Adjustment method for flatness measurement based on catenary parabola

Abstract

The application provides a kind of adjustment method based on catenary parabola flatness measurement, it is related to data processing technical field, the method includes pixel adjustment module and parabola adjustment module.Pixel adjustment module obtains the actual projection track of steel wire by image acquisition, perspective correction and distortion correction, homography mapping and center line extraction.Parabola adjustment module calculates the longitudinal coordinate difference value of actual track according to theoretical parabola, and outputs coordinated control instruction to electromagnetic force actuator along the steel wire through collaborative control unit calculation, to distribute the active adjustment of steel wire form by vertical electromagnetic force.Two modules work in closed loop cooperation according to fixed period, until the actual track of steel wire and the deviation of theoretical parabola is less than the set threshold, form stable, high-precision dynamic measurement reference.Active establishment and dynamic maintenance of catenary reference are realized, the precision and reliability of flatness measurement are improved, and an efficient and intelligent solution is provided for large-size component flatness detection.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to an adjustment method for flatness measurement based on a suspension parabola. Background Technology

[0002] Flatness is one of the key geometric tolerances for evaluating the surface quality of mechanical components. In aerospace, shipbuilding, large machine tools, wind power equipment, and other fields, the flatness accuracy of large-size components (such as wing panels, hull sections, and machine tool tables) directly affects their assembly quality, structural stability, and service performance. Therefore, developing high-precision, high-efficiency, and highly adaptable large-size flatness measurement technology is of significant engineering importance.

[0003] Currently, flatness measurement methods can be mainly divided into two categories: contact and non-contact. Contact methods (such as dial indicators and coordinate measuring machines) acquire data through physical contact between the probe and the surface being measured. Although they offer high accuracy, they have inherent limitations such as the measurement force easily causing deformation of thin-walled parts, low measurement efficiency, and stringent requirements for the on-site environment, making it difficult to meet the needs of rapid, in-situ inspection of large-sized components. Non-contact methods (such as laser trackers, laser interferometers, and digital photogrammetry) avoid the influence of contact force, but in large-scale measurements, they often face problems such as expensive equipment, complex systems, weak environmental interference resistance, or the need for multiple relocations and splicing, limiting their widespread adoption and on-site applicability.

[0004] Against this backdrop, the suspension cable method, a classic indirect method for measuring large-scale straightness and flatness, has regained attention. Its basic principle is to utilize thin steel wires stretched at both ends, which hang naturally under their own weight, forming a stable, continuous, and predictable catenary or parabola, serving as a natural baseline for spatial measurement. By measuring the deviation of each point on the measured surface relative to this baseline, the flatness error can be calculated. This method boasts significant advantages such as simple equipment, low cost, theoretically unlimited measurement range, and no need for complex environmental control.

[0005] However, the traditional suspension method faces a fundamental technical challenge in practical applications: how to accurately obtain and maintain the spatial position of the theoretical parabola under actual working conditions, namely, the problem of "accurate establishment and dynamic maintenance of the parabolic reference." Specifically, this manifests as follows:

[0006] Initial shape uncertainty: The theoretical shape (parabola) of a steel wire sags under its own weight is uniquely determined by parameters such as its material, diameter, span, and tension at both ends. However, in actual erection, due to various factors such as height errors at the end support points, installation eccentricity, uneven weight distribution of the steel wire itself, and inaccurate initial tension control, the initial static shape of the steel wire often deviates significantly from the theoretical parabola. If this deviation is not corrected, it will be directly transmitted as a systematic error to the final flatness measurement result.

[0007] Environmental disturbance sensitivity: Minor disturbances such as airflow, ambient temperature changes, and mechanical vibrations, which are difficult to completely avoid at the measurement site, can cause the suspension cable to produce continuous low-frequency oscillations or slow morphological drift. The static, "one-time" baseline in traditional methods cannot remain stable in dynamic environments, making the measurement baseline itself a source of error, which severely restricts measurement accuracy and is particularly unfavorable for long-term or online measurements.

[0008] Benchmark calibration and traceability are challenging: To convert relative displacement measured by images or sensors into absolute spatial coordinates, high-precision calibration of the parabolic benchmark is necessary. Traditional methods rely on placing a precise calibration object or a benchmark ruler of known length within the measurement field of view. However, this method is often cumbersome (requiring precise coplanar placement), highly intrusive (occupying measurement space), and the calibration result only represents the initial state, failing to compensate for errors introduced during measurement due to the deformation of the benchmark itself or drift of the imaging system.

[0009] To address the aforementioned problem of "difficulty in determining and maintaining the parabolic position," existing technologies have attempted some improvements, such as using higher-rigidity steel wires, adding wind shields, and implementing constant temperature control, aiming to suppress disturbances at the hardware level; or employing multiple measurements and averaging, and complex post-processing algorithms to compensate for fluctuations. However, these methods either significantly increase system complexity and cost, or remain passive compensation methods, failing to fundamentally achieve active stabilization, real-time correction, and high-precision calibration of the baseline.

[0010] Therefore, when using the suspension parabola for high-precision flatness measurement, the key to overcoming the accuracy bottleneck of the traditional suspension method and unleashing its potential in the field of large-size precision measurement lies in how to construct an intelligent adjustment method and system that can automatically, in real time, and accurately adjust the suspension shape to the theoretical parabolic state, and continuously compensate for environmental interference. This is not only a requirement for improving the accuracy of single measurement results, but also a core technical problem that must be overcome to realize the application of this technology from the laboratory to complex industrial sites and from intermittent measurement to online monitoring.

[0011] In view of the above, this application is hereby submitted. Summary of the Invention

[0012] To address the aforementioned problems in existing technologies, an adjustment method based on the flatness measurement of a suspension parabola is provided, with the aim of solving at least one of the above problems.

[0013] The technical solution to achieve the purpose of this invention is as follows:

[0014] This invention provides an adjustment method for flatness measurement based on a suspension parabola, which is implemented using a pixel adjustment module and a parabola adjustment module;

[0015] The pixel adjustment module acquires at least an image including the steel wire. Using a calibration reference or an auxiliary marker with known physical dimensions installed at the measurement station, it obtains positional information for at least two known coordinates. The acquired image undergoes perspective correction and distortion correction. Through homography transformation, the steel wire image is mapped onto a virtual correction plane parallel to the theoretical plane of the steel wire. In the coordinate system of this correction plane, a linear mapping relationship is established between the image pixel coordinates and the actual physical coordinates, ensuring that the actual physical dimension corresponding to each pixel on the measurement plane remains constant. Subsequently, the precise centerline of the steel wire in the corrected image is extracted as its current actual projection trajectory.

[0016] The parabola adjustment module includes a collaborative control unit and several electromagnetic force actuators evenly distributed along the axis of the measuring steel wire. It extracts the actual projected trajectory of the steel wire and generates a corresponding theoretical parabola based on the wire's length and trajectory. It calculates the difference in ordinate between the actual projected trajectory and the theoretical parabola at the corresponding abscissa positions. The control unit receives the sequence of these ordinate differences and converts them into control signals with positive and negative directions. After calculation by the control unit, it outputs a set of coordinated control commands, rather than independent incremental signals, to each electromagnetic force actuator. The aim is to adjust the distributed vertical force acting on the steel wire to make its overall shape smoothly and stably approach the preset theoretical parabola shape.

[0017] The pixel adjustment module and the parabolic adjustment module work together at a fixed cycle, sequentially executing a closed-loop process of image acquisition, perspective correction, trajectory extraction, difference calculation, and collaborative control output. This process is repeated until the parabolic adjustment module determines that the deviation of the actual projected trajectory of the steel wire from the theoretical parabolic baseline at each detection point is less than a set threshold. At this point, the baseline adjustment is complete. Finally, the stable and adjusted suspension cable shape is used as a high-precision measurement benchmark, and the system performs subsequent flatness detection and calculation.

[0018] Compared with the prior art, the beneficial effects of the present invention include:

[0019] (1) It realizes the active establishment and dynamic maintenance of the suspension parabolic baseline. The traditional suspension method relies on the passive drooping shape of the steel wire, which is affected by factors such as initial erection deviation and environmental disturbance, making it difficult to accurately stabilize the baseline. This invention forms a closed loop through the coordinated control of visual perception and electromagnetic force, which can monitor and actively adjust the shape of the steel wire in real time, so that it converges to the theoretical parabola quickly and smoothly, fundamentally solving the problem of accurate establishment and dynamic stability of the baseline;

[0020] (2) Improved the accuracy and reliability of flatness measurement. Through high-precision image correction, sub-pixel-level centerline extraction and elastic matching comparison, the system can accurately identify the micro-deviation between the steel wire and the theoretical parabola. Combined with predictive control based on physical models and distributed force control execution, it can achieve accurate compensation for deviations, significantly improve the overall measurement accuracy, and ensure reliable performance in industrial environments such as vibration and temperature changes.

[0021] (3) It has strong environmental adaptability and anti-interference ability. The system adopts dynamic calibration, real-time perspective correction, distortion residual compensation and other technologies, which can adaptively compensate for errors caused by camera pose drift, illumination changes and mechanical vibration. The elastic matching algorithm and quality factor evaluation mechanism further enhance the robustness of the system under non-ideal conditions and are suitable for complex industrial sites.

[0022] (4) It achieves full-process automation and intelligence. From image acquisition, coordinate mapping, trajectory extraction to closed-loop control, the entire process requires no manual intervention. The system autonomously completes benchmark establishment, status monitoring, and dynamic adjustment. The built-in self-diagnosis and parameter self-adaptation mechanisms further enhance the system's intelligence level and support long-term unattended operation.

[0023] (5) Balancing high precision and engineering practicality, compared with traditional solutions that rely on high-rigidity hardware, constant temperature environment or frequent calibration, this invention reduces the stringent requirements on hardware through intelligent compensation at the algorithm level, can work stably in conventional industrial environments, and has advantages such as low cost, flexible deployment and easy maintenance.

[0024] (6) It provides a highly reliable dynamic benchmark for subsequent flatness testing. The adjusted suspension cable shape is stable and traceable, providing a high-precision spatial reference benchmark for subsequent measurements. It effectively avoids measurement distortion caused by benchmark error transmission in traditional methods and provides reliable technical support for the precision flatness testing of large-size components. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the overall structure of the flatness measuring device;

[0027] Figure 2 This is a closed-loop flowchart of the pixel adjustment module and the parabolic adjustment module;

[0028] The components include: 1. Measuring device; 2. Magnetic base; 3. Part to be measured; 4. Counterweight; 5. Steel wire; and 6. Hook. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention.

[0030] Therefore, the following detailed description of embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely illustrates some embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0031] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the present invention can be combined with each other.

[0032] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0033] The present invention will be further described in detail below with reference to embodiments.

[0034] like Figures 1 to 2 As shown, the present invention provides an adjustment method for flatness measurement based on a suspension parabola, which is implemented using a pixel adjustment module and a parabola adjustment module;

[0035] The pixel adjustment module acquires at least an image including the steel wire. Using a calibration reference or an auxiliary marker with known physical dimensions installed at the measurement station, it obtains positional information for at least two known coordinates. The acquired image undergoes perspective correction and distortion correction. Through homography transformation, the steel wire image is mapped onto a virtual correction plane parallel to the theoretical plane of the steel wire. In the coordinate system of this correction plane, a linear mapping relationship is established between the image pixel coordinates and the actual physical coordinates, ensuring that the actual physical dimension corresponding to each pixel on the measurement plane remains constant. Subsequently, the precise centerline of the steel wire in the corrected image is extracted as its current actual projection trajectory.

[0036] The parabola adjustment module includes a collaborative control unit and several electromagnetic actuators evenly distributed along the axis of the steel wire. It extracts the actual projected trajectory of the steel wire and compares it with a pre-calculated theoretical parabola baseline based on the parabola formula for the self-weight of the suspension cable. The module calculates the difference in the vertical coordinates of the two lines at their corresponding abscissa positions. The control unit receives the sequence of these vertical coordinate differences and converts them into control signals with positive and negative directions. Given that the steel wire is a continuous flexible body, localized forces can cause changes in its overall shape. Therefore, this module adopts a collaborative control strategy based on a lumped parameter model. After calculation, the control unit outputs a set of coordinated control commands to each electromagnetic actuator, rather than independent incremental signals. This aims to adjust the distributed vertical forces acting on the steel wire to make its overall shape smoothly and stably approach the preset theoretical parabola shape.

[0037] The pixel adjustment module and the parabolic adjustment module work together at a fixed cycle, sequentially executing a closed-loop process of image acquisition, perspective correction, trajectory extraction, difference calculation, and collaborative control output. This process is repeated until the parabolic adjustment module determines that the deviation of the actual projected trajectory of the steel wire from the theoretical parabolic baseline at each detection point is less than a set threshold. At this point, the baseline adjustment is complete. Finally, the stable and adjusted suspension cable shape is used as a high-precision measurement benchmark, and the system performs subsequent flatness detection and calculation.

[0038] It should be noted that this invention addresses the core problem in traditional suspension cable flatness measurement: the difficulty in accurately establishing and dynamically maintaining a parabolic datum due to initial setup deviations, environmental disturbances, and calibration difficulties. It constructs a closed-loop adjustment system integrating high-precision visual perception, intelligent model comparison, and force control, transforming the traditional passive, static suspension cable datum into an actively calibrated, dynamically stable intelligent measurement datum. This invention integrates a pixel adjustment module and a parabolic adjustment module. The pixel adjustment module achieves high-precision, interference-resistant, real-time measurement and coordinate reconstruction of the steel wire's spatial morphology by introducing a series of visual perception and image processing technologies, including rapid calibration with non-coplanar stereo targets, an embedded dynamic calibration network, layered adaptive perspective and distortion correction, confidence-weighted coordinate mapping, and sub-pixel-level centerline extraction based on a physical model. The parabolic adjustment module, based on the actual trajectory of the steel wire provided by visual perception, intelligently generates an error vector field corresponding to the layout of the electromagnetic actuators through spatiotemporal registration and elastic matching algorithms. Furthermore, based on a deep understanding of the dynamics of the continuous flexible steel wire, a parameterized centralized model is constructed, and a model predictive control framework is adopted to calculate the optimal control command that achieves global coordination, anticipates coupling effects, and balances response speed and stability. Finally, through a distributed electromagnetic actuator network with bidirectional high-precision force control, adaptive compliant contact, and optimized layout, a coordinated vertical force is applied to the steel wire, causing its shape to smoothly and stably converge to the preset theoretical parabola. This invention, through the aforementioned closed-loop feedback and active adjustment mechanism, fundamentally solves the accuracy and stability problems of the suspension cable baseline, realizing a paradigm shift from passively adapting to the baseline to the baseline actively obeying the measurement. This provides a highly reliable dynamic measurement baseline for subsequent flatness detection, significantly improving the practicality, accuracy, and efficiency of the suspension cable method in large-size, high-precision, and industrial environments.

[0039] It should be further noted that this invention is based on and improved upon the following proprietary technology, combined with... Figure 1The diagram illustrates the existing technology, whose technical solution mainly includes a benchmark establishment system and a data acquisition and theoretical model. The benchmark establishment system involves attaching two adjustable magnetic bases 2 to the reference surfaces at both ends of the part being measured 3, with the attraction force set to be greater than 1.5 times the wire tension. A 0.2mm diameter steel wire 5 is used, one end connected to the horizontal rod of one magnetic base 2, and the other end passing over the horizontal rod of the other magnetic base 2 and connected to a 2.5kg counterweight 4. Quick positioning is achieved via a hook 6, with a positioning error ≤0.01mm. The tension adjustment knob is manually adjusted based on the counterweight weight and support spacing to tension the steel wire to its theoretically natural hanging state. The data acquisition and theoretical model involves placing the measuring device 1 on the upper surface of the part being measured and positioning the differential sensor directly below the steel wire 5. Based on the input wire parameters (diameter, material density), counterweight mass, and support spacing (e.g., 3000 mm), the weight per unit length p and end tension T0 of the wire are calculated, thus generating a theoretical sag parabolic model y = p·x² / T0. The position of the measuring device is adjusted so that the differential head contacts the wire and triggers a signal. The differential head contacts the wire at multiple points to obtain actual sag profile data. The collected actual sag data is compared with the theoretical parabolic model, and the least squares method is used for fitting to calculate the flatness deviation value of the measured surface (e.g., maximum deviation 0.015 mm), and a comparison graph of the actual and theoretical curves is plotted.

[0040] Preferably, in some embodiments of the present invention, in the pixel adjustment module, the system employs several high-resolution industrial area scan cameras, such as 3 or 10 high-resolution industrial area scan cameras, fixedly installed above the measurement station. Their optical axes are at a non-orthogonal angle (typically 30° to 60°) to the theoretical plane of the steel wire being measured, so as to simultaneously acquire sufficient axial field of view and significant vertical deviation information of the steel wire in a single image. The calibration reference is not a traditional two-dimensional planar calibration plate, but a three-dimensional target. This target consists of a base and one or more raised feature pillars. The top of each feature pillar is machined with a high-contrast circular marker (such as a solid black circle). The three-dimensional coordinates of all markers in the target's own coordinate system have been precisely calibrated using a higher-level instrument (such as a coordinate measuring machine).

[0041] During the initialization phase, the operator randomly places the 3D target on the workpiece table near the steel wire, without requiring it to be coplanar with or precisely aligned with the steel wire plane; it is only necessary to ensure that some of its marker points appear simultaneously with the steel wire in the camera's field of view. The system controls the camera to capture an initialization image that simultaneously includes both the steel wire and the 3D target. The calibration information acquisition process is a multi-parameter joint calculation step, which includes the following steps:

[0042] Step A1: The image processing algorithm first identifies all visible marker points on the stereo target in the initialized image and extracts their sub-pixel precision image coordinates.

[0043] Step A2 involves constructing a set of overdetermined equations based on the known 3D world coordinates of the marker points and their corresponding 2D image coordinates, combined with the camera's pinhole imaging model and a pre-selected lens distortion mathematical model (such as the Brown-Conrady model). Using nonlinear optimization algorithms such as the Levenberg-Marquardt algorithm, the extrinsic parameters of the camera relative to the target coordinate system (rotation matrix R and translation vector T) and the intrinsic parameters of the lens (focal length, principal point) and distortion coefficients are simultaneously calculated. This step completes the acquisition of camera pose calibration and lens distortion correction parameters in one step.

[0044] Step A3, based on the calculated high-precision camera model and combined with the known equations of the theoretical plane of the wire in space (e.g., a spatial plane defined by two fixed pivot points), directly calculates and generates a dedicated homography transformation matrix from the pixel coordinates of any captured wire image to the two-dimensional physical coordinates on the theoretical plane of the wire through three-dimensional geometric projection relationships. This matrix is ​​essentially a product of strongly coupling the camera imaging model with a specific measurement plane and is the core of achieving a constant mapping from pixels to physical dimensions.

[0045] By placing a stereo target and capturing a single initial image, the system not only completed camera calibration but also directly obtained a direct mapping relationship from the image to the measurement plane optimized for this measurement task, providing accurate transformation parameters for subsequent real-time perspective correction.

[0046] In existing technologies, vision-based linear reference measurement methods typically rely on two-dimensional planar calibration plates or on-site calibration methods based on known lengths for calibration. The two-dimensional planar calibration plate method requires the calibration plate to be precisely placed within or parallel to the plane being measured, which is cumbersome in practice and severely restricts the spatial layout of the measurement site. Calibration and measurement are two separate steps, and the calibration plate often needs to be removed during measurement, failing to compensate for errors introduced by slight camera vibrations or thermal drift. On-site calibration methods based on known lengths establish the pixel-to-physical size ratio by placing a reference object of known length within the field of view. This method cannot correct for perspective distortion and lens nonlinear distortion, and is only suitable for the ideal situation where the camera optical axis is perpendicular to the measured surface. Furthermore, the scaling factor may not be constant within the field of view, resulting in limited measurement accuracy.

[0047] This invention decouples the spatial constraints between the calibration object and the measurement plane, allowing the 3D target to be placed in any orientation at the measurement site without precise alignment with the wire plane. This significantly simplifies initialization, reduces requirements for site layout, and enables rapid deployment in space-constrained or complex industrial environments. Camera parameter calibration, lens distortion correction, and the generation of a dedicated mapping relationship to the theoretical plane of the wire are integrated into an automated joint calculation process. This avoids the multiple propagation of errors caused by step-by-step calibration, minimizing imaging geometric model errors from the source and laying a sub-pixel-level accuracy foundation for the entire vision measurement chain. The precisely solved homography transformation matrix actively compensates for perspective distortion introduced by the tilted camera installation. This allows the system to fully utilize the tilted perspective to obtain richer information on the vertical deformation of the wire, overcoming the limitations of traditional vertical perspective or simple proportional calibration methods, and expanding the dimensions of measurement information while ensuring accuracy. The dedicated homography transformation matrix, calculated in a single step, essentially encapsulates core information such as the camera model, distortion parameters, and measurement plane orientation. The matrix can be directly called in subsequent continuous measurement cycles, avoiding real-time repeated calculations. This not only reduces computational complexity and introduced uncertainties, but also ensures the consistency between the calibration environment and the measurement environment, effectively suppressing system errors and improving overall stability and long-term measurement efficiency.

[0048] Preferably, in some embodiments of the present invention, the pixel adjustment module may further employ an embedded real-time dynamic calibration network to adapt to long-term continuous measurement or situations where environmental disturbances exist. This network consists of a set of discretely distributed intelligent reference units, each fixedly installed in a non-critical area of ​​the measurement environment structure or the surface of the workpiece being measured, and its mounting plane is macroscopically parallel to the theoretical plane of the steel wire being measured.

[0049] Each intelligent reference unit integrates two core components: a micro-calibration pattern with a unique topological code (e.g., a variant based on Aztec or circular coding) and a high-precision miniature temperature-compensated attitude sensor (such as a MEMS inertial measurement unit) coplanarly integrated with the pattern. The precise geometric dimensions of the micro-calibration pattern and its physical coordinates in the mounting coordinate system are calibrated and archived before shipment; its coding design allows the visual recognition algorithm to quickly and uniquely identify its identity (ID) even under different viewing angles and partial occlusion. The attitude sensor monitors the unit's spatial orientation (pitch angle, roll angle) and temperature data in real time and transmits it to the central processing unit via a low-power wireless network (such as ZigBee). During system operation, the dynamic calibration network runs continuously. In each measurement cycle, the visual algorithm first identifies at least two intelligent reference units from the image, obtaining their pixel coordinates and IDs. Based on the IDs, a database is queried to obtain the corresponding physical coordinates and attitude data fed back by the sensor in real time. The system uses this attitude data to dynamically calculate or correct a virtual reference plane reflecting the current state of the actual physical reference surface. Subsequently, when calculating the homography transformation from the image pixel coordinate system to the world coordinate system, the algorithm not only relies on the pixel-physical coordinate correspondence but also introduces the information of this virtual reference plane as a strong constraint. This allows for real-time, first-order compensation for reference plane distortion caused by minor deformation of the mounting surface, environmental vibration, or thermal effects. In this way, the embedded network can periodically refresh the mapping model throughout the measurement process, achieving dynamic suppression of slow camera parameter drift and environmental disturbances, ensuring the long-term maintenance of mapping accuracy and reference stability during long-term measurements.

[0050] In existing technologies, traditional calibration methods primarily rely on a one-time camera calibration using a high-precision checkerboard or dot array calibration board before measurement. This method has inherent limitations. The first limitation is its static nature; once calibration is complete, the parameters are fixed. Parameter changes caused by thermal drift due to temperature variations, lens mechanical stress relaxation, or slight changes in installation position cannot be perceived or compensated for, introducing measurement errors that accumulate over time. The second limitation is its invasiveness; large calibration boards occupy significant space in the measurement field of view, competing for visual space with the measurement target (such as a suspension cable), or require repeated movement in and out during the measurement process, making operation cumbersome and unsuitable for online monitoring. The third limitation is environmental vulnerability; calibration depends on the ambient light conditions at the initial moment of measurement, and subsequent changes in lighting may affect the accuracy of feature point extraction, and it cannot correct for physical deformation of the reference surface caused by field vibrations.

[0051] Compared to the existing technologies described above, the embedded real-time dynamic calibration network of this invention, through distributed, state-aware intelligent units, transforms calibration from a one-time, off-threaded process into a continuous online process. This fundamentally overcomes the static and fixed defects of traditional methods and can sense and compensate for changes in the reference surface caused by temperature drift, mechanical deformation, or vibration in real time, effectively eliminating time-drift errors and significantly improving the stability and absolute accuracy of long-term measurements. The miniaturized and discrete reference units hardly obstruct the measurement subject, supporting simultaneous calibration and measurement without interrupting the process by introducing or removing large calibration boards, simplifying operation, improving efficiency, and enabling online monitoring. Its unique coding design and anti-occlusion recognition algorithm, combined with wireless transmission, ensure reliability and ease of deployment in complex industrial environments such as oil stains, flying debris, and uneven lighting. The attitude data fed back by each unit can be used to indirectly diagnose the stability of camera mounts or workpiece platforms, and the communication link status can also reflect network health. This enables the system to go beyond basic coordinate mapping functions, possessing deeper operational status monitoring and preliminary diagnostic capabilities.

[0052] Furthermore, in some embodiments of the present invention, in the pixel adjustment module, after initial calibration based on a stereo target or an embedded dynamic calibration network, the system obtains a high-precision camera model (intrinsic parameters, distortion coefficients) and an initial homography transformation matrix H0, which establishes a direct mapping relationship from the image plane to the theoretical plane of the wire. However, during continuous measurement, the camera may undergo slight pose changes due to environmental vibration, temperature drift, or mechanical stress relaxation, causing the pre-calculated H0 to gradually become invalid, thereby introducing perspective errors that accumulate over time. In addition, lens distortion is non-uniform on the imaging plane, and simple global correction models may retain errors at the edges of the field of view. To solve this problem, the present invention adopts a layered, progressive, real-time adaptive perspective correction and distortion residual correction, which not only performs geometric mapping but also realizes dynamic perception and compensation of error sources in the imaging process. The present invention adopts a two-layer correction architecture: the first layer is dynamic perspective correction based on feature tracking and pose estimation; the second layer is online estimation and compensation of distortion residuals based on the geometric characteristics of the wire image itself.

[0053] Dynamic perspective correction based on feature tracking and pose estimation includes the following steps:

[0054] Step B1: Within the measurement field of view, in addition to the steel wire being measured, the system pre-sets or automatically identifies several sets (e.g., 10 or 15 sets) of stable, high-contrast auxiliary visual features. These features may originate from coded patterns on intelligent reference units in the embedded dynamic calibration network, or from special reflective markers pre-attached to non-critical areas of the measurement platform or workpiece, or from inherent corner or texture features of the workpiece that appear stably in the image sequence.

[0055] In step B2, the system extracts the subpixel-level image coordinates of auxiliary features in the initial frame and establishes a 3D-2D correspondence database based on the initial calibration results (known or inversely calculated 3D coordinates of camera extrinsic parameters and feature points). After each subsequent frame is acquired, firstly, in the raw, distorted image without perspective correction, the KLT (Kanade-Lucas-Tomasi) optical flow method or a descriptor-based fast matching algorithm is used to track the image positions of these auxiliary features.

[0056] Step B3 utilizes the 2D coordinates and corresponding 3D coordinates of the tracked feature points, employing RANSAC (Random Sample Consensus) combined with the EPnP (Efficient PnP) algorithm to robustly estimate the rotation matrix R_t and translation vector T_t of the camera relative to the world coordinate system (or calibration target coordinate system) in the current frame. The RANSAC process effectively eliminates outliers caused by occlusion, sudden changes in illumination, or mismatches, ensuring the stability of pose estimation. When tracking auxiliary feature points, due to changes in illumination, slight occlusion, or image blurring, tracking algorithms (such as KLT optical flow) may produce erroneous tracking results (i.e., outliers). Directly using all these points (including erroneous points) to calculate the camera pose (R_t, T_t) will result in severely distorted results. Combining RANSAC with the PnP algorithm forms the RANSAC-EPnP process. The input is all tracked 2D-3D point pairs (mixed with erroneous tracking). RANSAC randomly selects 4 pairs of points, and EPnP is used to calculate a pose hypothesis. This pose is used to examine all point pairs, distinguishing which are correctly tracked interior points (small error) and which are incorrect exterior points (large error). This process is repeated multiple times to find the pose supported by the most correctly tracked points. Even if nearly half of the feature points fail to track, the system can still calculate the correct camera pose, ensuring the reliability of dynamic correction. It automatically filters out erroneous sensor data without manual intervention, demonstrating the system's intelligence and autonomy. It provides clean and accurate input data for the generation of the dynamic homography matrix, a crucial step in ultimately achieving sub-pixel accuracy and long-term stability.

[0057] Step B4 involves recalculating the homography matrix H_t from the image plane to the theoretical wire plane at the current moment, based on the real-time estimated current camera pose (R_t, T_t), the calibrated camera intrinsic parameter matrix K, and the spatial equation of the wire theoretical plane (defined by two fixed pivot points). The dynamic update of H_t essentially feeds back the real-time changes in the camera extrinsic parameters to the perspective transformation model, thereby compensating for the viewing angle change error caused by camera pose drift and ensuring that the perspective correction relationship always remains consistent with the current actual imaging geometry.

[0058] Even after initial lens distortion model correction, due to model imperfections or calibration errors, the theoretically straight wire image may still contain slight residual curves in the transformed corrected image. Online estimation and compensation of distortion residuals based on the inherent geometric characteristics of the wire image includes the following steps:

[0059] Step C1: Use the dynamically generated H_t to perform inverse perspective mapping on the original image of the current frame (which has undergone preliminary lens distortion correction) to obtain a preliminary corrected image. In this image, use the grayscale profiling method or the Steger algorithm to coarsely extract the centerline pixel set {p_i} of the steel wire;

[0060] Step C2: Assuming that after the H_t transformation, the ideal centerline of the steel wire should be a straight line, the coarsely extracted centerline point set {p_i} is fitted to a straight line L_ideal, and the vertical deviation δ_i from each point p_i to the fitted line is calculated. This deviation δ_i is considered a representation of the residual distortion in the current image. The system establishes a low-order polynomial model (e.g., second or third order) to describe the residual distortion field D_residual(x,y), which maps image coordinates to vertical deviations. Using the coordinates of {p_i} and the corresponding δ_i, the parameters of the D_residual model are fitted using the least squares method.

[0061] Step C3 involves applying the estimated residual distortion field D_residual to the pre-corrected image to fine-tune the pixel positions in the wire centerline region, eliminating δ_i. Then, high-precision sub-pixel centerline extraction is performed again on the finely corrected image to obtain the final actual projected trajectory of the wire used for comparison. This process can be iterated rapidly 1-2 times within a single frame until the straightness of the centerline meets a preset threshold.

[0062] Existing technologies for perspective correction and distortion correction generally employ one-time calibration plus static transformation, discrete processing, and reliance on highly stable hardware. One-time calibration plus static transformation involves calibrating the camera before measurement, calculating fixed distortion parameters and homography matrices, and maintaining these constants throughout the measurement process. This method is completely incapable of handling any changes in the camera or environment during measurement, and system accuracy degrades over time / within the environment. Discrete processing treats distortion correction and perspective correction as two independent, sequential image processing operations. This process may accumulate interpolation errors due to two resampling operations and is computationally inefficient. More importantly, it lacks a mechanism for posterior optimization using the geometric characteristics of the measured target itself. Reliance on highly stable hardware involves using ultra-rigid supports, temperature control devices, or active pose control platforms to suppress camera pose changes. This method is extremely costly, complex, and still cannot eliminate all microscopic deformations, while also reducing the engineering practicality and deployability of the solution.

[0063] This invention continuously tracks stable features in the environment and estimates camera pose in real time, dynamically updating the core parameters of perspective transformation (H_t). This transforms the correction process from an open-loop operation based on initial conditions to a closed-loop autonomous adjustment based on real-time feedback, fundamentally solving the long-term accuracy degradation problem caused by camera pose drift. It breaks through the traditional limitation of calibration relying solely on dedicated calibration objects, creatively using the steel wire itself as a reference benchmark for online evaluation and correction of residual imaging distortion. By fitting its theoretically straight projection to infer and compensate for the residual distortion field, high-order compensation for complex imaging errors is achieved. Dynamic perspective correction addressing changes in macroscopic geometric relationships and target-based residual distortion compensation addressing microscopic imaging defects are processed in layers and iterated sequentially. The two layers share intermediate results and work collaboratively, ensuring robustness to environmental changes while pursuing absolute accuracy under extreme conditions. The dynamic correction mechanism effectively suppresses systematic errors introduced by slow camera pose drift, temperature changes, and slight vibrations, ensuring the measurement benchmark remains stable during continuous long-term monitoring, making it particularly suitable for high-precision continuous measurement tasks in industrial applications. Without relying on ultra-stable installation platforms or constant temperature systems, the system utilizes intelligent compensation at the algorithm level to operate stably in conventional industrial environments with certain vibrations and temperature variations, significantly improving the economics, deployability, and environmental adaptability of the technical solution. The system can sense changes in its own imaging state, assess imaging quality, and autonomously adjust parameters and compensate for errors, demonstrating the evolution from static measurement tools to sensing systems with inherent intelligence.

[0064] Furthermore, in some embodiments of the present invention, in the pixel adjustment module, after completing image geometric normalization based on dynamic perspective correction and distortion residual compensation, the system obtains a virtual corrected image that is strictly parallel to the theoretical plane of the steel wire. To accurately convert the pixel coordinates (u', v') in this corrected image into two-dimensional physical coordinates (X, Y) on the theoretical plane of the steel wire, the present invention employs a hierarchical weighted fusion and dynamic self-calibration linear mapping method to achieve sub-pixel level coordinate transformation and construct a mapping confidence evaluation system with self-verification and online correction capabilities. Traditional methods typically use preset fixed scaling coefficients for simple scaling, which cannot effectively address the accuracy differences of homography transformation in different regions, optical focal length drift caused by ambient temperature, and mapping consistency issues between multi-camera systems. Therefore, the present invention employs a multi-source constraint fusion dynamic linear mapping model; this model includes hierarchical mapping coefficient calculation based on region confidence weighting and dynamic self-calibration of mapping coefficients based on closed-loop feedback.

[0065] Furthermore, in some embodiments of the present invention, in the pixel adjustment module, based on the calculation of layered mapping coefficients weighted by region confidence, during the initialization phase, the system utilizes marker points with known three-dimensional coordinates on the stereo target to establish a sparse set of reference mapping point pairs in the calibration image. Each point pair contains sub-pixel coordinates in the calibration image and their corresponding precise physical coordinates on the theoretical plane of the wire. These point pairs are non-uniformly distributed throughout the entire measurement field of view. Based on the aforementioned reference points, the system divides the calibration image into several overlapping sub-regions (e.g., generating triangular meshes through Delaunay triangulation). Within each triangular mesh, a local affine transformation model is fitted using the least squares method using its vertices and neighboring reference point pairs. This model uses a set of coefficients to describe the linear mapping relationship from pixel coordinates to physical coordinates. Simultaneously, the fitting residual of this local model is calculated as a confidence index of the mapping accuracy of the region; the smaller the residual, the higher the confidence. During real-time coordinate transformation, for any point in the calibration image, the system first determines the triangular mesh in which it resides (which may be located in the overlapping area of ​​multiple meshes). For each grid cell containing a point, a physical coordinate estimate is calculated using its local affine model. The final physical coordinates are obtained through a confidence-weighted fusion: the estimates of each grid cell are weighted according to their confidence weights, with the weights inversely proportional to the fitting residuals of the corresponding grid cells. This weighting mechanism ensures that the model in high-precision regions plays a dominant role in coordinate transformation, effectively suppressing errors at image edges or in areas with sparse features.

[0066] Furthermore, in some embodiments of the present invention, in the pixel adjustment module, dynamic self-calibration of the mapping coefficients based on closed-loop feedback is implemented. To address the potential thermal drift of camera internal parameters (especially focal length) over time, the present invention introduces an implicit self-calibration mechanism in the measurement closed loop. The system fixes one or more microscale reference scales of known physical length at the edge of the measurement station, with their direction parallel or perpendicular to the steel wire axis. In the calibration image of each measurement cycle, the system synchronously identifies and extracts images of these reference scales. Using the current mapping model, the physical distance between the two endpoints of the scale in the image is calculated and compared with the actual length of the scale to obtain the scale error. This error is fed back to the mapping model management system. The system maintains a scale drift compensation factor. When the scale errors detected in multiple consecutive measurement cycles all exceed a preset threshold and are in the same direction, the system determines that a significant scale drift has occurred. At this time, the system fine-tunes the global mapping scaling coefficient according to the error ratio, or performs consistent scaling of the scale-related parameters in all local affine models and updates the compensation factor. This process is gradual and conservative, ensuring that the system maintains stable mapping accuracy over long-term operation and avoiding misadjustment due to a single abnormal measurement.

[0067] In existing technologies, methods for establishing the mapping relationship between pixel coordinates and physical coordinates mainly include the global single scale factor method, the global homography matrix method based on a calibration board, and the multi-region calibration and lookup table method. The global single scale factor method calculates the pixel equivalent (e.g., millimeters / pixel) by placing a reference object of known length in the field of view and assuming that this scale is constant throughout the entire field of view. This method ignores residual nonlinearity that may exist after perspective correction, image edge resolution attenuation, and local scale changes caused by imperfect lens distortion correction, easily introducing significant errors in sub-pixel level precision measurements. The global homography matrix method based on a calibration board uses a calibration board to obtain a homography matrix (a 3×3 projection transformation matrix) from the image plane to the physical plane, which can describe transformations such as scaling, rotation, shearing, and perspective. However, this method fixes the mapping relationship, cannot cope with the slow drift of camera internal parameters (such as focal length) over time or temperature, and its mapping accuracy is not uniform throughout the entire field of view, lacking a targeted evaluation and compensation mechanism. Multi-region calibration and lookup table method divides the field of view into grids and stores coordinate transformation vectors at each node. It calculates the coordinates through bilinear interpolation. Although this method can improve local accuracy, it requires dense calibration points, the calibration process is cumbersome and time-consuming, and it is also a static mapping that cannot adapt to dynamic changes in system parameters.

[0068] To address the aforementioned issues, this invention employs a hierarchical weighted fusion and dynamic self-calibration linear mapping method. First, through a mapping architecture combining local modeling and global fusion, based on triangulation of reference points and fitting of local affine models, the nonlinear mapping differences in different regions within the field of view are precisely described and compensated. Then, confidence-weighted fusion ensures overall smoothness and coordinate transformation continuity, improving accuracy while avoiding the discontinuities and significant storage overhead of traditional lookup methods. Second, the concept of mapping confidence is introduced and used for decision fusion. The system includes a confidence estimate based on local model residuals for each coordinate transformation result. This information can be used for weighted calculations in downstream processing (such as trajectory filtering) or for system health status diagnosis, achieving an improvement from coordinate transformation to reliable output. Furthermore, an online self-calibration mechanism based on a built-in microscale reference is designed, embedding the microscale reference scale as a "calibration anchor point" into the measurement environment. This allows the system to synchronously complete in-situ monitoring and slow drift compensation of mapping scale accuracy during normal measurement cycles, realizing a shift from periodic interrupted maintenance to continuous, accompanying fine-tuning, and improving the reliability of long-term unattended operation. Achieving consistent high-precision coordinate transformation within the field of view, this system effectively compensates for the attenuation of mapping accuracy at image edges and sparsely featured regions through local model and confidence-weighted fusion. This ensures that the physical coordinate extraction accuracy of any point on the wire trajectory within the entire measurement area reaches the sub-pixel level, laying the foundation for subsequent high-precision comparison with theoretical parabolas. Improving the long-term stability and absolute accuracy of the measurement system, a built-in closed-loop self-calibration mechanism actively senses and compensates for mapping scale changes caused by factors such as camera thermal drift, suppressing the unavoidable time-varying errors in traditional static mapping methods, and ensuring consistent and accurate measurement results during long-term continuous operation. Enhancing the system's intelligent perception and self-diagnostic capabilities, confidence assessment provides quality indicators for each coordinate transformation, and error monitoring data during the self-calibration process serves as a health status signal for system performance, enabling the system to perform self-evaluation and maintenance. Optimizes engineering practicality and life-cycle maintenance costs, while ensuring high precision and reducing extreme requirements on camera lens optical quality, ambient temperature control and installation rigidity. Algorithm compensation allows for the use of more cost-effective hardware and stable operation in a more relaxed environment. Online self-calibration reduces reliance on manual periodic on-site recalibration, thereby reducing maintenance costs and complexity.

[0069] Furthermore, in some embodiments of the present invention, in the pixel adjustment module, after completing the high-precision linear mapping from pixel coordinates to physical coordinates, the system obtains a corrected image with accurate geometric relationships and constant scale, in which the steel wire image appears as a bright or dark band region with a specific width. The core task of this step is to extract the centerline of the steel wire. The goal is not only to locate a pixel-level centerline, but also to achieve sub-pixel-level extraction accuracy and ensure that the centerline trajectory still has strong noise resistance and high stability under typical industrial interferences such as high noise, uneven lighting, steel wire surface reflection, or local occlusion. To this end, the present invention proposes a fusion extraction method based on multi-scale linear enhancement and grayscale profile iterative modeling. Traditional centerline extraction methods (such as skeletonization, grayscale centroid method, or edge centering method) are prone to centerline breakage, position drift, or burrs when dealing with low contrast, uneven lighting, or complex noise. This invention constructs a two-stage progressive extraction process. The first stage performs robust initial localization based on the linear response of the multi-scale Hessian matrix, and the second stage performs sub-pixel-level iterative refinement based on grayscale profile physical model fitting. The multi-scale Hessian matrix is ​​used to accurately identify and enhance linear structures from the image.

[0070] Furthermore, in some embodiments of the present invention, in the pixel adjustment module, robust initial localization based on multi-scale linear enhancement is first performed. For the expected physical width of the wire in the corrected image (e.g., 1-3 mm, which can be converted to a pixel width range), a set of multi-scale linear filters based on the second derivative of the Gaussian function is constructed. By changing the standard deviation parameter of the Gaussian function, filter banks of different scales are generated, each scale specifically designed to enhance linear structures of a particular width while suppressing noise, speckles, and edge interference in the image.

[0071] Subsequently, the Hessian matrix of the input corrected image is calculated at multiple scales. This matrix is ​​composed of the second-order partial derivatives of the image (obtained through Gaussian kernel convolution at the corresponding scales). Eigenvalue decomposition is performed on the Hessian matrix at each pixel location, and a linear structure response value is constructed based on the eigenvalue relationships. This response value reaches its maximum at the center line of the wire and decays as it deviates from the center line, and is insensitive to changes in uniform illumination.

[0072] Next, for each pixel, the value that maximizes its linear response is selected from all preset scales, generating a linear response map and its corresponding optimal scale map. The linear response map forms a high-response ridge at the actual location of the wire;

[0073] Finally, on the linear response map, nonmaximum suppression combined with a path search algorithm (such as dynamic programming) is used to trace the high-response ridge along the extension direction of the wire, thereby obtaining a series of discrete, pixel-level precision initial center point positions, and recording the optimal scale information corresponding to each point.

[0074] Furthermore, in some embodiments of the present invention, in the pixel adjustment module, the sub-pixel refinement based on the grayscale profile physical model only achieves pixel-level accuracy for the initial center point obtained in the first stage, and may be subject to deviations due to local interference. To further improve accuracy, the present invention performs point-by-point refinement based on the physical characteristics of wire imaging. For each initial center point, a one-dimensional grayscale profile is intercepted along its normal direction (which can be estimated perpendicularly from the ridge direction at that point). Considering the blurring effect of the optical system, the cross-sectional grayscale profile of an ideal uniform cylindrical steel wire can be approximately modeled as a superposition model of a Gaussian function and a uniform background. The model parameters include the Gaussian center position (i.e., the true sub-pixel-level center offset), amplitude, width, and background grayscale.

[0075] Then, the actual sampled grayscale profile data is iteratively fitted with the Gaussian model using weighted least squares. During the fitting process, weights are assigned to the data points to reduce the influence of abnormal data points caused by local strong reflections, stains, or nearby interference on the steel wire surface. By using the model parameters obtained from the fitting, the sub-pixel-level precise offset of the centerline of the point relative to its initial position in the normal direction can be obtained.

[0076] In addition, the optimal scale information (implicit wire apparent width) obtained in the first stage is used to verify the consistency of the fitting results. If the fitting parameters (such as center offset or Gaussian width) of a certain point are seriously inconsistent with the expected range based on the scale, the point is regarded as an anomaly, and interpolation is performed based on the results of adjacent reliable points.

[0077] Finally, a lightweight filter (such as the Savitzky-Golay filter) based on the physical continuity and smoothness of the wire is applied to all the refined sub-pixel coordinate point sequences along the entire center line to obtain the final smooth, continuous and high-precision actual projection trajectory of the wire, which can be used by the subsequent parabolic adjustment module.

[0078] In existing technologies, the mainstream methods for extracting linear centerlines mainly include the gray-scale centroid method, edge detection centering method, skeletonization (central axis transformation) method, and Steger algorithm. The gray-scale centroid method uses the weighted center of grayscale values ​​as the center position on the gray-scale profile along the normal direction. Although computationally simple, it is very sensitive to asymmetric gray-scale distributions (e.g., unilateral reflections), and its extraction accuracy is usually limited to the pixel level. The edge detection centering method first detects the edges on both sides of the wire, then takes the midpoint between the two edges as the center. Its accuracy heavily depends on the accuracy of edge detection and the symmetry of the two edges. When the edges are blurred, broken, or unevenly illuminated, the error is large, and detecting the edges on both sides also increases computational complexity and the possibility of error accumulation. The skeletonization (central axis transformation) method refines the binarized wire region to obtain a single-pixel skeleton. It is extremely sensitive to the choice of binarization threshold, easily producing burrs, short branches, and topological deformations, and losing sub-pixel accuracy information. The Steger algorithm is based on the Hessian matrix and uses Taylor expansion to solve for the extreme brightness points to obtain the sub-pixel positions. It performs well in extracting thin lines, but may fail to calculate correctly or produce unstable results when the line width varies, the contrast is low, or the noise is strong, and the computational load is relatively large.

[0079] To address the limitations of the aforementioned methods, this invention employs a fusion extraction method based on multi-scale linear enhancement and grayscale profile iterative modeling. The first stage utilizes multi-scale Hessian response for linear structure enhancement and robust initial localization, effectively addressing noise and non-uniform background interference, providing stable initial points and normal direction estimates for the second stage. The second stage performs local fitting based on a physical model of wire imaging (such as a Gaussian distribution model), directly solving for the sub-pixel-level center position through parameter optimization, fully utilizing image grayscale information to improve accuracy. Abandoning purely geometric or grayscale moment-based localization approaches, this invention establishes a physical model of the grayscale distribution of the wire cross-section based on optical imaging principles, directly calculating the center position through model fitting. This makes the localization process closer to the actual imaging process, enhancing anti-interference capability and positioning accuracy. The optimal scale information extracted in the first stage is used as a constraint in the second stage to judge the rationality of the fitting results (such as scale consistency checks), forming an information loop and quality control between the two stages, effectively identifying and eliminating erroneous extraction points caused by strong local interference. A weighted strategy is employed in the iterative weighted least squares fitting to adaptively reduce the impact of abnormal grayscale values ​​such as reflective points and stains. Anomaly removal and interpolation strategies based on neighboring points ensure that the centerline remains intact even when local image damage occurs. Stable and reliable sub-pixel level extraction accuracy is achieved; through physical model fitting, the centerline positioning accuracy reaches sub-pixel level, providing a guarantee for subsequent high-precision comparison with the theoretical parabola. This is a key step in achieving micrometer-level accuracy for the overall measurement system. Excellent anti-interference capability and environmental robustness are demonstrated; the multi-scale linear enhancement stage is insensitive to uneven illumination and background changes. The model fitting stage suppresses the influence of local abnormal grayscale values ​​through a weighted mechanism, enabling stable operation in typical industrial vision environments with multiple interferences. The continuity and smoothness of the centerline trajectory are ensured; a topologically correct initial centerline is obtained through ridge tracing, followed by anomaly processing and lightweight smoothing filtering. The final output trajectory combines high accuracy and good continuity, conforming to the physical form of a flexible suspension cable, facilitating subsequent processing by the control module. Balancing algorithm accuracy and computational efficiency, the first-level filtering and response calculations can be parallelized, and the second-level model fitting performs lightweight optimization on one-dimensional profile data. The overall process complexity is controllable, and real-time processing (e.g., above 30Hz) can be achieved on standard industrial computing platforms, meeting the real-time requirements for dynamic adjustment of the closed loop.

[0080] Furthermore, in some embodiments of the present invention, the parabolic adjustment module in the pixel adjustment module also includes a comparison and difference sequence generation stage between the actual trajectory and the theoretical parabolic baseline. This stage is not a simple subtraction of geometric line values, but constitutes an intelligent differential engine with spatiotemporal alignment capability, fault tolerance mechanism, and error field analysis function. Its goal is to accurately extract the spatial distribution requirements of the driving force for closed-loop control from discrete, noisy measurement data and provide a basis for the system's state self-diagnosis. The system receives the actual projected trajectory of the steel wire, Trajectory_actual, from the pixel adjustment module. This trajectory is a set of high-precision two-dimensional point sequences {(X_i, Y_i)} with non-uniform sampling; the theoretical parabolic baseline, Trajectory_theoretical, is a continuous analytical curve Y_th(X)=aX²+bX+c defined by the parabolic formula of the suspension cable's self-weight. To overcome the three problems existing in direct comparison—non-one-to-one correspondence of sampling points, measurement noise interference, and unclear subsequent control requirements—the present invention adopts a comparison process of spatiotemporal registration-elastic matching-vectorized difference, which includes the following steps:

[0081] Step D1: At the beginning of each closed-loop cycle, the system binds a unified high-precision timestamp to the currently acquired raw image frame, the processed actual trajectory point sequence, and the theoretical parabolic parameters read from the parabola adjustment module (these parameters can be adaptively compensated based on ambient temperature and wire tension fine-tuning). This ensures that the comparison is based on the system state at the same physical moment, avoiding comparison deviations introduced by data processing pipeline delays or asynchronous parameter updates. The actual trajectory point sequence {(X_i, Y_i)} is defined in the measurement physical coordinate system established by the pixel adjustment module; the placement position of the electromagnetic force actuator and the theoretical parabolic formula are defined in the control physical coordinate system. There may be fixed translational and rotational relationships between the two coordinate systems due to installation. During the initialization phase, the system determines the rigid transformation matrix T_m2c between the two through a one-time calibration. Before comparison, all actual trajectory points are transformed to the control coordinate system through T_m2c to obtain {(X'_i, Y'_i)}, ensuring that the calculation is performed in the same spatial reference system as the actual actuator layout and theoretical benchmark, where i represents a natural number.

[0082] Step D2: Under the control coordinate system, based on the distribution range of the actual trajectory points' abscissas {X'_i}, the theoretical parabola Y_th(X) is adaptively densely discretized to generate a series of theoretical points {(X_j, Y_th(X_j))}. The discrete spacing ΔX is much smaller than the spacing of the electromagnetic actuators to ensure subsequent matching accuracy. To address the non-strict alignment problem of the abscissas caused by measurement noise and local deformation, the system adopts an elastic matching algorithm based on dynamic time warping. Within the allowable lateral elasticity range, for each actual point P_i(X'_i, Y'_i), the corresponding theoretical point Q_i(X_match_i, Y_th(X_match_i)) is optimally matched in the dense theoretical point set. The matching cost function comprehensively considers the lateral distance and the expected longitudinal error, thus tolerating small local lateral fluctuations in the actual trajectory and ensuring that the longitudinal difference ΔY_i=Y'_i-Y_th(X_match_i) truly reflects the vertical morphological deviation. To improve the accuracy of the comparison, when calculating ΔY_i, the value of Y_th(X_match_i) is not directly taken from the discrete theoretical point. Instead, when X'_i is not exactly equal to the x-coordinate of the discrete point, the theoretical y-coordinate value with sub-pixel accuracy is obtained by directly calculating through the theoretical parabola analytical formula or by using high-order interpolation, so that the difference ΔY_i itself has sub-pixel accuracy.

[0083] Step D3: After elastic matching and sub-pixel calculation, the system obtains the longitudinal difference ΔY_i corresponding to each actual measurement point P_i. Based on the known axial position X_actuator_k of each electromagnetic actuator, the ΔY_i of multiple nearby measurement points is aggregated to the corresponding actuator using inverse distance weighting or Gaussian weighting, forming a longitudinal error vector sequence {E_k} corresponding to the actuator layout, where E_k represents the vertical displacement (with sign) that the wire needs to be adjusted within the area controlled by actuator k.

[0084] The system further constructs a discrete representation of the error vector field distributed along the wire axis based on all (X'_i,ΔY_i) point pairs, and performs spatial analysis on it, including calculating the first derivative (error slope) and second derivative (error curvature) of the error field to determine the error distribution pattern, providing a basis for the collaborative control unit to select control strategies under different error modes.

[0085] In addition, the system calculates the overall quality factor Q for this comparison, which includes matching confidence, data integrity, and error field smoothness. Matching confidence is based on the average matching cost during the elastic matching process; data integrity is the proportion of valid matching points to the total number of measurement points; and error field smoothness is the spatial standard deviation of the error vector or the energy of high-frequency components. This quality factor Q will be transmitted to the collaborative control unit for dynamically adjusting the gain of the control algorithm or triggering the system's self-diagnostic process.

[0086] This invention establishes a unified spatiotemporal and coordinate framework across the entire process. By binding high-precision timestamps to data acquisition, processing, and control commands, it ensures that each stage is compared based on the state at the same physical moment. Simultaneously, during the initialization phase, a rigid transformation matrix from the measurement coordinate system to the control coordinate system is calibrated and applied, ensuring that the comparison is performed in a unified spatial reference system. An elastic matching algorithm is introduced to address non-ideal data alignment, drawing on and improving upon the concept of dynamic time warping, and applying it to spatial trajectory matching. This algorithm allows for limited lateral elastic alignment between the actual trajectory and the theoretical benchmark. By comprehensively weighing the cost function of lateral distance and expected longitudinal error, it optimizes the matching of the most suitable theoretical position for each actual measurement point. This mechanism effectively tolerates reasonable lateral fluctuations in actual suspension cables, accurately identifying pure vertical morphological deviations, and greatly enhancing the robustness and accuracy of error extraction in environments with vibration or disturbance. Furthermore, this invention upgrades from point-to-point differences to an error vector field. It not only calculates a high-precision longitudinal difference sequence but also constructs an error vector field distributed along the steel wire axis and analyzes its spatial characteristics (such as the first and second derivatives). This upgrades the output from a series of isolated error scalars to structured data containing spatial distribution pattern information. The collaborative control unit can then identify the overall shape of the error (e.g., overall tilt, local bending), providing in-depth information input for intelligent collaborative control that matches the error pattern. During the comparison process, an overall quality factor, including matching confidence, data integrity, and error field smoothness, is calculated and output in real time. This mechanism enables the system to self-judge the reliability of a single comparison result. The quality factor can be fed back to the control unit for dynamic adjustment of control parameters or to trigger diagnostic processes when data quality is abnormal, thus realizing the transformation of the system from a fixed output to a reliable and evaluable output, enhancing the adaptability and reliability of the entire control closed loop. Through strict spatiotemporal synchronization, coordinate system unification, and elastic sub-pixel matching, the extracted longitudinal difference ensures that it truly and accurately reflects the vertical deviation of the suspension cable relative to the theoretical parabola, avoiding error distortion caused by data misalignment and lateral fluctuations. The comparison accuracy reaches the sub-pixel level, matching the accuracy of the front-end visual extraction stage, providing a clean and reliable deviation signal for subsequent high-precision closed-loop control. The elastic matching algorithm is tolerant of reasonable lateral fluctuations, while the quality feedback mechanism allows the system to adopt robust strategies when it senses a decline in data quality. The combination of these two mechanisms ensures that the system can continuously generate stable and reliable control data even when faced with common industrial environmental disturbances such as minor vibrations and airflow disturbances. This effectively avoids control malfunctions caused by instantaneous data anomalies and improves the overall anti-interference capability of the system. The provided error vector field and its spatial characteristic analysis enable the cooperative control algorithm to understand the distribution patterns and morphology of errors.The control strategy can be upgraded from simple error-driven to shape optimization-driven. For example, coordinated bias force adjustment can be implemented for the overall tilt pattern, and focused couple correction can be implemented for the local bulge pattern, thereby achieving smarter, smoother, and more efficient overall shape control.

[0087] Furthermore, in some embodiments of the present invention, in the parabolic adjustment module, the collaborative control unit relies on a deep understanding and precise modeling of the continuous flexible body coupled dynamics characteristics of the measuring wire when performing control command calculations. The measuring wire, as the core controlled object, is a typical one-dimensional flexible body with continuous mass distribution, dominated by internal tension, and exhibiting geometric nonlinearity and dynamic coupling characteristics. Unlike traditional adjustment methods that treat the wire as a discrete point mass or a rigid body, the present invention employs a flexible body dynamics processing framework that integrates parametric model order reduction, spatiotemporal coupling effect quantification, and adaptive boundary condition perception to achieve high-precision morphological collaborative control. This framework includes the following steps:

[0088] Step E1 involves parameterized dynamics order reduction based on a hybrid model. Addressing the problem that precise dynamic models of continuous flexible bodies (such as partial differential equations) are too complex and difficult to use for real-time control, this invention employs a hybrid lumped-distributed parameter model for order reduction. The main body shape of the steel wire in static or quasi-static conditions is modeled as a catenary or parabola uniquely determined by the positions of the two fixed points, the wire linear density, and the internal tension, forming the macroscopic framework of the shape. The deviation between the actual shape of the steel wire and the static theoretical shape (i.e., the error to be controlled) is represented as a linear superposition of a series of pre-calculated or online identified spatial modes (such as sinusoidal modes or characteristic functions based on system identification). In the control model, not only is the direct effect of each electromagnetic actuator on the steel wire point directly below it considered, but also the modal influence matrix is ​​used to quantify how the force of a single actuator excites or inhibits various modes. This matrix, obtained through offline finite element analysis or online frequency response identification, encapsulates the spatially coupled dynamic characteristics of local forces propagating through the steel wire to both ends and exciting specific overall shapes.

[0089] Step E2: Online quantification and prediction of spatiotemporal coupling effects. To achieve smoother and more precise control, this invention dynamically predicts and compensates for coupling effects in the control loop. For each electromagnetic actuator position, the vertical displacement response at different times along the wire axis after a unit step force is applied is obtained through simulation or experiment, forming a spatiotemporal Green's function database for the system. In each control cycle, the co-control unit, based on the current error vector sequence and the previous control output, uses the aforementioned influence functions to predict the overall deformation trend of the wire in a short future time domain. This predictive function enables the controller to anticipate the potential effects of the current control action coupled through flexible body coupling in the future, thereby making compensation in advance, effectively avoiding overshoot and oscillation, and achieving predictive feedforward control.

[0090] Step E3: Adaptive sensing integration of boundary conditions and tension state. The dynamic characteristics of the steel wire strongly depend on the boundary conditions (stiffness and damping of the constraints at both ends) and internal tension. In the lumped parameter model, the fixed constraints at both ends of the steel wire are equivalent to virtual spring-damped boundaries with specific stiffness and damping, rather than ideal absolute rigid constraints. During system initialization, these equivalent boundary parameters can be identified online by exciting the steel wire and analyzing its free decaying vibration frequency and damping ratio. A model-based state observer is developed. This observer takes the output force sequence of the electromagnetic actuator and the multi-point, multi-frequency vibration signal of the steel wire extracted through image sequence analysis as input to estimate the equivalent average tension inside the steel wire and its variation trend online. When tension drift caused by temperature changes or material creep is detected, the system automatically adjusts the tension parameters in the dynamic model and can fine-tune the target parameters of the theoretical parabola (because its shape is determined by tension) accordingly, ensuring that the control model always matches the real physical state of the controlled object.

[0091] In existing technologies, the adjustment of suspension cables often adopts simple proportional-integral-derivative (PID) control or multiple-input multiple-output (MIMO) decoupling control. The control models are usually based on rigid assumptions or extremely simplified coupling relationships, which cannot accurately describe and predict the inherent complex spatiotemporal coupling dynamics of flexible bodies. This leads to problems such as oscillation, slow convergence, or instability under disturbances during the adjustment process, making it difficult to achieve high-precision and smooth morphological approximation.

[0092] Compared to existing technologies, this invention achieves a leap in control strategy from geometric error feedback to physical model prediction and compensation by introducing and processing the coupled dynamic characteristics of continuous flexible bodies. First, a precise reduced-order description of complex continuum dynamics is achieved through a hybrid parameterized model, enabling real-time model predictive control. Second, spatiotemporal coupling prediction is performed using a pre-built influence function database, endowing the control system with feedforward compensation capabilities and significantly improving response speed and control stability. Finally, by integrating adaptive sensing of boundary conditions and tension states, the control system possesses robustness against environmental changes and its own state drift, ensuring control accuracy and stability over long-term operation. These three aspects combined constitute the core theoretical foundation for achieving high-precision, high-efficiency, and highly robust coordinated adjustment of the steel wire's shape.

[0093] Furthermore, in some embodiments of the present invention, in the parabolic adjustment module, the collaborative control unit implements collaborative control based on a lumped parameter model. The core of this control scheme is to transform the infinite-dimensional shape control problem of a continuous flexible body into a multi-input multi-output real-time optimization problem based on a finite-dimensional state-space model. This scheme transcends independent reactive compensation for local errors, constructing an intelligent decision-making center capable of anticipating global coupling effects and generating coordinated control commands accordingly. This is crucial for achieving smooth, accurate, and stable convergence of the wire shape. Specifically, it includes the following:

[0094] Step F1 involves constructing a lumped-parameter model for control. The system uses the coordinates of several dominant modes selected in the aforementioned hybrid model as the core state vector. The modes are pre-obtained through offline finite element analysis or experimental modal analysis, which can characterize the main variable forms of the steel wire with minimal error. The dimensionality of the state vector is far less than the number of electromagnetic actuators, thus achieving a significant reduction in order from a high-dimensional physical action space to a low-dimensional morphological description space. Based on this, the lumped-parameter state-space equations of the system are established. These equations use the modal coordinates and their derivatives as state variables, and the control forces of each actuator as inputs. The modal mass matrix, modal damping matrix, and modal stiffness matrix encapsulate the inertia, internal friction, and stiffness (including tension contribution) characteristics of the steel wire. A control influence matrix precisely quantifies the excitation efficiency of each actuator for each mode. This control influence matrix is ​​a concentrated representation of the spatial coupling dynamics of the steel wire. Finally, the system is expressed in a standard first-order state-space form, laying the foundation for applying optimization algorithms in modern control theory.

[0095] Step F2 involves collaborative optimization based on model predictive control. The collaborative control unit performs a model predictive control calculation once per control cycle. Its core is to solve an open-loop optimal control problem within a finite future time window. Within this framework, the system uses discretized state-space equations as the predictive model to simulate the future evolution trajectory of the wire's modal states under different control force sequences. The optimization process aims to minimize a carefully designed objective function, which is a weighted sum of tracking error, control energy, and control smoothing terms. The tracking error term represents the deviation between the predicted shape and the target parabolic shape in modal space, driving the wire to approximate its theoretical shape. The control energy term represents the amplitude of the control force, aiming to minimize actuator energy consumption and heat generation. The control smoothing term represents the rate of change of the control force, forcing the generation of smooth control commands, avoiding abrupt changes, and ensuring stable adjustment. Simultaneously, the optimization process strictly adheres to the output force amplitude constraints of each actuator.

[0096] In each control cycle, the system solves the constrained optimization problem described above, obtaining a series of optimal control forces in the future time domain. However, only the first optimal control force corresponding to the current moment is actually output to each actuator. In the next cycle, the system updates the state estimate based on the latest visual measurement data and performs a new round of rolling optimization and solution. This "rolling time domain, feedback correction" mechanism enables it to naturally handle multivariate coupling and physical constraints, and continuously output globally coordinated control commands.

[0097] Step F3, State Estimation and Parameter Adaptation: Since the system cannot directly measure modal coordinates, a reduced-dimensional state observer is designed. This observer takes the wire displacement measurements obtained from a few key locations by the vision system and the known current control force as inputs, and combines the aforementioned state-space model and measurement model to estimate the complete modal state vector (including modal coordinates and their rates of change) in real time, providing necessary feedback information for model predictive control.

[0098] In addition, the system establishes a slow adaptive outer loop to continuously monitor long-term control performance (such as steady-state error and control force statistical characteristics) and the residuals of the state observer. When continuous performance degradation due to slow changes in wire properties (such as material aging and tension drift) is detected, the outer loop can trigger online fine-tuning of key model parameters (such as equivalent damping and control influence coefficients) to ensure that the internal control model always matches the actual physical state of the controlled object, thereby ensuring control accuracy and robustness under long-term operation.

[0099] In existing technologies, independent closed-loop control or simple decoupling control are often used to adjust the suspension configuration using multiple actuators. The former cannot handle the strong coupling between actuators, which can easily lead to oscillations and "control snatching" phenomena; the latter relies on a precise and fixed coupling model, lacks consideration for changes in object characteristics and global optimization objectives, and is difficult to achieve fast, smooth, and energy-efficient optimal adjustment.

[0100] Compared to existing technologies, the cooperative control scheme based on a lumped parameter model described in this invention reduces the complex continuum control problem to a finite-dimensional optimization problem through state-space modeling and solves it using a model predictive control framework. This scheme offers the following advantages: First, it significantly reduces the dimensionality of the optimization problem through modal coordinates and embeds coupled dynamics within the control influence matrix, enabling the controller to fundamentally understand and handle the interactions between multiple actuators. Second, by optimizing future control sequences within a finite time domain, the model predictive control framework can explicitly handle actuator constraints and actively balance tracking accuracy, control energy consumption, and motion smoothness, thereby achieving fast, stable, and efficient morphological convergence. Finally, by combining a state observer with an online fine-tuning mechanism for model parameters, the system can reliably estimate all states with partial measurements and adapt to slow changes in the controlled object's characteristics, ensuring long-term control robustness and accuracy retention.

[0101] Furthermore, in some embodiments of the present invention, the electromagnetic force actuator system in the parabolic adjustment module employs a distributed vertical electromagnetic force application method. This actuator system is not a traditional rigid mechanical adjustment mechanism, but rather uses a vertical electromagnetic force generation and application network with high dynamic response, precise force control capability, non-contact or micro-contact application mode, and distributed collaborative characteristics. It decomposes macroscopic shape control into multiple microscopic, independently programmable force application units, and organizes these units into an intelligent force field capable of shaping a flexible steel wire through a collaborative control algorithm. It includes the following steps:

[0102] Step G1 employs a dual-mode magnetic circuit topology and integrated multi-dimensional sensing and closed-loop force control. In the dual-mode magnetic circuit topology, each electromagnetic actuator uses a hybrid magnetic circuit design combining an E-type electromagnet with a permanent magnet bias. The central post of the E-type core is perpendicular to the steel wire, while the two side posts are coupled to the permanent magnet. At zero control current, the permanent magnet provides a constant static bias force (e.g., attraction), keeping the steel wire essentially taut without active adjustment and suppressing low-frequency oscillations. The control coil is wound around the central post; by applying forward / reverse current, a bidirectional adjustable dynamic electromagnetic force can be superimposed on the bias force, achieving precise bidirectional control of tension and pressure. This design allows bidirectional force output to be achieved with a unipolar drive circuit, simplifying power electronics design. In this integrated multidimensional sensing and closed-loop force control system, each actuator module integrates three sensing units: a high-linearity Hall current sensor for real-time monitoring of coil current; a miniature laser triangular displacement sensor for precise measurement of the instantaneous air gap between the core pole face and the wire surface; and a patch-type temperature sensor for monitoring coil temperature rise. The control unit, based on a precise electromagnetic force model, uses current and air gap as primary feedback. Through high-speed digital proportional-integral-derivative control combined with a feedforward compensation algorithm, it achieves closed-loop precise control of the output force of each actuator. Its dynamic force control accuracy can reach ±0.5% of full scale, with a response bandwidth exceeding 100Hz.

[0103] Step G2 employs an adaptive flexible contact and vertical force guiding mechanism. In the passive compliance and active collision avoidance mechanism, the actuator end-effector is connected to the core via a two-degree-of-freedom passive compliance mechanism, consisting of a radial flexible hinge and an axial preload spring. This mechanism allows the actuator end-effector to float within a certain range radially (in a plane perpendicular to the wire axis and vertically) to accommodate minor lateral fluctuations in the wire, ensuring that the force direction always remains automatically vertical and avoiding the introduction of harmful lateral friction. Simultaneously, the axial spring preload ensures that the actuator end-effector moves accordingly and maintains slight contact in the event of unexpected large vertical jumps in the wire, avoiding rigid impacts. In the contact state monitoring and protection based on air gap feedback, the system monitors the air gap value of each actuator in real time. During normal operation, the air gap is maintained at a set value, representing a "micro-contact" state. When the air gap abnormally increases or decreases sharply, the control algorithm immediately switches to protection mode, reducing force output or performing rapid buffering to prevent damage to the wire or actuator. This design enables the system to adapt to forced vibrations and transient disturbances in the wire.

[0104] Step G3 employs a distributed layout and thermal-mechanical coupling management. In the variable-density non-uniform layout strategy, actuators are not installed at strictly equal intervals, but rather using a non-uniform layout scheme based on the theoretical parabolic curvature variation. In areas with greater parabolic curvature, the actuator density is higher to provide stronger local shape adjustment capabilities; in the middle section with less curvature, the density is lower. This optimized layout based on the geometric characteristics of the controlled object minimizes the number of actuators and reduces cost and power consumption while ensuring control performance. In thermal management collaborative control, the drive circuits of all actuators are integrated on a water-cooled heat dissipation substrate. The system monitors the temperature and current of each actuator and implements a dynamic thermal equilibrium strategy by fine-tuning the weight matrix R of the control energy term in the collaborative control algorithm online. When the temperature of an actuator is too high, the system can appropriately reduce its control weight while meeting shape control performance requirements, transferring some control tasks to adjacent actuators with lower temperatures, achieving joint optimization of control load and thermal load, and ensuring long-term continuous and stable operation of the system.

[0105] In the existing technology, the actuators used to adjust the shape of slender flexible bodies are mostly rigid screw pushers or simple electromagnetic attraction devices. The former has problems such as gaps, friction, slow response and possible damage to the object, while the latter can often only provide unidirectional force and has insufficient control accuracy and coordination.

[0106] Compared to existing technologies, the distributed vertical electromagnetic force actuator system described in this invention has the following advantages: First, through a hybrid magnetic circuit and closed-loop force control design, it achieves bidirectional, high-precision, and fast-response vertical force application to the steel wire, providing an ideal physical execution terminal for precise morphological control. Second, the adaptive compliant mechanism and air-gap-based contact protection mechanism enable the system to operate safely and compliantly in non-contact or micro-contact states, effectively avoiding damage to the high-precision measuring steel wire and adapting to its dynamic fluctuations. Finally, through curvature-based optimized layout and thermal-mechanical collaborative management, while ensuring global control capabilities, it optimizes system resources (number of actuators, energy consumption, heat distribution), improving the reliability, economy, and long-term operational stability of the entire actuator network. This actuator system, closely integrated with the aforementioned visual perception, intelligent comparison, and model predictive controller, together constitutes a complete, efficient, and robust closed-loop system for precise suspension morphological adjustment.

[0107] It should be noted that technical features that are not fully explained will be addressed using conventional technical methods.

[0108] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described herein. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the specific embodiments described above. Therefore, any modifications or equivalent substitutions to the present invention, as well as all technical solutions and improvements that do not depart from the spirit and scope of the invention, are covered within the scope of the claims of the present invention.

Claims

1. An adjustment method for flatness measurement based on a suspension parabola, which is implemented using a pixel adjustment module and a parabola adjustment module, characterized in that, The pixel adjustment module acquires at least an image including the steel wire, performs perspective correction and distortion correction on the acquired image, and maps the steel wire image in the image to a virtual correction plane parallel to the theoretical plane of the steel wire through homography transformation. In the coordinate system of the correction plane, a linear mapping relationship between the image pixel coordinates and the actual physical coordinates is established to ensure that the actual physical size corresponding to each pixel on the measurement plane is constant. Subsequently, the precise center line of the steel wire in the corrected image is extracted as its current actual projection trajectory. The parabola adjustment module includes a collaborative control unit and several electromagnetic force actuators evenly distributed along the axis of the measuring steel wire. It extracts the actual projected trajectory of the steel wire and generates a corresponding theoretical parabola based on the length and trajectory of the steel wire. It calculates the difference in the vertical coordinates between the actual projected trajectory and the theoretical parabola at the corresponding horizontal coordinate positions. The control unit receives the sequence of vertical coordinate differences and converts them into control signals with positive and negative directions. After the control unit solves the problem, it outputs a set of coordinated control commands to each electromagnetic force actuator, so that the overall shape of the steel wire smoothly and stably approaches the preset theoretical parabolic shape. The pixel adjustment module and the parabolic adjustment module work together at a fixed cycle, sequentially executing a closed-loop process of image acquisition, perspective correction, trajectory extraction, difference calculation, and collaborative control output; until the parabolic adjustment module determines that the deviation of the actual projected trajectory of the steel wire from the theoretical parabolic baseline at each detection point is less than the set threshold, the system then performs subsequent flatness detection and calculation.

2. The adjustment method for flatness measurement based on a suspension parabola according to claim 1, characterized in that, When performing perspective correction and distortion correction, the pixel adjustment module adopts a calibration method based on a three-dimensional target. The three-dimensional target includes multiple marker points with known three-dimensional coordinates. By capturing a single shot, the internal and external parameters of the camera and the lens distortion coefficient are obtained simultaneously, and the homography transformation matrix from the image pixel coordinates to the physical coordinates of the wire theoretical plane is directly calculated and generated.

3. The adjustment method for flatness measurement based on a suspension parabola according to claim 1, characterized in that, The pixel adjustment module also includes an embedded real-time dynamic calibration network, which consists of multiple fixed-mount intelligent reference units. Each unit includes a uniquely coded micro-calibration pattern and an attitude sensor, used to monitor and compensate for reference surface distortion caused by environmental vibration, temperature changes or mounting surface deformation in real time.

4. The adjustment method for flatness measurement based on a suspension parabola according to claim 1, characterized in that, Perspective correction and distortion correction employ a layered, progressive structure: The first layer is dynamic perspective correction based on feature tracking and pose estimation. By tracking stable auxiliary visual features within the field of view, the camera pose is estimated in real time and the homography transformation matrix is ​​dynamically updated. The second layer is based on the online estimation and compensation of residual distortion based on the geometric characteristics of the wire image itself. A residual distortion field model is established by fitting the deviation of the wire centerline, and pixel-level fine-tuning compensation is performed.

5. The adjustment method for flatness measurement based on a suspension parabola according to claim 1, characterized in that, A linear mapping relationship between image pixel coordinates and actual physical coordinates is established using a hierarchical weighted fusion and dynamic self-calibration method, including: Based on the calculation of hierarchical mapping coefficients with region confidence weighting, the corrected image is divided into multiple sub-regions, a local affine transformation model is fitted in each region, and the confidence weight is determined based on the fitting residual. Dynamic self-calibration of mapping coefficients based on closed-loop feedback monitors scale error and dynamically adjusts mapping coefficients by identifying a fixed microscale reference scale in the image to compensate for scale changes caused by camera thermal drift.

6. The adjustment method for flatness measurement based on a suspension parabola according to claim 1, characterized in that, The precise centerline of the steel wire in the corrected image was extracted using a fusion extraction method based on multi-scale linear enhancement and grayscale profile iterative modeling. The first stage is based on the linear response of the multi-scale Hessian matrix to perform robust initial localization and generate a set of pixel-level center points. The second level is based on the physical model fitting of the grayscale profile for sub-pixel refinement. The grayscale profile is fitted by a Gaussian model, and the sub-pixel level offset of the center point is calculated.

7. The adjustment method for flatness measurement based on a suspension parabola according to claim 1, characterized in that, The parabola adjustment module also includes a comparison and difference sequence generation process between the actual trajectory and the theoretical parabola benchmark. It employs a spatiotemporal registration-elastic matching-vectorized difference process, comprising the following steps: Step D1: Unify the timestamp and coordinate system, transforming the actual trajectory points to be consistent with the control coordinate system; Step D2: Use an elastic matching algorithm based on dynamic time warping to match the theoretical point for each actual point and calculate the sub-pixel level ordinate difference; Step D3: Aggregate the differences according to the actuator position to form an error vector sequence, and construct an error vector field for spatial analysis.

Images