displacement and pressure feedback data of the distributed hydraulic jacks are monitored and output in real time. Based on the actual displacement and pressure feedback data, a new round of steel arch frame data acquisition is triggered; Based on the newly collected steel arch frame data, the current pose deviation of the steel arch frame is recalculated. If the current pose deviation is greater than the set deviation threshold, the iteration adjustment is repeated until the error value is less than or equal to the set deviation threshold, at which point the steel arch frame locking signal is output.

2. The method for intelligent installation accuracy control of tunnel steel arch frames based on machine vision according to claim 1, characterized in that, The steps for generating a denoised 3D point cloud by performing vibration noise compensation on the original 3D point cloud data based on the motion state data include: Extract the triaxial angular velocity components and timestamp sequence from the motion state data to generate an angular velocity-time relationship matrix; The displacement compensation of the point cloud is calculated based on the tunnel environment damping coefficient and the angular velocity-time relationship matrix. The point cloud displacement compensation is applied to the original three-dimensional point cloud data to generate a vibration-compensated intermediate point cloud. Extract the triaxial acceleration components from the motion state data and calculate the linear vibration offset of the equipment. The linear vibration offset is superimposed on the vibration compensation intermediate point cloud to output a denoised 3D point cloud.

3. The method for intelligent installation accuracy control of tunnel steel arch frames based on machine vision according to claim 2, characterized in that, Based on the denoised 3D point cloud and the visual image data, the steps of identifying and extracting the coordinates of the top vertex, left arch foot, and right arch foot of the steel arch frame, constructing the topological skeleton of the steel arch frame, and outputting the structured feature set include: The denoised 3D point cloud is divided into spatial meshes, and the outline edge point set of the steel arch frame is extracted; The visual image data is mapped to the contour edge point set using pixel-space coordinate mapping to identify candidate points in the arched foot region. Based on the candidate points in the arch foot region, fit a local surface, calculate the curvature extremum points, and output the coordinates of the left arch foot point and the right arch foot point. Search for the extreme points of the arch height in the denoised 3D point cloud, and verify the arch vertex coordinates by combining the texture features of the visual image data; Connect the top apex, left arch foot, and right arch foot to construct a topological skeleton of the steel arch frame containing spatial vector relationships, and output a structured feature set containing coordinates and vector topology.

4. The method for intelligent installation accuracy control of tunnel steel arch frames based on machine vision according to claim 1, characterized in that, The steps of registering and comparing the structured feature set with the preset steel arch frame design model to calculate the pose deviation between the actual pose and the designed pose of the steel arch frame include: Based on the coordinates of the arch apex, left arch foot, and right arch foot in the structured feature set, a local coordinate system for the steel arch frame is established. Extract the coordinates of the reference points from the pre-designed steel arch frame design model and construct the design coordinate system; Calculate the rigid body transformation matrix from the local coordinate system of the steel arch frame to the design coordinate system; The spatial vector relationships in the structured feature set are mapped to the design coordinate system through the rigid body transformation matrix; By comparing the coordinates of the mapped arch apex, left arch foot, and right arch foot with the coordinates of the corresponding points in the preset steel arch frame design model, a six-degree-of-freedom pose deviation quantity containing displacement deviation components and angle deviation components is generated. Similarity matching is performed between the mapped spatial vector relationship and the vector topology of the design model to correct the angular component error in the six-degree-of-freedom pose deviation, thus obtaining the final pose deviation.

5. The method for intelligent installation accuracy control of tunnel steel arch frames based on machine vision according to claim 4, characterized in that, The steps of selecting the control mode based on the magnitude of the pose deviation and generating the corresponding hydraulic adjustment command set include: The pose deviation is decomposed into displacement deviation components and angle deviation components, and the projection component of the displacement deviation component in the axis of the distributed hydraulic jack is calculated. If the maximum projected component is greater than the first threshold, then select the speed control mode to generate the basic displacement command; If the maximum projected component is not greater than the first threshold but is greater than the second threshold, select the position closed-loop mode to generate the basic displacement command. If the maximum projected component is not greater than the second threshold, then the optimized fine-tuning mode is selected to generate the basic displacement command; The displacement compensation vector of the distributed hydraulic jack is calculated based on the angular deviation components. The coordinated displacement, execution direction vector, and preset execution priority of each hydraulic jack are encapsulated into a hydraulic adjustment instruction set.

6. The method for intelligent installation accuracy control of tunnel steel arch frames based on machine vision according to claim 1, characterized in that, The steps for constructing a hysteresis compensation model based on historical response data of the hydraulic system, correcting the transmission delay of the hydraulic adjustment command set, and generating a corrected hydraulic adjustment command set include: Extract command sending timestamps, actual displacement, and pressure feedback data from the historical execution records of the hydraulic system, and align them according to the time sequence to generate a timestamped command-response dataset; The command-response dataset is segmented into time windows, and the temporal correlation features between command displacement and actual displacement within each time window are extracted. A prediction model for instruction transmission delay is constructed based on the aforementioned time-domain correlation characteristics; The hydraulic adjustment command set is input into the prediction model to obtain the transmission delay estimate for each hydraulic adjustment command; The timestamp of the hydraulic adjustment command is corrected based on the estimated command transmission delay to obtain the corrected hydraulic adjustment command set.

7. A method for intelligent installation accuracy control of tunnel steel arch frames based on machine vision according to any one of claims 1 to 6, characterized in that, Before the step of outputting the steel arch locking signal, the following is also included: Real-time acquisition of tunnel environmental parameters, including surrounding rock stress distribution data, ambient temperature and humidity values, and equipment vibration spectrum data; Align the current pose deviation with the tunnel environment parameters in the spatiotemporal dimension, establish a mapping relationship between the rate of change of environment parameters and pose deviation, and generate an environment-pose coupling matrix. Based on environmental deformation cases in the historical construction database, the similarity between the environmental feature vector of the tunnel environmental parameters and the environmental deformation cases is calculated, and the typical deformation pattern identifier with the highest feature matching degree is output. Based on the environment-pose coupling matrix, the reference offset of the corresponding key point of the steel arch frame is retrieved according to the typical deformation mode identifier, and a compensation vector set is generated by combining the feature matching degree weight. The compensation vector set is superimposed on the current pose deviation, the target displacement of the hydraulic jack is recalculated, and the hydraulic adjustment command after environmental correction is generated. After executing the hydraulic adjustment command after environmental correction, obtain the current pose deviation of the steel arch frame in the new round; Determine whether the current pose deviation in the new round is greater than the set deviation threshold; If the value is greater than the value, the tunnel environment parameters and the current pose deviation for the next round are extracted, and the mapping coefficients of the environment-pose coupling matrix are updated.

8. The method for intelligent installation accuracy control of tunnel steel arch frames based on machine vision according to claim 7, characterized in that, If the current pose deviation in the new round is less than or equal to a set deviation threshold, the method further includes: Obtain the actual pose data and current pose deviation of adjacent installed steel arch frames; Based on the pose data of adjacent installed steel arch frames, a mechanical coupling model of the arch frame group is established to solve the interaction force between the current steel arch frame and the adjacent steel arch frames. Based on the interaction forces, the deformation trend of the current steel arch frame during the hydraulic adjustment process is predicted, and a deformation compensation vector is generated. The deformation compensation vector is superimposed on the current pose deviation of the steel arch frame to reconstruct the target displacement of the hydraulic adjustment command, thereby obtaining the hydraulic adjustment command after collaborative correction. Execute the hydraulic adjustment command after the collaborative correction, and monitor the displacement changes of adjacent arch frames in real time; If the displacement change of adjacent arch frames exceeds the safety threshold, the parameter recalibration of the mechanical coupling model of the arch frame group is triggered. Output the locking signal of the steel arch frame that satisfies the cooperative stability condition.

9. A machine vision-based intelligent installation accuracy control system for tunnel steel arch frames, characterized in that, The control system includes: The multi-source data acquisition module is used to synchronously trigger multi-source sensors through a precise time protocol to acquire visual image data, raw 3D point cloud data, and motion state data of the steel arch frame. Based on the calibration parameters of the multi-source sensors, a spatiotemporal mapping relationship is established, and a spatiotemporally aligned multimodal dataset is output. The multi-source sensor calibration parameters include the pre-calibrated rigid body transformation matrices between each sensor, namely rotation matrices and translation vectors. The point cloud vibration suppression module is used to perform vibration noise compensation on the original three-dimensional point cloud data based on the motion state data to generate a denoised three-dimensional point cloud. The multimodal feature extraction module is used to identify and extract the coordinates of the top vertex, left arch foot, and right arch foot of the steel arch frame based on the denoised 3D point cloud and the visual image data, construct the topological skeleton of the steel arch frame, and output a structured feature set. The pose deviation calculation module is used to register and compare the structured feature set with the preset steel arch frame design model, and calculate the pose deviation between the actual pose and the design pose of the steel arch frame. The decision control module is used to select the control mode according to the magnitude of the pose deviation and generate the corresponding hydraulic adjustment instruction set. The hysteresis compensation module is used to construct a hysteresis compensation model based on the historical response data of the hydraulic system, correct the transmission delay of the hydraulic adjustment command set, and generate a corrected hydraulic adjustment command set. The collaborative hydraulic execution module is used to control the distributed hydraulic jacks to perform displacement actions according to the modified hydraulic adjustment instruction set, and to monitor and output the actual displacement and pressure feedback data of the distributed hydraulic jacks in real time. The iterative verification module is used to recalculate the current pose deviation of the steel arch frame based on the new round of collected steel arch frame data. If the current pose deviation is greater than the set deviation threshold, it returns to iterative adjustment until the error value is less than or equal to the set deviation threshold, at which point a steel arch frame locking signal is output.

10. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 8.

Images