Bridge installation positioning and guiding fine control system and method based on machine vision

Abstract

The application provides a bridge installation positioning guide fine control system and method based on machine vision, the system comprises a birefringent photoelastic cooperation target, a polarization-macroscopic double optical path vision imaging terminal, an FPGA time sequence controller, an upper computer industrial computer system and a three-dimensional hydraulic jack cluster, the method alternately collects natural environment light geometric images and polarization interference fringe images through the FPGA time sequence controller, extracts a full-field two-dimensional strain tensor to perform local elastic displacement field integral calculation to compensate for a characteristic point physical offset, uses a nonlinear optimization algorithm to reconstruct a real rigid body pose matrix of a beam section, maps a two-dimensional fringe density gradient matrix to a physical distribution topology space, and attenuates a local virtual stiffness coefficient in a local second-order rigid-flexible coupling guide model according to a local gradient integral value, so that the application eliminates spatial positioning deviation caused by elastic distortion, realizes adaptive flexible retreat of the three-dimensional hydraulic jack cluster, and ensures the safety of the closure operation.

Description

Technical Field

[0001] This invention relates to the field of bridge construction monitoring and automated installation technology, specifically to a precision control system and method for bridge installation positioning guidance based on machine vision. Background Technology

[0002] During the installation and closure phases of large bridge components, machine vision systems are typically required for spatial positioning and guidance. Existing visual positioning technologies are generally based on the assumption of an ideal rigid body, which means that the positioning target does not deform during assembly. However, in actual bridge assembly conditions, due to the extremely large self-weight of the components and the complex stress boundaries, the positioning target placed in the stress area of ​​the component is prone to local elastic deformation along with the bridge base. This material deformation will cause uncontrollable physical offset of the spatial feature points on the target, making the traditional visual rigid body projection calculation model fail, thus introducing significant spatial solution deviations and severely reducing the spatial positioning accuracy of bridge components under complex stress conditions.

[0003] To simultaneously acquire the stress state and pose information of components, existing systems often require the introduction of multimodal image acquisition equipment. Existing dual-modal or multimodal vision acquisition systems mostly adopt software-level synchronous or decoupled asynchronous triggering mechanisms. During the dynamic assembly of bridge beam segments, due to the slight vibration and continuous spatial displacement of components, asynchronous acquisition will cause the geometric pose image acquired under natural ambient light and the stress interference image acquired under polarized light to be difficult to strictly align in terms of timestamps. This misalignment of timestamps can easily cause motion blur in the image, and cannot guarantee the reliability of multimodal data in subsequent cross-domain fusion and mapping calculations.

[0004] Furthermore, in the underlying execution stage of bridge closure, existing hydraulic jack cluster control systems mostly adopt rigid position control models with fixed parameters. During multi-point collaborative advancement operations, local rigid contact often occurs between the beam segment to be assembled and the fixed pier segment due to attitude errors. Since the traditional control framework lacks an online adaptive adjustment mechanism based on interface stress, when a local rigid collision causes a surge in stress, the hydraulic control nodes cannot actively reduce the system stiffness and produce flexible yielding. This not only prevents the concentrated stress at the closure interface from being effectively released, but also easily causes damage to the mechanical actuator due to local overload, posing a safety hazard to the bridge assembly project. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a precision control system and method for bridge installation positioning guidance based on machine vision. This solves the problem that elastic deformation caused by local stress concentration during the installation and closure phase of large bridge components leads to the failure of the rigid body assumption in traditional visual positioning algorithms, resulting in decreased positioning accuracy and overload damage to the execution end.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The first aspect of this invention provides a machine vision-based bridge installation positioning guidance precision control system, comprising: a birefringent photoelastic cooperative target, set in the stress area of ​​the bridge beam segment to be assembled; a polarization-macroscopic dual-optical-path vision imaging terminal, set on a reference datum surface of a fixed bridge pier segment and aligned with the birefringent photoelastic cooperative target; an FPGA timing controller, electrically connected to the polarization-macroscopic dual-optical-path vision imaging terminal, for outputting a hardware synchronization trigger signal to the polarization-macroscopic dual-optical-path vision imaging terminal; a three-dimensional hydraulic jack cluster, with its mechanical actuators supported at the bottom of the bridge beam segment to be assembled; and a host computer industrial control system, communicatively connected to both the polarization-macroscopic dual-optical-path vision imaging terminal and the three-dimensional hydraulic jack cluster, for receiving image data and executing data processing and control logic, as well as issuing displacement control commands to the three-dimensional hydraulic jack cluster.

[0008] Preferably, the polarization-macroscopic dual-optical-path visual imaging terminal includes: an active circularly polarized light source module, arranged on the front light-transmitting side of the polarization-macroscopic dual-optical-path visual imaging terminal, for emitting circularly polarized light of a preset wavelength to the birefringent photoelastic cooperative target; a beam splitter prism, disposed in the internal optical path of the polarization-macroscopic dual-optical-path visual imaging terminal and located in the reflection and return optical path of the birefringent photoelastic cooperative target, for separating the reflected and return light into a transmitted beam propagating along a first direction and a reflected beam propagating along a second direction; and a main camera, disposed at the output of the transmitted beam from the beam splitter prism. In the optical path, a geometric image of the natural ambient light of the birefringent photoelastic cooperative target is acquired when the active circularly polarized light source module is off; a circularly polarized analyzer is set in the output optical path of the reflected beam of the beam splitter, and the polarization filtering direction is orthogonal to the polarization state of the active circularly polarized light source module, used to filter out the background ambient light and transmit the birefringent polarization interference signal light; a secondary camera is set in the output optical path of the circularly polarized analyzer on the side away from the beam splitter, used to acquire the polarization interference fringe image transmitted through the circularly polarized analyzer when the active circularly polarized light source module is on.

[0009] Preferably, the birefringent photoelastic cooperative target includes: a reflective substrate, bonded to the surface of the stress area of ​​the bridge beam segment to be assembled, and made of epoxy resin binder doped with high reflectivity metal particles, thereby forming a diffuse reflection substrate on the surface of the bridge beam segment to be assembled; a photoelastic coating, layered and laid on the outer side of the reflective substrate away from the surface of the bridge beam segment to be assembled, the material being an epoxy resin-based birefringent polymer, having a fixed physical thickness and a calibrated strain fringe constant; and a spatially encoded pattern, attached to the surface of the photoelastic coating away from the reflective substrate, consisting of multiple asymmetrically distributed geometric primitives, which form a set of feature points for machine vision spatial positioning under natural ambient light.

[0010] The second aspect of this invention provides a machine vision-based fine control method for bridge installation positioning guidance, applied to the aforementioned machine vision-based fine control system for bridge installation positioning guidance, comprising the following steps: after system power-on initialization, the FPGA timing controller outputs a time-division multiplexed trigger signal to the polarization-macroscopic dual-optical-path visual imaging terminal; within odd-numbered frame triggering periods, the polarization-macroscopic dual-optical-path visual imaging terminal acquires a geometric image of the natural ambient light of a birefringent photoelastic cooperative target; within even-numbered frame triggering periods, the polarization-macroscopic dual-optical-path visual imaging terminal activates its internal active circularly polarized light source module. The system acquires polarization interference fringe images of a birefringent photoelastic cooperative target; a polarization-macroscopic dual-optical-path visual imaging terminal sends the ambient light geometry image and the polarization interference fringe image to a host computer system in the form of a data stream; the host computer system extracts isochromatic and isocliptural feature parameters from the polarization interference fringe image; combining the optical and mechanical constants of the birefringent photoelastic cooperative target, the host computer system converts the isochromatic and isocliptural feature parameters into a full-field two-dimensional strain tensor of the target region; the host computer system then applies the full-field two-dimensional strain tensor to the ambient light geometry... The coordinates of feature points in the image are used to perform local elastic displacement field integral calculations. The host computer system compensates for the physical offset of feature points caused by material deformation based on the integral calculation results, and uses a nonlinear optimization algorithm to reconstruct the true rigid body pose matrix of the bridge beam segment to be assembled. The host computer system constructs an admittance control model based on the true rigid body pose matrix and the preset target pose, calculating the position deviation and virtual restoring force. The host computer system also calculates the two-dimensional fringe density gradient matrix of the polarization interference fringe image and maps it to the physical coordinates corresponding to the three-dimensional hydraulic jack cluster. The topological space is distributed; the host computer system adjusts the local virtual stiffness coefficients in the admittance control model online based on the local gradient integral values ​​in the physical topological space. When the gradient integral value of the corresponding region increases, the local virtual stiffness coefficient of the corresponding control node decreases; the host computer system generates displacement control commands based on the admittance control model containing the local virtual stiffness coefficients and sends the displacement control commands to the three-dimensional hydraulic jack cluster; the three-dimensional hydraulic jack cluster drives the actuator to generate spatial displacement and non-uniform retreat according to the displacement control commands, and completes the closure positioning of the bridge beam segment to be assembled.

[0011] Preferably, the process of outputting the time-division multiplexing trigger signal to send the image in the form of a data stream specifically includes: the FPGA timing controller generating a high-frequency square wave control signal, and the FPGA timing controller outputting the high-frequency square wave control signal as a time-division multiplexing trigger signal to the polarization-macroscopic dual-optical-path visual imaging terminal; during the odd-numbered frame triggering period, the FPGA timing controller controls the active circularly polarized light source module to be in a power-off state through the high-frequency square wave control signal, and the FPGA timing controller triggers the global shutter of the main camera for synchronous exposure, and controls the secondary camera to be in a stopped-exposure state, so that the main camera acquires the spatial coding pattern of the birefringent photoelastic cooperative target under natural ambient light, generating a geometric image of natural ambient light; during the even-numbered frame triggering period, the FPGA timing controller controls the active circularly polarized light source module to be in a power-off state through the high-frequency square wave control signal. When powered on, the active circularly polarized light source module emits circularly polarized light of a preset wavelength towards the birefringent photoelastic cooperative target. The FPGA timing controller triggers the global shutter of the secondary camera for synchronous exposure and controls the primary camera to be in a stopped-exposure state. This allows the secondary camera to acquire the optical signal transmitted through the circularly polarized analyzer and generate a polarization interference fringe image. The polarization-macroscopic dual-optical-path visual imaging terminal alternately arranges the geometric image of natural ambient light and the polarization interference fringe image according to the triggering sequence of the high-frequency square wave control signal, generating a multimodal image data stream. The polarization-macroscopic dual-optical-path visual imaging terminal sends the multimodal image data stream to the host industrial control computer system. Through the alternating triggering of the hardware-level high-frequency square wave control signal, the timestamps of the geometric features of natural ambient light and the polarization features of optical stress are strictly aligned in the same physical space coordinates.

[0012] Preferably, the process of extracting isochromatic line feature parameters and isoclinal feature parameters from a polarization interference fringe image and converting them into a full-field two-dimensional strain tensor for the target region specifically includes: the host computer / industrial control system receiving the polarization interference fringe image and performing digital image processing on it to extract the isochromatic line fringe order distribution and the isoclinal principal stress direction angle distribution; the host computer / industrial control system reading the pre-calibrated strain fringe constant and photoelastic coating thickness of the birefringent photoelastic cooperative target, and calculating the difference between the first principal strain and the second principal strain at any pixel point within the target region by dividing the product of the isochromatic line fringe order distribution and the strain fringe constant by the photoelastic coating thickness according to the laws of strain optics; the host computer / industrial control system combining elasticity mechanics... The compatibility equation is numerically integrated over the difference between the first and second principal strains to calculate the sum of the first and second principal strains. Independent values ​​for the first and second principal strains are then separated from each other using the difference and the sum. The host computer system performs a two-dimensional tensor coordinate transformation based on the independent first and second principal strain values ​​and the angular distribution of the principal stress directions along the isochoric lines. This generates a full-field two-dimensional strain tensor for the target region in Cartesian coordinates. This full-field two-dimensional strain tensor includes X-axis normal strain components, Y-axis normal strain components, and XY-plane shear strain components, achieving cross-domain quantization conversion of optical interference pattern parameters to underlying structural mechanical strain tensor parameters.

[0013] Preferably, the process of performing local elastic displacement field integral calculation based on the coordinates of feature points in the natural ambient light geometric image using the full-field two-dimensional strain tensor to reconstruct the true rigid body pose matrix of the bridge beam segment to be assembled using a nonlinear optimization algorithm specifically includes: the host computer system identifying the unstressed boundary points or physical geometric centers of the birefringent photoelastic cooperative target in the natural ambient light geometric image, and setting the unstressed boundary points or physical geometric centers as the integration reference origin for the local elastic displacement field integral calculation; and ensuring that the photoelastic coating of the birefringent photoelastic cooperative target satisfies the thin film physical boundary conditions. Under these conditions, the host computer system sets the normal deformation of the birefringent photoelastic target surface as a high-order infinitesimal quantity and ignores it, retaining only the in-plane deformation component. The host computer system extracts multiple geometric feature points for spatial positioning from the ambient light geometric image and plans a spatial integration path from the integration reference origin to each geometric feature point in the full-field two-dimensional strain tensor. Line integration is performed on the full-field two-dimensional strain tensor along the spatial integration path to calculate the local elastic deformation displacement vector of each geometric feature point caused by elastic stress tension or compression. The host computer system extracts the three-dimensional nominal coordinates of multiple geometric feature points on the birefringent photoelastic cooperative target in an ideal rigid body state without force. It then adds these three-dimensional nominal coordinates to the local elastic deformation displacement vector to obtain the actual three-dimensional physical coordinates of the geometric feature points after deformation under force. The host computer system also extracts the two-dimensional pixel observation coordinates of the corresponding geometric feature points from the natural ambient light geometric image. Based on the actual three-dimensional physical coordinates and the two-dimensional pixel observation coordinates, it constructs a reprojection error objective function incorporating a deformation compensation term. The Levenberg-Marquardt nonlinear least squares algorithm is used to iteratively solve the reprojection error objective function, searching for the rotation matrix and translation vector that minimize the reprojection error objective function. The rotation matrix and translation vector at the minimum value are then combined to form the true rigid body pose matrix of the bridge beam segment to be assembled. By introducing the local elastic deformation displacement vector into the visual reprojection calculation model in the form of feedforward compensation parameters, the physical offset of the spatially encoded feature points under force is compensated, fundamentally eliminating the spatial calculation deviation caused by the elastic distortion of the target itself in the overall rigid body pose reconstruction process.

[0014] Preferably, the process of constructing an admittance control model based on the real rigid body pose matrix and the preset target pose, and then calculating the position deviation and virtual restoring force, specifically includes: performing a six-degree-of-freedom spatial difference operation on the real rigid body pose matrix and the target pose matrix of the bridge beam segment to be assembled, and calculating a six-dimensional geometric deviation vector containing three translational degrees of freedom deviations and three rotational degrees of freedom deviations; retrieving the pre-configured position loop proportional gain matrix and position loop differential gain matrix, and adding the product obtained by multiplying the six-dimensional geometric deviation vector by the position loop proportional gain matrix to the product obtained by multiplying the first-order time derivative of the six-dimensional geometric deviation vector by the position loop differential gain matrix. The product of the two is used to calculate the virtual restoring force vector that drives the three-dimensional hydraulic jack cluster to approach the target pose of the bridge beam segment to be assembled. In the underlying hydraulic servo control framework, a second-order rigid-flexible coupled admittance control model containing a virtual mass matrix, a virtual damping matrix, and a virtual stiffness matrix is ​​constructed. In the second-order rigid-flexible coupled admittance control model, the product of the second-order time derivative of the hydraulic system's end displacement correction and the virtual mass matrix, plus the product of the first-order time derivative of the hydraulic system's end displacement correction and the virtual damping matrix, plus the product of the hydraulic system's end displacement correction and the virtual stiffness matrix, is numerically equal to the virtual restoring force vector.

[0015] Preferably, the process of mapping the two-dimensional stripe density gradient matrix to the physical distribution topology space corresponding to the three-dimensional hydraulic jack cluster and generating displacement control commands to be issued to the three-dimensional hydraulic jack cluster specifically includes: constructing a homogeneous transformation matrix between the two-dimensional image pixel coordinate system where the birefringent photoelastic cooperative target is located and the physical space coordinate system of the bottom of the bridge beam segment to be assembled; mapping the image pixel region in the two-dimensional stripe density gradient matrix to the physical force influence surface corresponding to each control node in the three-dimensional hydraulic jack cluster according to the homogeneous transformation matrix; dividing the two-dimensional stripe density gradient matrix into multiple local stress gradient sub-matrices corresponding one-to-one with the control nodes of the three-dimensional hydraulic jack cluster; performing area integration on the local stress gradient sub-matrices corresponding to each control node of the three-dimensional hydraulic jack cluster in the physical distribution topology space to calculate the local gradient integral value of the corresponding control node; and the host computer system performing second-order rigid-flexible coupling admittance control based on the local gradient integral value. The diagonal elements of the virtual stiffness matrix in the model undergo online attenuation updates, and these diagonal elements represent the local virtual stiffness coefficients of the corresponding control nodes. The local virtual stiffness coefficients of the corresponding control nodes are reduced according to the dynamic update mathematical expression of the local virtual stiffness coefficients. The value of the local virtual stiffness coefficient is equal to the initial stiffness setting threshold minus the product of the stiffness attenuation gain coefficient and the local gradient integral value. The reduced local virtual stiffness coefficient is then substituted into the second-order rigid-flexible coupling admittance control model to calculate the numerically increased end-effector displacement correction. This end-effector displacement correction is converted into a displacement control command and sent to the three-dimensional hydraulic jack cluster. By utilizing the optical interference distribution characteristics to dynamically adjust the stiffness parameters in the control model online, the three-dimensional hydraulic jack cluster can automatically reduce virtual stiffness and generate adaptive flexible yielding actions when encountering a stress surge caused by a local rigid collision during multi-point collaborative operation. This ensures the safety and reliability of the bridge assembly and closure operation and stress release.

[0016] This invention provides a precision control system and method for bridge installation positioning guidance based on machine vision. It has the following beneficial effects:

[0017] 1. This invention extracts the full-field two-dimensional strain tensor from polarization interference fringe images and calculates the local elastic deformation displacement vectors of each geometric feature point along the spatial integration path. These local elastic deformation displacement vectors are then used as feedforward compensation parameters in the iterative solution of the reprojection error objective function. This design quantitatively compensates for the offset of spatially encoded feature points under stress at a fundamental physical and mechanical level, eliminating the calculation deviation caused by the elastic distortion of the target itself in the overall rigid body pose reconstruction process, and improving the spatial positioning accuracy of large bridge beam segments under complex stress conditions.

[0018] 2. This invention uses an FPGA timing controller to output a high-frequency square wave control signal, alternately controlling the on / off state of the active circularly polarized light source module within odd- and even-frame trigger cycles, and simultaneously triggering the global shutters of the corresponding main and secondary cameras for exposure. This time-division multiplexing dual-modal image synchronous acquisition mechanism ensures strict alignment of the timestamps of the geometric features of natural ambient light and the polarization features of optical stress at the hardware level, overcoming the image motion blur problem during the dynamic assembly of large bridges and ensuring the reliability of the multimodal data stream in subsequent mapping and solving.

[0019] 3. This invention maps the calculated two-dimensional stripe density gradient matrix forward to the physical distribution topology of a three-dimensional hydraulic jack cluster, and based on the local gradient integral values ​​corresponding to each control node, attenuates the local virtual stiffness coefficients in the second-order rigid-flexible coupling admittance control model online. This control logic enables the three-dimensional hydraulic jack cluster to automatically reduce the stiffness of the control model and generate adaptive flexible yielding when encountering stress concentration caused by local rigid contact during multi-point collaborative operation, thereby realizing stress release at the closure interface and preventing damage to the mechanical actuator due to overload. Attached Figure Description

[0020] Figure 1 This is a system architecture diagram of the present invention;

[0021] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see the appendix Figure 1-2 The present invention provides a precision control system and method for bridge installation positioning guidance based on machine vision, including a birefringent photoelastic cooperative target, a polarization-macro dual-optical-path visual imaging terminal, an FPGA timing controller, a host computer industrial control system, and a three-dimensional hydraulic jack cluster.

[0024] The birefringent photoelastic cooperative target is set in the stress area of ​​the bridge beam segment to be assembled.

[0025] The polarization-macroscopic dual-optical-path visual imaging terminal is set on the reference reference surface of the fixed bridge pier section, and the polarization-macroscopic dual-optical-path visual imaging terminal is aligned with the birefringent photoelastic cooperative target.

[0026] The FPGA timing controller is electrically connected to the polarization-macro dual-optical-path visual imaging terminal and is used to output a hardware synchronization trigger signal to the polarization-macro dual-optical-path visual imaging terminal.

[0027] The host computer industrial control system communicates with the polarization-macro dual-optical-path vision imaging terminal to receive image data and execute data processing and control logic.

[0028] The communication input end of the three-dimensional hydraulic jack cluster is connected to the host computer industrial control system, and the mechanical execution end of the three-dimensional hydraulic jack cluster is supported at the bottom of the bridge beam segment to be assembled.

[0029] Based on the aforementioned system architecture, the macro-level workflow includes the following processing steps:

[0030] S101, after the system is powered on and initialized, the timing controller outputs a time-division multiplexing trigger signal to the polarization-macro dual-optical-path visual imaging terminal.

[0031] S102, within the odd-numbered frame triggering period, the polarization-macroscopic dual-optical-path visual imaging terminal acquires the natural ambient light geometric image of the birefringent photoelastic cooperative target.

[0032] S103, during the even-numbered frame triggering period, the polarization-macroscopic dual-optical-path visual imaging terminal activates its internal active circular polarization light source module to acquire polarization interference fringe images of the birefringent photoelastic cooperative target.

[0033] The S104 polarization-macroscopic dual-optical-path visual imaging terminal sends the geometric image of natural ambient light and the polarization interference fringe image to the host computer system in the form of a data stream.

[0034] S105, the host computer / industrial control system extracts isochromatic line feature parameters and isoclimatic line feature parameters from the polarization interference fringe image.

[0035] S106, combining the optical and mechanical constants of the birefringent photoelastic cooperative target, the host computer industrial control system converts the isochromatic line characteristic parameters and isoclinal line characteristic parameters into the full-field two-dimensional strain tensor of the target region.

[0036] S107, the host computer system performs local elastic displacement field integral calculation based on the coordinates of feature points in the geometric image of natural ambient light using the full-field two-dimensional strain tensor.

[0037] S108, the host computer system compensates for the physical offset of feature points caused by material deformation based on the integral calculation results, and uses a nonlinear optimization algorithm to reconstruct the real rigid body pose matrix of the bridge beam segment to be assembled.

[0038] S109, the host computer system constructs an admittance control model based on the real rigid body pose matrix and the preset target pose, and calculates the position deviation and virtual restoring force.

[0039] S110, the host computer industrial control system calculates the two-dimensional fringe density gradient matrix of the polarization interference fringe image, and maps the two-dimensional fringe density gradient matrix to the physical distribution topology space corresponding to the three-dimensional hydraulic jack cluster.

[0040] S111, the host computer system adjusts the local virtual stiffness coefficient in the admittance control model online based on the local gradient integral value in the physical distribution topology space. When the gradient integral value in the corresponding region increases, the local virtual stiffness coefficient of the corresponding control node decreases.

[0041] S112, the host computer system generates displacement control commands based on the admittance control model including local virtual stiffness coefficients, and sends the displacement control commands to the three-dimensional hydraulic jack cluster.

[0042] S113, the three-dimensional hydraulic jack cluster drives the execution end to generate spatial displacement and non-uniform retreat according to the displacement control command, and completes the closure positioning of the bridge beam segment to be assembled.

[0043] Regarding the birefringent photoelastic cooperative target mentioned in the aforementioned overall architecture, the birefringent photoelastic cooperative target specifically includes a reflective sublayer to a spatially encoded pattern.

[0044] The birefringent photoelastic cooperative target includes a reflective substrate, a photoelastic coating, and a spatially encoded pattern.

[0045] The reflective substrate is bonded to the surface of the stress area of ​​the bridge beam segment to be assembled. The reflective substrate is made of epoxy resin binder doped with high reflectivity metal particles, thereby forming a diffuse reflection substrate on the surface of the bridge beam segment to be assembled, which is used to provide a reflection return channel for polarized light paths.

[0046] The photoelastic coating is layered and laid on the outer side of the reflective substrate away from the surface of the bridge beam segment to be assembled. The photoelastic coating is made of epoxy resin-based birefringent polymer. The photoelastic coating has a fixed physical thickness and a calibrated strain fringe constant, so that the photoelastic coating generates internal strain synchronously when the bridge beam segment to be assembled undergoes elastic deformation, and generates polarization interference fringes corresponding to the principal stress difference in the polarized light field.

[0047] The spatial coding pattern is attached to the surface of the photoelastic coating away from the reflective substrate. The spatial coding pattern is composed of multiple asymmetrically distributed geometric primitives. The geometric primitives are solidified on the surface of the photoelastic coating through micro-etching or high-contrast screen printing. Under natural ambient light, the geometric primitives form a set of feature points for machine vision spatial positioning, so that the spatial coding pattern as a whole undergoes synchronous physical displacement with the elastic deformation of the photoelastic coating, and is used to output visual observation coordinates coupled with the structural strain state.

[0048] Regarding the polarization-macroscopic dual-optical-path visual imaging terminal described in the aforementioned overall architecture, the polarization-macroscopic dual-optical-path visual imaging terminal specifically includes an active circularly polarized light source module and a secondary camera.

[0049] The polarization-macroscopic dual-optical-path visual imaging terminal includes an active circularly polarized light source module, a beam splitter prism, a main camera, a circularly polarized analyzer, and a secondary camera.

[0050] An active circularly polarized light source module is arranged on the front light-transmitting side of the polarization-macro dual-optical-path visual imaging terminal, and the emission optical axis of the active circularly polarized light source module is aligned with the birefringent photoelastic cooperative target, which is used to emit circularly polarized light of a preset wavelength to the birefringent photoelastic cooperative target.

[0051] The beam splitter is positioned in the internal optical path of the polarization-macroscopic dual-optical-path visual imaging terminal, and is located in the reflection and return optical path of the birefringent photoelastic cooperative target, so that the beam splitter receives the reflected and return light and separates the reflected and return light into a transmitted beam that travels along the first direction and a reflected beam that travels along the second direction.

[0052] The main camera is positioned on the output light path of the transmitted beam of the beam splitter and is used to acquire the natural ambient light geometric image of the birefringent photoelastic cooperative target when the active circular polarization light source module is turned off.

[0053] The circular polarization analyzer is placed on the output light path of the reflected beam of the beam splitter, and the polarization filtering direction of the circular polarization analyzer is orthogonal to the polarization state of the active circular polarization light source module. It is used to filter out the background ambient light and transmit birefringent polarization interference signal light.

[0054] The secondary camera is positioned on the outgoing light path of the circular polarization analyzer on the side away from the beam splitter, and is used to acquire images of polarization interference fringes transmitted through the circular polarization analyzer when the active circular polarization light source module is turned on.

[0055] After the system is powered on and initialized as described in the aforementioned macroscopic workflow, a time-division multiplexing trigger signal is output to the polarization-macroscopic dual-optical-path visual imaging terminal to send the natural ambient light geometric image and polarization interference fringe image to the host computer system in the form of a data stream. Specifically, this includes generating a high-frequency square wave control signal to send a multimodal image data stream.

[0056] This invention provides a process for acquiring geometric images of natural ambient light and polarization interference fringe images, including image acquisition within odd-numbered frame triggering periods and image acquisition within even-numbered frame triggering periods.

[0057] After the system is powered on and initialized, the FPGA timing controller generates a high-frequency square wave control signal, and outputs the high-frequency square wave control signal as a time-division multiplexing trigger signal to the polarization-macroscopic dual-optical-path vision imaging terminal.

[0058] During the odd-numbered frame triggering period, the FPGA timing controller controls the active circularly polarized light source module to be in a power-off state through a high-frequency square wave control signal. The FPGA timing controller also triggers the global shutter of the main camera for synchronous exposure and controls the secondary camera to be in a stopped exposure state, so that the main camera can acquire the spatial coding pattern of the birefringent photoelastic cooperative target under natural ambient light and generate a geometric image of natural ambient light.

[0059] During the even-numbered frame triggering period, the FPGA timing controller controls the active circularly polarized light source module to be powered on via a high-frequency square wave control signal. This causes the active circularly polarized light source module to emit circularly polarized light of a preset wavelength toward the birefringent photoelastic cooperative target. Meanwhile, the FPGA timing controller triggers the global shutter of the secondary camera for synchronous exposure and controls the primary camera to be in a stopped exposure state. This allows the secondary camera to acquire the optical signal transmitted through the circularly polarized analyzer and generate a polarization interference fringe image.

[0060] The polarization-macroscopic dual-optical-path visual imaging terminal alternately arranges the geometric image of natural ambient light and the polarization interference fringe image according to the triggering sequence of the high-frequency square wave control signal, generating a multimodal image data stream, and the polarization-macroscopic dual-optical-path visual imaging terminal sends the multimodal image data stream to the host computer industrial control system.

[0061] The aforementioned macroscopic workflow involves extracting isochromatic and isoclinal feature parameters from polarization interference fringe images and converting these parameters into a full-field two-dimensional strain tensor for the target region. Specifically, this includes extracting image feature parameters and constructing a two-dimensional strain tensor.

[0062] This invention provides a transformation process for a full-field two-dimensional strain tensor, including the extraction of characteristic parameters and the transformation of two-dimensional tensor coordinates.

[0063] The host computer system receives the polarization interference fringe image and performs digital image processing on it to extract the isochromatic fringe order distribution and the principal stress direction angle distribution of the isoclipt lines. The isochromatic fringe order distribution represents the relative difference between the first principal strain and the second principal strain at each pixel on the surface of the birefringent photoelastic cooperative target, and the principal stress direction angle distribution of the isoclipt lines represents the principal strain direction angle at each pixel on the surface of the birefringent photoelastic cooperative target.

[0064] The host computer system reads the pre-calibrated strain fringe constant and photoelastic coating thickness of the birefringent photoelastic cooperative target, and calculates the difference between the first principal strain and the second principal strain of any pixel point in the target area by dividing the product of the isochromatic line fringe order distribution and the strain fringe constant by the photoelastic coating thickness according to the law of strain optics.

[0065] The host computer system combines the elasticity compatibility equation to numerically integrate the difference between the first principal strain and the second principal strain, calculates the sum of the first principal strain and the second principal strain, and separates the independent values ​​of the first principal strain and the second principal strain through the difference between the first principal strain and the second principal strain and the sum of the first principal strain and the second principal strain.

[0066] The host computer system performs a two-dimensional tensor coordinate transformation based on the independent first principal strain value, the independent second principal strain value, and the isoclinic principal stress direction angle distribution. This transformation generates a full-field two-dimensional strain tensor of the target region in Cartesian coordinates. This full-field two-dimensional strain tensor includes X-axis normal strain components, Y-axis normal strain components, and XY-plane shear strain components. The X-axis normal strain component is the product of the average of the first and second principal strain values ​​plus half the principal strain difference and twice the cosine of the principal stress direction angle. The Y-axis normal strain component is the product of the average of the first and second principal strain values ​​minus half the principal strain difference and twice the cosine of the principal stress direction angle. The XY-plane shear strain component is the product of half the principal strain difference and twice the sine of the principal stress direction angle. This full-field two-dimensional strain tensor serves as the data basis for subsequent algorithms to compensate for the physical offset of material deformation.

[0067] The aforementioned macroscopic workflow describes the calculation of the local elastic displacement field integral based on the coordinates of feature points in the geometric image of natural ambient light using the full-field two-dimensional strain tensor. Specifically, this includes establishing an integral reference origin and calculating the local elastic deformation displacement vector.

[0068] This invention provides an integral calculation process for a local elastic displacement field, including the establishment of the integral reference origin and the solution of the local elastic deformation displacement vector.

[0069] The host computer system identifies the unforced boundary points or physical geometric centers of the birefringent photoelastic cooperative target in the ambient light geometric image, and sets these points as the integration reference origin for calculating the local elastic displacement field. Under the condition that the photoelastic coating of the birefringent photoelastic cooperative target satisfies the thin film physical boundary conditions, the host computer system sets the normal deformation of the birefringent photoelastic cooperative target surface as a high-order infinitesimal quantity and ignores it, retaining only the in-plane deformation component. The host computer system extracts the spatial positioning parameters from the ambient light geometric image. Multiple geometric feature points are identified, and a spatial integration path from the integration reference origin to each geometric feature point is planned in the full-field two-dimensional strain tensor. Line integration is then performed on the full-field two-dimensional strain tensor along the spatial integration path to calculate the local elastic deformation displacement vector of each geometric feature point caused by elastic stress tension or compression. This local elastic deformation displacement vector represents the spatial distortion offset of the geometric feature point in the two-dimensional physical plane relative to the ideal force-independent state, and the local elastic deformation displacement vector is used as a position compensation parameter input into the subsequent nonlinear optimization algorithm.

[0070] The aforementioned macroscopic workflow involves compensating for the physical offset of feature points caused by material deformation based on integral calculation results and reconstructing the true rigid body pose matrix of the bridge beam segment to be assembled using a nonlinear optimization algorithm. Specifically, this includes extracting nominal coordinates to output the true rigid body pose matrix.

[0071] This invention provides a process for reconstructing the pose matrix of a real rigid body, including the construction of a reprojection error objective function and the iterative solution of a nonlinear least squares algorithm.

[0072] The host computer system extracts the three-dimensional nominal coordinates of multiple geometric feature points on the birefringent photoelastic cooperative target in an ideal rigid body state without force, and adds the three-dimensional nominal coordinates to the local elastic deformation displacement vector to obtain the actual three-dimensional physical coordinates of the geometric feature points after deformation under force. The host computer system also extracts the two-dimensional pixel observation coordinates of the corresponding geometric feature points in the geometric image of the natural ambient light, and constructs a reprojection error objective function with a deformation compensation term based on the actual three-dimensional physical coordinates and the two-dimensional pixel observation coordinates.

[0073] The mathematical expression for the objective function of reprojection error with the addition of a deformation compensation term is:

[0074]

[0075] In the objective function of reprojection error For rotation matrix, It is a translation vector. Let be the objective function for the reprojection error to be optimized. This represents the total number of geometric feature points. The index of the geometric feature point. These are the two-dimensional pixel observation coordinates of the geometric feature points. This is the perspective projection mapping function from three-dimensional space to a two-dimensional pixel plane. The intrinsic parameter matrix of the main camera, The three-dimensional nominal coordinates of the geometric feature points. This is the local elastic deformation displacement vector. The symbol represents the L2 norm square operator for calculating the Euclidean distance.

[0076] The host computer system uses the Levenberg-Marquardt nonlinear least squares algorithm to iteratively solve the reprojection error objective function, searching for the rotation matrix and translation vector that minimize the reprojection error objective function. The rotation matrix and translation vector in the minimum state are then combined to form the true rigid body pose matrix of the bridge beam segment to be assembled. This allows the true rigid body pose matrix of the bridge beam segment to be assembled to eliminate the computational interference caused by the physical elastic distortion of the birefringent photoelastic cooperative target on the spatial positioning coordinates.

[0077] The aforementioned macroscopic workflow involves constructing an admittance control model based on the real rigid body pose matrix and the preset target pose, and then calculating the position deviation and virtual restoring force. Specifically, this includes calculating the pose deviation and constructing the admittance control model.

[0078] This invention provides a process for constructing an admittance control model, including the calculation of pose deviation and virtual restoring force, and the establishment of a second-order rigid-flexible coupled admittance control model.

[0079] The host computer system receives the actual rigid body pose matrix and the pre-set target pose matrix of the bridge beam segment to be assembled, and performs a six-degree-of-freedom spatial difference operation on the actual rigid body pose matrix and the target pose matrix of the bridge beam segment to be assembled to calculate a six-dimensional geometric deviation vector containing three translational degrees of freedom deviations and three rotational degrees of freedom deviations.

[0080] The host computer system retrieves the pre-configured position loop proportional gain matrix and position loop differential gain matrix, and adds the product obtained by multiplying the six-dimensional geometric deviation vector with the position loop proportional gain matrix to the product obtained by multiplying the first-order time derivative of the six-dimensional geometric deviation vector with the position loop differential gain matrix, to calculate the virtual restoring force vector that drives the three-dimensional hydraulic jack cluster to approach the target pose of the bridge beam segment to be assembled.

[0081] The host computer industrial control system constructs a second-order rigid-flexible coupled admittance control model in the underlying hydraulic servo control framework, which includes a virtual mass matrix, a virtual damping matrix, and a virtual stiffness matrix. In the second-order rigid-flexible coupled admittance control model, the product of the second-order time derivative of the hydraulic system's end displacement correction and the virtual mass matrix, plus the product of the first-order time derivative of the hydraulic system's end displacement correction and the virtual damping matrix, plus the product of the hydraulic system's end displacement correction and the virtual stiffness matrix, is numerically equal to the virtual restoring force vector. This allows the second-order rigid-flexible coupled admittance control model to output the hydraulic system's end displacement correction based on the virtual restoring force vector.

[0082] The process described in the aforementioned macroscopic workflow involves calculating the two-dimensional fringe density gradient matrix of the polarization interference fringe image and mapping the two-dimensional fringe density gradient matrix to the physical distribution topology space corresponding to the three-dimensional hydraulic jack cluster. Specifically, this includes extracting the series distribution of isochromatic fringe lines and establishing a topological mapping relationship.

[0083] This invention provides a mapping process for a two-dimensional fringe density gradient matrix, including the calculation of the two-dimensional fringe density gradient matrix and the mapping of the physical distribution topological space.

[0084] The host computer system extracts the fringe order distribution of the isochromatic lines in the polarization interference fringe image, performs two-dimensional spatial differentiation on the fringe order distribution, and calculates the two-dimensional fringe density gradient matrix of the polarization interference fringe image. The gradient value in the two-dimensional fringe density gradient matrix represents the degree of local stress concentration in the area covered by the birefringent photoelastic cooperative target.

[0085] The host computer industrial control system constructs a homogeneous transformation matrix between the two-dimensional image pixel coordinate system where the birefringent photoelastic cooperative target is located and the physical space coordinate system of the bottom of the bridge beam segment to be assembled. Based on the homogeneous transformation matrix, the image pixel region in the two-dimensional stripe density gradient matrix is ​​positively mapped to the physical force influence surface corresponding to each control node in the three-dimensional hydraulic jack cluster.

[0086] The host computer industrial control system establishes a mapping function from the pixel coordinates of photoelastic stripes to the physical space topology of the three-dimensional hydraulic jack cluster, and divides the two-dimensional stripe density gradient matrix into multiple local stress gradient sub-matrices that correspond one-to-one with the control nodes of the three-dimensional hydraulic jack cluster, so that the optical interference features in the two-dimensional image domain are transformed into stress distribution features in the physical distribution topology space of the three-dimensional hydraulic jack cluster.

[0087] The aforementioned macroscopic workflow involves adjusting the local virtual stiffness coefficients in the admittance control model online based on the local gradient integral values ​​in the physical distribution topology space until the closure and positioning of the bridge beam segments to be assembled is completed. Specifically, this includes calculating the local gradient integral values ​​to drive the execution end to generate spatial displacement and non-uniform yielding.

[0088] This invention provides an online adjustment process for local virtual stiffness coefficients, including the calculation of local gradient integral values ​​and the decay update of the virtual stiffness matrix.

[0089] The host computer system performs area integration on the local stress gradient submatrix corresponding to each control node of the three-dimensional hydraulic jack cluster in the physical distribution topology space, and calculates the local gradient integral value of the corresponding control node.

[0090] The host computer system performs online attenuation updates on the diagonal elements of the virtual stiffness matrix in the second-order rigid-flexible coupled admittance control model based on the local gradient integral value, and the diagonal elements are the local virtual stiffness coefficients of the corresponding control nodes.

[0091] The mathematical expression for the dynamic update of the local virtual stiffness coefficient is:

[0092]

[0093] In the mathematical expression for the dynamic update of the local virtual stiffness coefficient, For the first Each control node at time The local virtual stiffness coefficient, Set a threshold for the initial rigidity. This is the stiffness attenuation gain coefficient. For the first Each control node is responsible for controlling the physical force-affected surface. Let L be the gradient norm of the local stress gradient submatrix within the physical force-affected surface. This represents the area integration operator on a surface affected by physical forces.

[0094] When the local joint surface of the bridge beam segment to be assembled causes local stress concentration due to rigid contact, the local gradient integral value of the corresponding control node increases, and the host computer system reduces the local virtual stiffness coefficient of the corresponding control node according to the dynamic update mathematical expression of the local virtual stiffness coefficient.

[0095] The host computer system substitutes the reduced local virtual stiffness coefficient into the second-order rigid-flexible coupling admittance control model. Under the same input condition of virtual restoring force, the system calculates the numerically increased end displacement correction through the second-order rigid-flexible coupling admittance control model and converts the end displacement correction into a displacement control command, which is then sent to the three-dimensional hydraulic jack cluster.

[0096] The corresponding control nodes in the three-dimensional hydraulic jack cluster increase the yield displacement according to the displacement control command, so that the three-dimensional hydraulic jack cluster synchronously generates local spatial deflection during the process of moving towards the target pose, until the six-dimensional geometric deviation vector approaches zero and the two-dimensional strain tensor distribution in the whole field is uniform, thus completing the closure positioning of the bridge beam segment to be assembled.

[0097] For the data communication protocol between the host computer industrial control system and the three-dimensional hydraulic jack cluster, as well as the underlying closed-loop drive circuit of the hydraulic servo valve, those skilled in the art can use industrial Ethernet and conventional proportional-integral-derivative controllers to implement it. The data communication protocol and the underlying closed-loop drive circuit are well-known technologies in this field and will not be described in detail here.

[0098] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine vision-based bridge installation positioning and guidance fine control system, used to execute a machine vision-based bridge installation positioning and guidance fine control method, characterized in that, include: A birefringent photoelastic cooperative target is placed in the stress area of ​​the bridge beam segment to be assembled; The polarization-macro dual-optical-path visual imaging terminal is set on the reference reference surface of the fixed bridge pier section and aligned with the birefringent photoelastic cooperative target. An FPGA timing controller is electrically connected to a polarization-macro dual-optical-path visual imaging terminal and is used to output a hardware synchronization trigger signal to the polarization-macro dual-optical-path visual imaging terminal. A three-dimensional hydraulic jack cluster, with its mechanical actuators supported at the bottom of the bridge beam segment to be assembled; The host computer industrial control system is connected to the polarization-macro dual-optical-path vision imaging terminal and the three-dimensional hydraulic jack cluster, respectively. It is used to receive image data, execute data processing and control logic, and issue displacement control commands to the three-dimensional hydraulic jack cluster.

2. The precision control system for bridge installation positioning guidance based on machine vision according to claim 1, characterized in that, The polarization-macroscopic dual-optical-path visual imaging terminal includes: An active circularly polarized light source module is arranged on the front light-transmitting side of the polarization-macroscopic dual-optical-path visual imaging terminal, and is used to emit circularly polarized light of a preset wavelength to the birefringent photoelastic cooperative target; A beam splitter is set in the internal optical path of the polarization-macroscopic dual-optical-path visual imaging terminal and is located in the reflection and return optical path of the birefringent photoelastic cooperative target. It is used to separate the reflected and return light into a transmitted beam that propagates along the first direction and a reflected beam that propagates along the second direction. The main camera is positioned on the output light path of the transmitted beam of the beam splitter and is used to acquire the natural ambient light geometric image of the birefringent photoelastic cooperative target when the active circular polarization light source module is turned off. A circular polarization analyzer is placed on the output light path of the reflected beam of the beam splitter, and the polarization filtering direction is orthogonal to the polarization state of the active circular polarization light source module. It is used to filter out background ambient light and transmit birefringent polarization interference signal light. The secondary camera is positioned on the outgoing light path of the circular polarization analyzer on the side away from the beam splitter, and is used to acquire images of polarization interference fringes transmitted through the circular polarization analyzer when the active circular polarization light source module is turned on.

3. The precision control system for bridge installation positioning guidance based on machine vision according to claim 1, characterized in that, The birefringent photoelastic cooperative target includes: The reflective substrate is bonded to the surface of the stress area of ​​the bridge beam segment to be assembled, and is made of epoxy resin binder doped with high reflectivity metal particles, thereby forming a diffuse reflective substrate on the surface of the bridge beam segment to be assembled. The photoelastic coating is layered and laid on the outer side of the reflective substrate away from the surface of the bridge beam segment to be assembled. The material is an epoxy resin-based birefringent polymer with a fixed physical thickness and a calibrated strain fringe constant. The spatial coding pattern, attached to the surface of the photoelastic coating away from the reflective substrate, consists of multiple asymmetrically distributed geometric primitives. Under natural ambient light, these geometric primitives form a set of feature points for spatial positioning in machine vision.

4. A machine vision-based bridge installation positioning guidance fine control method, applied to the machine vision-based bridge installation positioning guidance fine control system as described in any one of claims 1 to 3, characterized in that, Includes the following steps: S101, after the system is powered on and initialized, the FPGA timing controller outputs a time-division multiplexing trigger signal to the polarization-macro dual-optical-path vision imaging terminal; S102, within the odd-numbered frame triggering period, the polarization-macroscopic dual-optical-path visual imaging terminal acquires the natural ambient light geometric image of the birefringent photoelastic cooperative target; S103, during the even-numbered frame triggering period, the polarization-macroscopic dual-optical-path visual imaging terminal activates the internal active circular polarization light source module to acquire polarization interference fringe images of the birefringent photoelastic cooperative target. S104, the polarization-macroscopic dual-optical-path visual imaging terminal sends the geometric image of natural ambient light and the polarization interference fringe image to the host computer system in the form of a data stream. S105, The host computer / industrial control system extracts isochromatic line feature parameters and isoclimatic line feature parameters from the polarization interference fringe image; S106, combining the optical and mechanical constants of the birefringent photoelastic cooperative target, the host computer and industrial control system convert the isochromatic line characteristic parameters and isoclipt line characteristic parameters into the full-field two-dimensional strain tensor of the target region; S107, the host computer system performs local elastic displacement field integral calculation based on the coordinates of feature points in the geometric image of natural ambient light using the full-field two-dimensional strain tensor; S108, the host computer system compensates for the physical offset of feature points caused by material deformation based on the integral calculation results, and uses a nonlinear optimization algorithm to reconstruct the real rigid body pose matrix of the bridge beam segment to be assembled. S109, the host computer system constructs an admittance control model based on the real rigid body pose matrix and the preset target pose, and calculates the position deviation and virtual restoring force; S110, the host computer industrial control system calculates the two-dimensional fringe density gradient matrix of the polarization interference fringe image and maps the two-dimensional fringe density gradient matrix to the physical distribution topology space corresponding to the three-dimensional hydraulic jack cluster. S111, the host computer industrial control system adjusts the local virtual stiffness coefficient in the admittance control model online based on the local gradient integral value in the physical distribution topology space. When the gradient integral value in the corresponding region increases, the local virtual stiffness coefficient of the corresponding control node decreases. S112, the host computer system generates displacement control commands based on the admittance control model including local virtual stiffness coefficients, and sends the displacement control commands to the three-dimensional hydraulic jack cluster. S113, the three-dimensional hydraulic jack cluster drives the execution end to generate spatial displacement and non-uniform retreat according to the displacement control command, and completes the closure positioning of the bridge beam segment to be assembled.

5. The machine vision-based bridge installation positioning guidance fine control method according to claim 4, characterized in that, The steps S101 to S104, specifically, include outputting the time-division multiplexing trigger signal to transmit the image in the form of a data stream. The FPGA timing controller generates a high-frequency square wave control signal, and outputs the high-frequency square wave control signal as a time-division multiplexing trigger signal to the polarization-macro dual-optical-path vision imaging terminal. During the odd-numbered frame triggering period, the FPGA timing controller controls the active circularly polarized light source module to be in a power-off state through a high-frequency square wave control signal. The FPGA timing controller also triggers the global shutter of the main camera for synchronous exposure and controls the secondary camera to be in a stopped exposure state, so that the main camera can collect the spatial coding pattern of the birefringent photoelastic cooperative target under natural ambient light and generate a geometric image of natural ambient light. During the even-numbered frame triggering period, the FPGA timing controller controls the active circularly polarized light source module to be powered on via a high-frequency square wave control signal, so that the active circularly polarized light source module emits circularly polarized light of a preset wavelength toward the birefringent photoelastic cooperative target. The FPGA timing controller also triggers the global shutter of the secondary camera for synchronous exposure and controls the main camera to be in the stopped exposure state, so that the secondary camera can collect the optical signal transmitted through the circularly polarized analyzer and generate a polarization interference fringe image. The polarization-macroscopic dual-optical-path visual imaging terminal alternately arranges the geometric image of natural ambient light and the polarization interference fringe image according to the triggering sequence of the high-frequency square wave control signal, generating a multimodal image data stream, and the polarization-macroscopic dual-optical-path visual imaging terminal sends the multimodal image data stream to the host computer industrial control system.

6. The machine vision-based bridge installation positioning guidance fine control method according to claim 4, characterized in that, The steps S105 to S106, which involve extracting isochromatic and isoclinal feature parameters from the polarization interference fringe image and converting them into a full-field two-dimensional strain tensor for the target region, specifically include: The host computer system receives the polarization interference fringe image and performs digital image processing on it to extract the order distribution of the isochromatic fringes and the principal stress direction angle distribution of the isoclimax lines in the polarization interference fringe image. The host computer system reads the strain fringe constant and photoelastic coating thickness of the birefringent photoelastic cooperative target, and calculates the difference between the first principal strain and the second principal strain of any pixel in the target area by dividing the product of the isochromatic line fringe order distribution and the strain fringe constant by the photoelastic coating thickness according to the law of strain optics. The host computer system combines the elasticity compatibility equation to numerically integrate the difference between the first principal strain and the second principal strain, calculates the sum of the first principal strain and the second principal strain, and separates the independent first principal strain value and the independent second principal strain value through the difference between the first principal strain and the second principal strain and the sum of the first principal strain and the second principal strain. The host computer system performs a two-dimensional tensor coordinate transformation based on the independent first principal strain value, the independent second principal strain value, and the angular distribution of the principal stress direction along the isoclinic lines. This generates a full-field two-dimensional strain tensor of the target region in the Cartesian coordinate system. The full-field two-dimensional strain tensor includes the X-axis normal strain component, the Y-axis normal strain component, and the XY plane shear strain component.

7. The machine vision-based bridge installation positioning guidance fine control method according to claim 4, characterized in that, The execution step S107, which involves performing local elastic displacement field integration calculation based on the coordinates of feature points in the geometric image of natural ambient light using the full-field two-dimensional strain tensor, specifically includes: The host computer system identifies the unforced boundary point or physical geometric center of the birefringent photoelastic cooperative target in the geometric image of natural ambient light, and sets the unforced boundary point or physical geometric center as the integral reference origin for the calculation of the local elastic displacement field. Under the condition that the photoelastic coating of the birefringent photoelastic cooperative target satisfies the physical boundary conditions of the thin film, the host computer system sets the normal deformation of the surface of the birefringent photoelastic cooperative target as a high-order infinitesimal quantity and ignores it, retaining only the in-plane deformation component. The host computer system extracts multiple geometric feature points for spatial positioning from the geometric image of the natural ambient light and plans the spatial integration path from the integral reference origin to each geometric feature point in the full-field two-dimensional strain tensor. By performing line integration on the full-field two-dimensional strain tensor along the spatial integration path, the local elastic deformation displacement vector of each geometric feature point caused by elastic stress tension or compression is calculated.

8. The machine vision-based bridge installation positioning guidance fine control method according to claim 7, characterized in that, The execution step S108, which involves compensating for the physical offset of feature points caused by material deformation based on the integral calculation results and reconstructing the true rigid body pose matrix of the bridge beam segment to be assembled using a nonlinear optimization algorithm, specifically includes: The host computer system extracts the three-dimensional nominal coordinates of multiple geometric feature points on the birefringent photoelastic cooperative target in the ideal rigid body state without force, and adds the three-dimensional nominal coordinates to the local elastic deformation displacement vector to obtain the actual three-dimensional physical coordinates of the geometric feature points after force deformation. The host computer system extracts the two-dimensional pixel observation coordinates of the corresponding geometric feature points in the geometric image of natural ambient light, and constructs a reprojection error objective function with deformation compensation term based on the actual three-dimensional physical coordinates and the two-dimensional pixel observation coordinates. The Levenberg-Marquardt nonlinear least squares algorithm is used to iteratively solve the reprojection error objective function. The rotation matrix and translation vector that minimize the reprojection error objective function are searched, and the rotation matrix and translation vector in the minimum state are combined to form the true rigid body pose matrix of the bridge beam segment to be assembled.

9. The machine vision-based bridge installation positioning guidance fine control method according to claim 4, characterized in that, The execution step S109, which involves constructing an admittance control model based on the real rigid body pose matrix and the preset target pose, and calculating the position deviation and virtual restoring force, specifically includes: A six-degree-of-freedom spatial difference operation is performed on the actual rigid body pose matrix and the target pose matrix of the bridge beam segment to be assembled to calculate a six-dimensional geometric deviation vector containing three translational degrees of freedom deviations and three rotational degrees of freedom deviations. The pre-configured position loop proportional gain matrix and position loop differential gain matrix are retrieved, and the product obtained by multiplying the six-dimensional geometric deviation vector with the position loop proportional gain matrix is ​​added to the product obtained by multiplying the first-order time derivative of the six-dimensional geometric deviation vector with the position loop differential gain matrix to calculate the virtual restoring force vector that drives the three-dimensional hydraulic jack cluster to approach the target pose of the bridge beam segment to be assembled. In the underlying hydraulic servo control framework, a second-order rigid-flexible coupled admittance control model is constructed, which includes a virtual mass matrix, a virtual damping matrix, and a virtual stiffness matrix. In the second-order rigid-flexible coupled admittance control model, the product of the second-order time derivative of the hydraulic system's end displacement correction and the virtual mass matrix, plus the product of the first-order time derivative of the hydraulic system's end displacement correction and the virtual damping matrix, plus the product of the hydraulic system's end displacement correction and the virtual stiffness matrix, is numerically equal to the virtual restoring force vector.

10. The machine vision-based bridge installation positioning guidance fine control method according to claim 4, characterized in that, The execution steps S110 to S112, which involve mapping the two-dimensional stripe density gradient matrix to the physical distribution topology space corresponding to the three-dimensional hydraulic jack cluster, adjusting the local virtual stiffness coefficients in the admittance control model online based on the local gradient integral values ​​in the physical distribution topology space, and generating displacement control commands to be sent to the three-dimensional hydraulic jack cluster, specifically include: A homogeneous transformation matrix is ​​constructed between the two-dimensional image pixel coordinate system where the birefringent photoelastic cooperative target is located and the physical space coordinate system at the bottom of the bridge beam segment to be assembled. Based on the homogeneous transformation matrix, the image pixel region in the two-dimensional stripe density gradient matrix is ​​positively mapped to the physical force influence surface corresponding to each control node in the three-dimensional hydraulic jack cluster. The two-dimensional stripe density gradient matrix is ​​divided into multiple local stress gradient sub-matrices that correspond one-to-one with the control nodes of the three-dimensional hydraulic jack cluster. In the physical distribution topology space, the area integral operation is performed on the local stress gradient submatrix corresponding to each control node of the three-dimensional hydraulic jack cluster to calculate the local gradient integral value of the corresponding control node. The host computer system performs online attenuation updates on the diagonal elements of the virtual stiffness matrix in the second-order rigid-flexible coupling admittance control model based on the local gradient integral value. The diagonal elements are the local virtual stiffness coefficients of the corresponding control nodes. The reduced local virtual stiffness coefficients are substituted into the second-order rigid-flexible coupling admittance control model to calculate the numerically increased end displacement correction. The end displacement correction is then converted into a displacement control command and sent to the three-dimensional hydraulic jack cluster.

Images