Multi-sensor fusion-based real-time control system for walking posture of bridge-building machine

Abstract

The application relates to the technical field of bridge construction equipment, and discloses a real-time control system for the walking posture of a bridge builder based on multi-sensor fusion, which comprises the following modules: a heterogeneous sensor and data acquisition module, which acquires kinematics and structural strain data of the bridge builder in real time; an offline modeling and calibration module, which pre-establishes an initial dynamics model and key mapping relationship; an online state estimation and model adaptive module, which performs state estimation, model correction and disturbance identification and outputs comprehensive state information; a multi-constraint predictive control decision module, which solves an optimization problem containing structural safety constraints and generates optimal control instructions; and an execution and feedback module, which analyzes the optimal control instructions to drive an execution mechanism and adjusts the posture to form a real-time closed loop. The application adopts a scheme of fusing multi-source data and adaptively correcting a model online, solves the fundamental defect that the model does not match the actual system state, and causes the control precision to obviously decrease with the passage of time and environmental changes.

Description

Technical Field

[0001] This invention relates to the field of bridge construction equipment technology, specifically to a real-time control system for the walking posture of a bridge-building machine based on multi-sensor fusion. Background Technology

[0002] As modern bridge engineering develops towards longer spans and modular designs, the size and weight of single precast box girders are increasing daily. As the core equipment for the precise transport and erection of these giant components, the stability and safety of the bridge-building machine's operation are paramount. Any slight deviation in posture or excessive structural vibration could cause irreversible damage to the equipment itself or the already constructed bridge structure.

[0003] Currently, the industry widely adopts automated control solutions based on Programmable Logic Controllers (PLCs) and servo hydraulic technology. These solutions, through preset program logic, achieve automated sequential control of the hydraulic actuators of the bridge-building machine, replacing tedious manual operations and reliably completing travel and beam placement tasks under standard, ideal working conditions. In simple point-to-point movement applications, this technology provides a stable and cost-effective solution, forming the foundation of current automated control for bridge-building machines.

[0004] However, as the working environment becomes more complex and the operational requirements become more demanding, the inherent limitations of existing technical solutions become apparent. First, the mathematical model underlying its control algorithm typically simplifies the massive bridge-building machine into an ideal rigid body. This is far removed from the significant flexibility exhibited by the equipment under heavy loads. This mismatch between the model and the physical entity makes it impossible for the control system to accurately predict and suppress vibrations and deformations caused by its own flexibility. During dynamic processes such as start-up, shutdown, or speed changes, the attitude control accuracy is difficult to guarantee. Second, in actual open-air operations, bridge-building machines often encounter unforeseen disturbances such as sudden wind loads or uneven settlement of the supporting foundation. Existing systems handle such disturbances with passive hysteresis compensation, meaning they can only correct after attitude deviations occur, failing to identify the source of the disturbance and take proactive feedforward suppression measures. This results in poor compensation effects and slow response. Finally, in pursuing attitude tracking accuracy, existing control logic does not take into account the stress state of the structure itself. The control system is blind to the structural safety risks that its own behavior may cause. This creates a disconnect between control decisions and structural safety, lacking predictive protection for structural safety and harboring potential safety risks under extreme conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion. This system solves the technical problems of insufficient posture control accuracy and structural safety risks in complex working conditions caused by existing technologies, which use simplified ideal models, can only passively and lag-compensate for complex disturbances such as wind loads and foundation settlement, and separate control decisions from structural safety.

[0006] To achieve the above objectives, this invention provides the following technical solution: a real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion. This system constructs a cyber-physical closed loop through the coordinated operation of the following modules: The heterogeneous sensing and data acquisition module is used to collect the kinematic data and structural strain data of the bridge-building machine in real time and generate synchronous sensor data. The offline modeling and calibration module is used to establish and output the initial dynamic model and key mapping relationships based on the design parameters of the bridge-building machine; The online state estimation and model adaptation module is used to receive the synchronous sensor data and, in combination with the initial dynamic model and key mapping relationship, generate comprehensive state information including the augmented state estimate at the current moment, the real-time corrected dynamic model, and the equivalent perturbation force feedforward signal. The multi-constraint predictive control decision module is used to solve the multi-constraint optimization problem online based on the comprehensive state information and the preset target attitude trajectory, so as to generate the optimal control command; The execution and feedback module is used to receive and parse the optimal control command to drive the actuators on the bridge-building machine to produce physical movements, thereby adjusting the attitude of the bridge-building machine.

[0007] In one optional technical solution, the heterogeneous sensing and data acquisition module specifically includes: A real-time dynamic differential positioning receiver and an inertial measurement unit of the global satellite navigation system are jointly installed at the main beam stiffness position of the bridge-building machine to measure the kinematic data; A multi-channel fiber Bragg grating sensor array is distributed along the main beam and supporting columns of the bridge-building machine to measure the strain data of the structure. The data synchronization and preprocessing unit is used to perform high-precision time synchronization, temperature compensation, and filtering preprocessing on the kinematic data and the structural strain data to generate the synchronized sensor data.

[0008] In one optional technical solution, the offline modeling and calibration module is specifically used for: Based on the three-dimensional computer-aided design drawings of the bridge-building machine, a finite element model containing the geometric structure, material properties, and connection relationships between components of the bridge-building machine is established. Modal analysis and model order reduction are performed on the finite element model to generate an initial dynamic model in state-space form that describes the dynamic characteristics of the system; By applying a series of typical loads and disturbance conditions to the finite element model for simulation analysis, a stress-state mapping function that can quickly deduce structural stress from the system state and an external disturbance-strain field sensitivity matrix that describes the quantitative relationship between a specific external disturbance and the sensor strain response are calibrated. These two together constitute the key mapping relationship.

[0009] In one optional technical solution, the online state estimation and model adaptation module is specifically used for: An unscented Kalman filter algorithm is used to nonlinearly fuse the kinematic data and the structural strain data to estimate an augmented state vector in real time that includes the rigid motion and flexible deformation of the bridge-building machine. This vector is used as the augmented state estimate. The strain residual between the measured structural strain and the predicted strain estimated based on the current state is calculated. When the strain residual continues to exceed a preset threshold, a parameter identifier is triggered to update the key parameters of the initial dynamic model online and generate the real-time corrected dynamic model. Analyze the real-time spatial distribution field of the structural strain data. When the real-time distribution field exhibits a specific topological pattern that deviates from the normal working condition, use the external disturbance-strain field sensitivity matrix in the key mapping relationship to solve the inverse problem in order to calculate the equivalent disturbance force feedforward signal that causes the anomaly.

[0010] Furthermore, the augmented state vector includes: the three-dimensional position vector and three-dimensional velocity vector of the bridge-building machine in the global coordinate system, the attitude Euler angle vector of the bridge-building machine and the angular velocity vector in the body coordinate system, and the modal coordinate vector of a preset order describing the flexible deformation of the main beam of the bridge-building machine and its first-order time rate of change vector.

[0011] Furthermore, updating the initial dynamic model parameters includes: A parameter identifier based on recursive least squares is used to update the matrix parameters related to modal stiffness in the initial dynamic model online using the strain residuals.

[0012] Furthermore, calculating the equivalent disturbance force feedforward signal includes: The difference between the real-time strain field vector at the current moment and the reference strain field vector under normal operating conditions is used to obtain a strain deviation vector. The strain deviation vector is then multiplied by the generalized inverse matrix of the external disturbance-strain field sensitivity matrix in the key mapping relationship to calculate the equivalent disturbance force feedforward signal.

[0013] In one optional technical solution, the multi-constraint predictive control decision module is specifically used for: Construct a cost function to minimize attitude tracking error and control energy consumption within a finite future prediction time domain; A set of constraints is set, including: system dynamic constraints requiring the system state evolution to follow the real-time corrected dynamic model in the comprehensive state information; control input constraints reflecting the performance limits of physical actuators such as hydraulic cylinders; and structural safety constraints ensuring that the predicted stress values ​​of key structural components do not exceed the allowable stress limits of the materials. The cost function and the set of constraints together constitute a quadratic programming problem. This quadratic programming problem is solved online using an efficient numerical optimization solver to obtain an optimal control input sequence. The first control command of the optimal control input sequence is then extracted as the optimal control command.

[0014] Furthermore, the structural safety constraints include: At each step within the prediction time domain, the predicted system state value is quickly mapped to the predicted stress value of the key component using the structural stress-state mapping function in the key mapping relationship. The predicted stress value is limited to always being less than or equal to the preset allowable material stress safety limit throughout the entire prediction time domain.

[0015] This invention provides a real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion, which has the following beneficial effects: This invention employs a scheme that integrates multi-source data and online adaptive model correction. By utilizing online state estimation and model adaptation modules, kinematic data and structural strain data are deeply fused, and the offline dynamic model is corrected in real time based on measured strain. This results in a real-time updated system model that accurately reflects the dynamic coupling of rigid body motion and flexible deformation. Compared with existing technologies that mostly rely on offline established fixed ideal models that do not change with operating conditions, this invention solves the fundamental defect that the model does not match the actual system state, leading to a significant decrease in control accuracy over time and with environmental changes.

[0016] This invention constructs a disturbance identification and solution mechanism based on strain field topology analysis. By analyzing the morphological characteristics of the distributed strain field of the whole machine in real time and combining it with the pre-calibrated disturbance-strain field sensitivity matrix for inverse solution, it realizes rapid identification and accurate quantification of unpredictable external disturbances such as uneven settlement of the foundation, which are difficult to measure directly. This is completely different from the existing technology that treats all disturbances as unknown interferences and can only perform passive and delayed position error compensation. It solves the limitation of traditional control methods that cannot actively distinguish the source of disturbances, resulting in delayed compensation response and poor control effect.

[0017] This invention integrates structural safety constraints as a rigid boundary condition directly into the online optimization solution process of model predictive control. This ensures that every control command generated by the system, while driving the bridge-building machine to accurately track the target attitude, has already guaranteed that the structural stress will not exceed the safety limit in the future. Existing technologies usually separate attitude control from structural safety analysis, with the control system only responsible for tracking and the safety system responsible for alarms or emergency shutdowns afterward. This invention solves the dilemma of the separated approach, which sacrifices control performance for conservative operation or faces the risk of loss of control and structural damage under extreme conditions. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the system functional modules according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the deployment of the heterogeneous sensing and data acquisition module according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the offline modeling and calibration module of this invention. Figure 4 This is an internal logic block diagram of the online state estimation and model adaptation module in an embodiment of the present invention; Figure 5 This is a schematic diagram of the principle of the multi-constraint predictive control decision module in an embodiment of the present invention; Figure 6 This is a closed-loop schematic diagram of the execution and feedback module in an embodiment of the present invention.

[0019] Explanation of icon numbers: 10. Heterogeneous sensing and data acquisition module; 20. Offline modeling and calibration module; 30. Online state estimation and model adaptation module; 40. Multi-constraint predictive control decision module; 50. Execution and feedback module. Detailed Implementation

[0020] 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.

[0021] Figure 1 This is a functional module diagram of a bridge-building machine travel posture real-time control system based on multi-sensor fusion according to an embodiment of the present invention, as shown below. Figure 1As shown, this embodiment of the invention provides a real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion. The system may include: a heterogeneous sensing and data acquisition module 10; an offline modeling and calibration module 20; an online state estimation and model adaptation module 30; a multi-constraint predictive control decision module 40; and an execution and feedback module 50.

[0022] The control system is a closed-loop control system deployed on the bridge-building machine. The heterogeneous sensing and data acquisition module 10 is physically installed on different structural parts of the bridge-building machine. The offline modeling and calibration module 20 is a one-time offline software module; its generated model and parameters are pre-stored in the system. The online state estimation and model adaptation module 30 and the multi-constraint predictive control decision module 40 are core software modules deployed in the onboard industrial controller. The execution and feedback module 50 includes the system's hardware actuators and their drive units.

[0023] The vehicle-mounted industrial controller can be a high-performance industrial PC (IPC) with high floating-point computing power and multiple real-time communication interfaces, running a real-time operating system (such as Linux-RT or QNX) to ensure deterministic scheduling and execution of the control algorithm. In another embodiment, a high-end programmable logic controller (PLC) supporting complex algorithm computation can be used in conjunction with a motion control module. In yet another embodiment, the vehicle-mounted industrial controller can also be a customized hardware system based on an embedded platform (such as the NVIDIA Jetson series or Xilinx Zynq series), integrating GPU or FPGA units for parallel computing to accelerate the numerical optimization solution process in model predictive control.

[0024] The overall workflow of the system follows a cyclical information loop. In each control cycle, the heterogeneous sensing and data acquisition module 10 first acquires the real-time kinematic data and structural strain data of the bridge-building machine.

[0025] The collected data is fed into the online state estimation and model adaptation module 30. The online state estimation and model adaptation module 30 performs three parallel tasks: first, it fuses multi-source data to estimate an augmented state vector that includes rigid body motion and flexible deformation in real time; second, it uses measured structural strain data to correct and optimize key parameters of the system dynamics model online; and third, it identifies unforeseen external disturbances by analyzing the distributed strain field and calculates the equivalent disturbance force.

[0026] Subsequently, the online state estimation and model adaptation module 30 transmits three key pieces of information—the augmented state estimate at the current moment, the corrected system dynamics model, and the calculated equivalent disturbance feedforward signal—to the multi-constraint predictive control decision module 40.

[0027] The multi-constraint predictive control decision module 40 receives the above information and the externally set target attitude trajectory. Based on the above information, it constructs and solves an online finite-time domain optimization problem containing multiple constraints, and generates a set of optimal control command sequences that can accurately track the target attitude and ensure that the structural stress is within a safe range.

[0028] The multi-constraint predictive control decision module 40 sends only the first control instruction of the above optimal control instruction sequence to the execution and feedback module 50.

[0029] After receiving the optimal control command, the execution and feedback module 50 parses it and drives the hydraulic actuator on the bridge-building machine to produce corresponding actions, thereby physically adjusting the attitude of the bridge-building machine.

[0030] The result of this physical adjustment will instantly change the actual posture and internal stress distribution of the bridge-building machine. This new physical state will be captured by the heterogeneous sensing and data acquisition module 10 in the next control cycle, thereby starting a new round of "perception, estimation, decision-making, and execution" closed-loop process, enabling the system to continuously and continuously perform optimal control of the bridge-building machine's posture.

[0031] To further clarify the implementation method and technical details of the system of the present invention, the specific structure and implementation principle of each functional module constituting the system will be described in detail below.

[0032] like Figure 2 As shown, Figure 2 This is a schematic diagram of the deployment of the heterogeneous sensing and data acquisition module 10 according to an embodiment of the present invention. The heterogeneous sensing and data acquisition module 10 is used to acquire the global kinematic state and internal strain state of the bridge-building machine in real time during its movement, and to provide a high-quality data stream that has been synchronized and preprocessed for subsequent modules.

[0033] Specifically, the heterogeneous sensing and data acquisition module 10 includes a set of kinematic sensors and a set of internal structural state sensors. The kinematic sensors include at least one high-precision GPS-RTK receiver and at least one industrial-grade inertial measurement unit (IMU). To accurately measure the rigid body motion of the entire machine, the GPS-RTK antenna and the IMU unit are jointly mounted at a predetermined location with high structural stiffness on the main beam structure of the bridge-building machine. This location minimizes the impact of local structural vibrations and elastic deformations on the global pose measurement.

[0034] The internal structural sensors are multi-channel fiber Bragg grating (FBG) sensor arrays. This array is constructed by connecting multiple optical fibers in series and is distributed along the critical stress paths of the bridge-building machine. A specific deployment method involves: multiple sets of FBG strain sensors are arranged longitudinally along the outer surfaces of the upper and lower flanges of the main girder; FBG strain sensors are arranged at specific angles (e.g., 45 degrees) in the critical shear areas of the main girder web; and FBG strain sensors are arranged axially along the sidewalls of each main support column. This deployment method effectively captures the main bending, torsional, and compressive deformation modes generated by the bridge-building machine under its own weight and external loads.

[0035] To ensure the accuracy of subsequent multi-source data fusion, the heterogeneous sensing and data acquisition module 10 integrates a data synchronization and preprocessing unit. This data synchronization and preprocessing unit uses Precision Time Protocol (PTP, IEEE 1588 standard) to perform high-precision clock synchronization on all data acquisition devices in the system (including GPS-RTK receivers, IMU data logging units, and FBG demodulators), ensuring that all sensor data packets have a unified, microsecond-level precision synchronization timestamp.

[0036] The preprocessing in the data synchronization and preprocessing unit is used to solve and compensate the acquired raw sensor data. For the FBG sensor array, the raw measurement value is the offset of the grating center wavelength. This offset is affected by both mechanical strain and ambient temperature. To accurately separate the mechanical strain, the preprocessing unit uses the following method for temperature compensation.

[0037] Center wavelength offset of FBG sensor With mechanical strain and temperature change The relationship can be represented as: ; in, The initial center wavelength of the grating is . The effective photoelastic coefficient of the optical fiber material. The coefficient of thermal expansion of the optical fiber material. is the thermo-optic coefficient of the optical fiber material.

[0038] By deploying a pair of FBG sensors (sensor 1 and sensor 2) symmetrically at the upper and lower flanges of the same longitudinal section of the main beam, the strains of the two sensors under pure bending conditions are equal in magnitude but opposite in sign. For physical properties that change with the same temperature, a system of equations can be constructed: ; By solving the simultaneous equations, the pure bending strain of the cross section can be calculated at the same time. and local temperature changes For a single-point deployed FBG sensor, a nearby FBG sensor, unaffected by mechanical stress, is used as a pure temperature sensor for compensation.

[0039] After the calculation and compensation are completed, the data preprocessing unit further applies a digital low-pass filter to the calculated strain data, GPS-RTK position and velocity data, and IMU attitude angle and angular velocity data to filter out high-frequency measurement noise. Finally, the heterogeneous sensing and data acquisition module 10 packages all the data that has undergone time synchronization, calculation compensation, and filtering into a data frame with a unified data structure and timestamp, and sends it in real time to the online state estimation and model adaptation module 30 via industrial Ethernet.

[0040] like Figure 3 As shown, Figure 3 This is a flowchart illustrating the workflow of the offline modeling and calibration module 20 in this embodiment of the invention. The offline modeling and calibration module 20 is a software module that runs before system deployment. Its function is to provide the online control system with a computationally efficient initial predictive model that accurately describes the dynamic characteristics of the bridge-building machine, and to pre-calibrate the key mapping relationships required for system operation. The generated results are stored in the system as initial parameters.

[0041] The implementation of the offline modeling and calibration module 20 includes the following steps: Step 1: Establish a high-fidelity finite element method (FEM) model of the entire bridge-building machine. This step is based on the 3D computer-aided design (CAD) drawings of the bridge-building machine and is performed in a finite element analysis software environment. First, the geometric model of the bridge-building machine is imported into the software, and the material properties of each component, including elastic modulus, Poisson's ratio, and material density, are set according to the actual material (e.g., Q345 steel). Then, the finite element model of the entire bridge-building machine is meshed using element types suitable for large-scale structural analysis (e.g., beam elements, plate / shell elements, and solid elements), and the mesh is refined at critical connection points and high-stress areas. Finally, appropriate boundary conditions and connection constraints are set according to the actual contact between the bridge-building machine and the ground or bridge piers.

[0042] Step 2: Perform model reduction to generate a reduced-order state-space model (ROM) suitable for real-time control, i.e., the initial dynamic model. Due to the limited degrees of freedom of the complete finite element model (…),… The magnitude of the problem is extremely large and cannot be directly used for online calculation. Therefore, modal synthesis is used to reduce the model order. First, modal analysis is performed on the above finite element model to solve its generalized eigenvalue problem of free vibration: ; in, and These are the global mass matrix and stiffness matrix of the finite element model, respectively. It is the natural angular frequency of the structure. It is the corresponding mode shape vector.

[0043] By solving the above problem, a series of natural frequencies and mode shapes can be obtained. From these modes, the modes that have a significant impact on the overall attitude of the bridge-building machine (especially the vertical deflection, lateral bending, and torsion of the main girder) are selected. Low-frequency elastic modes. This... The modal shape vectors are combined column-wise to form the modal matrix. .

[0044] Using this modal matrix, the high-dimensional dynamic equations in physical coordinates are transformed into low-dimensional equations in modal coordinates. The modal coordinate vector of the system is defined as follows: Then the physical displacement vector of the system It can be approximated as Substituting this relationship and performing a coordinate transformation, we obtain the reduced-order state-space model (ROM), whose continuous-time form is as follows: ; in, Represents the state vector The first derivative with respect to time, i.e., the rate of change of the state variable; the state vector. It includes modal coordinates and their rates of change; To control the input vector (such as the force or speed of each hydraulic cylinder); Output vectors for the model (e.g., displacements of key points, attitude angles, and strain values ​​at specific locations); matrices It is a constant matrix obtained by transforming the modal parameters (modal mass, modal stiffness, modal damping) and the force distribution matrix, which together constitute the initial dynamic model of the system.

[0045] Step 3: Perform offline calibration of key mapping relationships. This step establishes the two key function relationships required by the online module.

[0046] The first is the structural stress-state mapping function. In the finite element method software, a series of typical static load conditions are applied to the model. Under each load condition, the state of the model is recorded. (i.e., the modal coordinates of each order) and the von Mises equivalent stress values ​​of key structural monitoring points (e.g., the midpoints of the upper and lower edges of the main beam). The corresponding data between them. By performing polynomial fitting or training a small neural network on this data, a mapping function that can quickly calculate the structural stress from the estimated system state is obtained. .

[0047] Calibration stress-state mapping function The specific steps may include: First, selecting several key monitoring points for structural stress concentration in the established high-fidelity finite element model (FEM). Then, applying a series of typical static load combinations to the FEM model (e.g., box girders of different weights, center of gravity shifts at different locations), performing static analysis on each load condition, and recording the stress values ​​at these key monitoring points. And the state vector of the reduced-order state-space model (ROM) of the system at this time. By collecting a large amount of Data pairs are fitted with mapping functions using machine learning or system identification methods. The mapping function can be a polynomial function, a radial basis function (RBF) network, or a shallow feedforward neural network. This approach avoids time-consuming finite element calculations in online control, enabling rapid stress prediction.

[0048] The second is the external disturbance-strain field sensitivity matrix. To identify unforeseen disturbances (such as uneven settlement at support points), such disturbances are simulated in a finite element model. Specifically, a unit vertical displacement or force is applied at one support point of the bridge-building machine, and the resulting strain response at all FBG sensor installation locations is calculated. This strain response vector is then used as the sensitivity matrix. This is a column. Repeating this process for all possible perturbation locations constructs the complete sensitivity matrix. This sensitivity matrix describes the effect of a unit perturbation on the entire strain field, and its generalized inverse matrix is... It is then used to solve the inverse problem of perturbation identification online.

[0049] Finally, the offline modeling and calibration module 20 will generate the ROM matrix ( ), stress mapping function and perturbation sensitivity matrix As the initial configuration file, it is stored in the vehicle-mounted industrial controller for use by the online state estimation and model adaptation module 30.

[0050] Please see Figure 4 , Figure 4 This is an internal logic block diagram of the online state estimation and model adaptation module 30 according to an embodiment of the present invention. The online state estimation and model adaptation module 30 is a core software module deployed in the vehicle-mounted industrial controller. This module operates in real time during each control cycle. Its function is to integrate multi-source data provided by the heterogeneous sensing and data acquisition module 10 to optimally estimate the complete state of the system, and to correct the system model online based on actual measurement results, identify external disturbances, and provide the most accurate and comprehensive information input for subsequent control decisions.

[0051] The implementation of the online state estimation and model adaptation module 30 includes the following steps: Step 1: Perform augmented state estimation based on Unscented Kalman Filter (UKF). To comprehensively describe the complex dynamics of the bridge-building machine, which simultaneously involves rigid body motion and flexible deformation, this system defines an augmented state vector. Its composition is as follows: ; in, These represent the three-dimensional position and velocity vectors of the bridge-building machine in the global coordinate system; These are the attitude Euler angle vector and the angular velocity vector in the body coordinate system, respectively; These are the modal coordinate vectors describing the flexible deformation of the structure and their rates of change, respectively.

[0052] Due to the highly nonlinear coupling between the system's kinematics and dynamics, an unscented Kalman filter algorithm is used for state estimation. In discrete time... The nonlinear state transition function of the system and measurement function It can be represented as: ; ; in, It is the control input from the previous moment. It is a combined measurement vector containing GPS / IMU and FBG readings provided by the heterogeneous sensing and data acquisition module 10 at the current moment; and These are process noise and measurement noise, respectively.

[0053] The UKF algorithm uses a set of deterministically sampled Sigma points to transmit the mean and covariance of the state distribution, thus avoiding the errors introduced by linearizing the nonlinear function. The algorithm periodically performs two steps: prediction and update, outputting the optimal estimate of the augmented state vector at the current time step at the end of each cycle. and its covariance matrix.

[0054] Step 2: Perform online adaptive correction of model parameters based on strain residuals. Since the offline-established ROM may deviate from the actual system due to changes in operating conditions, structural aging, or environmental temperature, the online state estimation and model adaptation module 30 includes an online model correction mechanism. This mechanism uses high-precision FBG strain measurements as a benchmark to fine-tune the key parameters of the ROM.

[0055] Specifically, in the UKF update step, the online state estimation and model adaptation module 30 calculates the predicted strain value based on the current state estimate. Simultaneously, the predicted value is compared with the actual measured strain value provided by the heterogeneous sensing and data acquisition module 10. By comparison, the strain residual vector is obtained. .

[0056] When the norm of the residual vector consistently exceeds a preset threshold, a parameter identifier based on Recursive Least Squares (RLS) is activated. This parameter identifier updates the parameters in the ROM that have the greatest impact on model accuracy, such as the diagonal elements related to modal stiffness. It is assumed that the parameters to be identified constitute a parameter vector. Its update rule is: ; ; ; in, It is a system parameter vector. This is the system parameter vector from the previous time step. It is the strain prediction residual. It is Kalman gain. It is the covariance matrix of the parameter estimates. The covariance matrix of the parameter estimates from the previous time step. It is a forgetting factor. It is the Jacobian matrix of strain prediction with respect to parameters. The standard mathematical notation represents the transpose operation of a matrix. It is an identity matrix.

[0057] Updated parameters This will be immediately used to reconstruct the ROM's state matrix, generating a set of dynamic model matrices that are corrected in real time. .

[0058] Step 3: Identify and solve unforeseen disturbances based on strain field topology analysis. The purpose of this step is to identify external disturbances that are difficult to measure directly by sensors, such as uneven settlement of a single support foundation. The online state estimation and model adaptation module 30 first collects the strain readings of all FBG sensors at the current moment, forming a high-dimensional real-time strain field vector.

[0059] Subsequently, a pattern recognition algorithm, such as Principal Component Analysis (PCA), is used to reduce the dimensionality and extract features from the real-time strain field vector, and then compare it with a strain field pattern pre-stored under normal operating conditions. If the topological characteristics of the current strain field significantly deviate from the normal pattern, it is determined that the system has been subjected to an unforeseen disturbance.

[0060] When a disturbance is identified, the online state estimation and model adaptation module 30 activates an inverse problem solver. This solver utilizes the disturbance-strain field sensitivity matrix pre-calibrated by the offline modeling and calibration module 20. This is used to inversely calculate the equivalent perturbation force causing the current abnormal strain field distribution. The solution process can be expressed as: ; in, It is the deviation vector between the current measured strain field and the normal reference strain field. The standard mathematical notation represents the transpose operation of a matrix. The calculated equivalent perturbation vector... It contains information about the location of the disturbance and the magnitude of the force / torque.

[0061] Finally, at the end of each control cycle, the online state estimation and model adaptation module 30 outputs three information packets to the multi-constraint predictive control decision module 40: the optimal augmented state estimate. The dynamic model matrix after real-time correction ( ), and the calculated equivalent disturbance feedforward vector .

[0062] like Figure 5 As shown, Figure 5 This is a schematic diagram of the multi-constraint predictive control decision module 40 according to an embodiment of the present invention. The multi-constraint predictive control decision module 40 is the core decision software module deployed in the vehicle-mounted industrial controller. The function of this module 40 is to receive real-time state, correction model, and disturbance information from the online state estimation and model adaptation module 30, and, in conjunction with an externally set target attitude, generate control commands that optimally drive the bridge-building machine's attitude and ensure structural safety by solving an optimization problem with multiple constraints online.

[0063] The multi-constraint predictive control decision module 40 receives three information packets from the online state estimation and model adaptation module 30 at each control cycle: the optimal augmented state estimate at the current time. The dynamic model matrix after real-time correction ; and the calculated equivalent disturbance feedforward vector .

[0064] Based on the received information, the multi-constraint predictive control decision module 40 first constructs an optimization problem in a finite time domain. The goal of this problem is to find a future control time domain... Control input sequence within This makes the cost function of a comprehensive performance index... To reach a minimum. The specific form of this cost function is: ; in, For prediction in the time domain; At the current moment Predicted future Step control input; This corresponds to the system output prediction value; It is a preset reference attitude trajectory; The last one (i.e., the last) The system terminal outputs the predicted value (step). It is the terminal reference trajectory point at the very end of the prediction time domain; It is a positive definite weight matrix, which is used to adjust the penalty weights for attitude tracking error, control energy consumption and terminal state error, respectively.

[0065] In solving the above optimization problem, a series of strict constraints must be met to ensure the physical realizability and absolute security of the control.

[0066] The first constraint is a system dynamics constraint. At each step in the prediction time domain, the system's state evolution must follow a corrected discrete state-space model, including perturbation feedforwards, provided by the online state estimation and model adaptation module 30. ; in, In the current control cycle For the future The predicted value of the system state vector at each step. The state matrix is ​​corrected in real time. The input matrix after real-time correction For at any time For the future The state prediction value of the step, This is a distribution matrix that maps the equivalent disturbance force to the system state. This constraint ensures that control decisions are based on accurate predictions of the system's future behavior.

[0067] The second constraint is the control input constraint. This constraint reflects the performance limits of the physical actuator; for example, the extension speed or output torque of a hydraulic actuator is limited to a fixed range. The mathematical form of this constraint is: ; in, and These are the lower and upper bound vectors that control the allowed inputs, respectively.

[0068] The third constraint is the state and output constraint. This constraint is used to limit the key state quantities (such as the vibration amplitude of specific parts) and output quantities (such as the attitude angle deviation of the main beam) of the bridge-building machine during operation to within the allowable process range.

[0069] The fourth constraint is a structural safety constraint. To prevent structural damage due to control actions or external disturbances, this constraint requires that the predicted stress values ​​of critical components of the bridge-building machine must not exceed the allowable stress of its materials throughout the entire prediction time domain. This constraint utilizes a stress-state mapping function pre-calibrated by the offline modeling and calibration module 20. To achieve: ; in, The predicted stress of key components is calculated based on the state prediction values. This is a preset allowable safety limit for material stress. This constraint directly incorporates structural safety into the scope of control decisions.

[0070] Combining the above cost function and all constraints, a quadratic programming (QP) problem is formed. The multi-constraint predictive control decision module 40 calls an efficient numerical optimization solver (such as the interior point method or the effective set method) in each control cycle to solve the QP problem online.

[0071] To ensure the real-time requirements of each control cycle, the numerical optimization solver is configured to return a solution within a defined time. As one implementation, open-source or commercial QP solvers specifically designed for embedded model predictive control, such as OSQP and qpOASES, can be used. These solvers are optimized for real-time applications. During implementation, the maximum number of iterations of the solver needs to be limited, and its worst-case execution time (WCET) must be significantly shorter than the sampling period of the control system (e.g., if the system sampling period is 100 milliseconds, the solver must complete the calculation within 50 milliseconds) to ensure stable output of control commands.

[0072] After the solution is completed, the optimal control input sequence is obtained. Based on the rolling time-domain principle of model predictive control, the multi-constraint predictive control decision module 40 extracts only the first control command of the sequence. The system then uses this as the final control decision for the current moment and outputs it to the execution and feedback module 50. In the next control cycle, the system will repeat the entire optimization process based on the new measurement value.

[0073] Please see Figure 6 , Figure 6 This is a closed-loop schematic diagram of the execution and feedback module 50 in an embodiment of the present invention. The execution and feedback module 50 is the interface between the control system of the present invention and the physical entity of the bridge-building machine. The function of the execution and feedback module 50 is to receive and accurately execute the high-level control commands from the multi-constraint predictive control decision module 40, and complete the control closed loop of the entire system with the physical actions generated by these commands.

[0074] The specific functional implementation of the execution and feedback module 50 includes the following steps: Step 1: Parse and issue control commands. The execution and feedback module 50 receives the optimal control command vector from the multi-constraint predictive control decision module 40. This vector is an abstract numerical vector, where each component corresponds to a target control quantity of a hydraulic actuator, such as the target extension speed (in meters per second) or the target output force (in Newtons).

[0075] The execution and feedback module 50 contains an instruction parsing unit. This unit converts the abstract target control quantity into a physical electrical signal that the underlying hardware driver unit can recognize, based on preset hardware calibration parameters. For example, if the actuator is a hydraulic cylinder controlled by an electro-hydraulic proportional servo valve, the instruction parsing unit will output a target speed value. Through a mapping function Converted into a specific voltage or current signal value : ; The mapping function During the system debugging phase, experimental calibration revealed a nonlinear relationship between the servo valve's orifice opening and the output flow rate (which in turn determines the hydraulic cylinder speed). The generated electrical signal was analyzed. It was then sent to the corresponding hydraulic valve actuator.

[0076] Step 2: Execution of physical actions. After receiving electrical signals from the command parsing unit, the hydraulic valve actuators on the bridge-building machine precisely drive the electro-hydraulic proportional servo valves. Based on the magnitude and polarity of the signal, the servo valves control the flow rate and direction of the high-pressure oil output from the hydraulic pump station into the hydraulic cylinders.

[0077] Controlled hydraulic fluid flow drives the piston rod of the hydraulic cylinder to extend or retract, thereby achieving precise adjustment of the height of a single support leg of the bridge-building machine, or driving the movement of other related mechanisms. The coordinated actions of all hydraulic actuators together constitute the final physical adjustment of the overall posture of the bridge-building machine.

[0078] Step 3: Constructing the physical feedback closed loop of the system. The physical actions of the execution and feedback module 50 directly change the spatial pose (position and attitude angle) of the bridge-building machine and its own structural stress state. This new physical state will be immediately measured by the heterogeneous sensing and data acquisition module 10 installed on the bridge-building machine.

[0079] Specifically, GPS-RTK and IMU capture the changes in rigid body motion caused by the executed actions, while the FBG sensor array synchronously measures the changes in structural strain caused by attitude adjustment and stress redistribution. This new sensor data acquired by the heterogeneous sensing and data acquisition module 10 will serve as the initial input for the next control cycle, entering the online state estimation and model adaptation module 30.

[0080] In this way, the output (physical action) of the execution and feedback module 50 becomes the input (new sensor measurement value) of the entire control system through the bridge-building machine itself, thus forming a complete and closed physical information feedback loop, enabling the system to continuously make the next round of rolling optimization decisions based on the actual execution effect.

[0081] To further illustrate the collaborative working process of the technical solution of this invention, a specific working scenario example will be used below.

[0082] This embodiment describes a scenario where a heavy-duty bridge erecting machine is performing a pier-crossing operation, lifting a precast box girder from a girder transport vehicle and moving it forward to a target pier. During this process, the front support leg of the bridge erecting machine needs to cross an existing pier. The operating environment is subject to continuous lateral wind loads, and there is a slight, unknown uneven settlement of the foundation in the area where the front support leg lands.

[0083] Before the operation begins, the operator sets the target attitude trajectory for this operation through the human-machine interface. This means that throughout the entire pier crossing process, the lateral and longitudinal inclination angles of the bridge-building machine's main beam must remain near zero degrees to ensure smooth transport of the box girder. The initial dynamic model, stress mapping function, and disturbance sensitivity matrix generated by the offline modeling and calibration module 20 have been pre-loaded into the system.

[0084] Once the crossing operation is started, the system enters a real-time closed-loop control state.

[0085] Within any given control cycle, the heterogeneous sensing and data acquisition module 10 operates in real time. Its GPS-RTK and IMU units, deployed on the main girder, measure the kinematic information of the bridge-building machine's overall forward movement. Simultaneously, fiber Bragg grating (FBG) sensor arrays arranged along the main girder and columns measure the complex strain distribution resulting from the combined effects of the box girder's self-weight, wind load, and foundation settlement. All data, after internal time synchronization and preprocessing, is sent to the core controller.

[0086] The online state estimation and model adaptation module 30 within the controller receives this data frame. First, its internal unscented Kalman filter (UKF) algorithm fuses the kinematic data from GPS / IMU and the structural strain data from FBG to accurately estimate the augmented state vector at the current moment, which includes the six-degree-of-freedom pose of the entire machine and the multi-order elastic modal coordinates of the main beam.

[0087] Simultaneously, the disturbance identification unit of the online state estimation and model adaptation module 30 analyzes the topology of the real-time acquired FBG strain field. This disturbance identification unit discovers a prominent, fixed deviation component between the current strain field distribution pattern and the pattern under conditions solely caused by self-weight and known wind loads. By matching and inversely solving this deviation component with a pre-calibrated disturbance sensitivity matrix, the system identifies that this deviation is caused by ground settlement at a certain support leg at the front, and calculates the equivalent vertical disturbance force.

[0088] Subsequently, the online state estimation and model adaptation module 30 packages the optimal state estimate, the real-time corrected system dynamics model, and the calculated foundation settlement equivalent disturbance feedforward signal together and sends them to the multi-constraint predictive control decision module 40.

[0089] The multi-constraint predictive control decision module 40 receives the above information. Based on this information, it constructs and solves a model predictive control (MPC) optimization problem. In the prediction step, the multi-constraint predictive control decision module 40 uses the corrected model and incorporates feedforward terms for wind load and foundation settlement disturbance to predict that, in the future, if optimal control is not applied, the main beam of the bridge-building machine will tilt due to foundation settlement, and under wind load, the stress on one side of the main beam will approach the safety limit.

[0090] In the optimization solution process, the multi-constraint predictive control decision module 40 prioritizes maintaining the main beam's horizontal position, while simultaneously treating the operating speed, stroke range, and predicted stress values ​​at key structural points of all hydraulic actuators as strict constraints. Through numerical solution, the multi-constraint predictive control decision module 40 calculates an optimal sequence of control commands for the coordinated action of multiple support legs. These control commands not only compensate for normal center of gravity shifts but, more importantly, proactively instruct the hydraulic cylinders beneath the support legs experiencing settlement to extend at an additional rate to counteract the settlement effect. Simultaneously, it instructs the hydraulic cylinders on the windward and leeward sides to perform slight differential adjustments to resist the overturning moment generated by wind loads.

[0091] Finally, the multi-constraint predictive control decision module 40 sends the first instruction of the optimal control sequence to the execution and feedback module 50. The execution and feedback module 50 interprets it as a specific voltage signal, drives the corresponding electro-hydraulic servo valve, and precisely controls the movement of each hydraulic cylinder.

[0092] The results of the above actions, namely the new posture and stress state of the bridge-building machine, are captured by the heterogeneous sensing and data acquisition module 10 in the next control cycle, thus forming a continuous rolling optimization closed loop.

[0093] In this embodiment, the system of the present invention achieves effective identification and proactive compensation for known and unknown composite disturbances through deep fusion of multi-sensor information and online adaptive cognition. By introducing hard constraints on future structural safety into the control decision, it ensures that the bridge-building machine can complete its tasks with high precision under complex working conditions while also guaranteeing its own structural safety.

[0094] 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 real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion, characterized in that, include: The heterogeneous sensing and data acquisition module is used to collect the kinematic data and structural strain data of the bridge-building machine in real time and generate synchronous sensor data. The offline modeling and calibration module is used to establish and output the initial dynamic model and key mapping relationships based on the design parameters of the bridge-building machine; The online state estimation and model adaptation module is used to receive the synchronous sensor data and, in combination with the initial dynamic model and key mapping relationship, generate comprehensive state information including the augmented state estimate at the current moment, the real-time corrected dynamic model, and the equivalent perturbation force feedforward signal. The multi-constraint predictive control decision module is used to solve the multi-constraint optimization problem online based on the comprehensive state information and the preset target attitude trajectory, so as to generate the optimal control command; The execution and feedback module is used to receive and parse the optimal control command to drive the actuator on the bridge building machine to produce physical movements, thereby adjusting the attitude of the bridge building machine; The offline modeling and calibration module is specifically used for: Based on the three-dimensional computer-aided design drawings of the bridge-building machine, a finite element model containing the geometric structure, material properties, and connection relationships between components of the bridge-building machine is established. Modal analysis and model order reduction are performed on the finite element model to generate the initial dynamic model in state-space form that describes the dynamic characteristics of the system; Offline simulations were performed on the finite element model to calibrate a structural stress-state mapping function and an external disturbance-strain field sensitivity matrix, which served as the key mapping relationship.

2. The real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion as described in claim 1, characterized in that, The heterogeneous sensing and data acquisition module specifically includes: A real-time dynamic differential positioning receiver of the global satellite navigation system is installed at the main beam stiffness position of the bridge-building machine to measure the position data of the bridge-building machine in the global coordinate system; An inertial measurement unit is installed close to the real-time dynamic differential positioning receiver to measure the acceleration and angular velocity data of the bridge-building machine in the machine coordinate system. A multi-channel fiber Bragg grating sensor array is distributed along the main beam and supporting columns of the bridge-building machine to measure the strain data of the structure. The data synchronization and preprocessing unit is used to perform high-precision time synchronization, temperature compensation, and filtering preprocessing on the kinematic data and the structural strain data to generate the synchronized sensor data.

3. The real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion as described in claim 1, characterized in that, The online state estimation and model adaptation module is specifically used for: An unscented Kalman filter algorithm is used to fuse the kinematic data and the structural strain data to estimate the augmented state vector containing the rigid motion and flexible deformation of the bridge-building machine, which is then used as the augmented state estimate. The strain residual between the measured value of the structural strain data and the predicted value obtained based on the augmented state estimate is calculated, and when the strain residual exceeds a preset threshold, the parameters of the initial dynamic model are updated to generate the real-time corrected dynamic model. The real-time distribution field of the structural strain data is analyzed, and when the real-time distribution field deviates from the normal pattern, the external disturbance-strain field sensitivity matrix in the key mapping relationship is used to solve in reverse to calculate the equivalent disturbance force feedforward signal.

4. The real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion according to claim 1, characterized in that, The multi-constraint predictive control decision module is specifically used for: Construct a cost function to minimize attitude tracking error and control energy consumption in the future prediction time domain; A set of constraints is set, including: system dynamic constraints that follow the real-time correction dynamic model in the comprehensive state information, control input constraints that reflect the physical limits of the actuator, and structural safety constraints that ensure that the predicted stress values ​​of key parts do not exceed the allowable stress limits. The cost function and the set of constraints are used to form a quadratic programming problem. The quadratic programming problem is solved online to obtain the optimal control input sequence, and the first control command of the optimal control input sequence is extracted as the optimal control command.

5. The real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion according to claim 1, characterized in that, The execution and feedback module is specifically used for: The optimal control command is parsed into one or more physical electrical signals that drive the actuator according to preset hardware calibration parameters; The physical electrical signal is sent to the corresponding hardware driver unit to control the action of the actuator, thereby completing the physical adjustment of the bridge-building machine's posture.

6. The real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion according to claim 3, characterized in that, The augmented state vector includes: The bridge-building machine's three-dimensional position vector and three-dimensional velocity vector in the global coordinate system, the bridge-building machine's attitude Euler angle vector and angular velocity vector in the body coordinate system, and the modal coordinate vector and first-order rate of change vector of a preset order describing the flexible deformation of the bridge-building machine's main beam.

7. The real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion according to claim 3, characterized in that, The parameters for updating the initial dynamic model include: A parameter identifier based on recursive least squares is used to update the matrix parameters related to modal stiffness in the initial dynamic model online using the strain residuals.

8. The real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion according to claim 3, characterized in that, The step of using the external disturbance-strain field sensitivity matrix to solve in reverse order to calculate the equivalent disturbance force feedforward signal includes: The strain deviation vector is obtained by subtracting the real-time strain field vector at the current moment from the reference strain field vector under normal operating conditions. The strain deviation vector is then multiplied by the generalized inverse of the external disturbance-strain field sensitivity matrix to calculate the equivalent disturbance force feedforward signal.

9. The real-time control system for the traveling posture of a bridge-building machine based on multi-sensor fusion according to claim 4, characterized in that, The structural safety constraints include: At each step within the prediction time domain, the predicted system state value is mapped to the predicted stress value of the key component using the structural stress-state mapping function in the key mapping relationship. The predicted stress value is always less than or equal to the preset allowable stress safety limit for the material.

Images