Assembly method of steel structure fabricated building

Abstract

The invention relates to the technical field of intelligent construction, and discloses a construction steel structure fabricated building assembly method which comprises the steps that real-time pose data of a steel structure component is collected through a multi-source sensor array; performing multi-rate fusion processing on the pose data to generate six-dimensional state estimation containing position and pose errors; constructing an error matrix in the Lie group SE (3) space based on the state estimation, wherein the error matrix represents the deviation between the component pose and the target pose; and establishing a matrix differential equation model according to the error matrix, and solving an optimal compensation control quantity in a future time domain in a rolling manner through a model prediction control algorithm. Through combination of Lie group space error modeling and rolling time domain optimization, the system can sense the pose deviation of a component in real time and dynamically adjust a control strategy, limitation of traditional static error compensation and online parameter identification and disturbance suppression under complex working conditions are broken through, and uncertainty caused by dynamic changes of a construction environment is effectively dealt with.

Description

Technical Field

[0001] The present invention relates to the field of intelligent construction technology, and in particular to an assembly method for a steel structure assembled building. Background Art

[0002] In the field of building steel structure assembly, traditional control methods are mostly based on the principle of static error compensation, which makes it difficult to adapt to the dynamic changes in complex construction environments.

[0003] Existing technologies typically employ independent degree-of-freedom control strategies, lacking systematic modeling of the coupling effects of component spatial positions, making it difficult to effectively eliminate cumulative errors. Conventional PID control methods suffer from adjustment lag when dealing with time-varying parameters and nonlinear disturbances, and parameter tuning relies on manual experience, making it difficult to meet high-precision assembly requirements.

[0004] Furthermore, existing systems often rely on fixed control models, unable to track changes in component physical properties in real time and prone to precision drift over long-term operations. Furthermore, traditional methods lack effective decoupling methods for the coupled effects of spatial deformation and temporal drift, resulting in a single control resource allocation strategy and difficulty achieving multi-objective collaborative optimization. These technical bottlenecks severely restrict the construction efficiency and quality stability of prefabricated buildings, particularly in complex engineering scenarios such as large spans and special-shaped structures. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides an assembly method for a prefabricated steel structure building, which solves the problems of poor adaptability of traditional assembly methods in dynamic environments, difficulty in compensating for space-time error coupling, and insufficient parameter self-tuning capabilities.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for assembling a steel structure prefabricated building, comprising the following steps: S1, collect real-time position data of steel structure components through a multi-source sensor array; S2. Perform multi-rate fusion processing on the pose data to generate a six-dimensional state estimate including position and pose errors; S3, constructing an error matrix in the Lie group SE(3) space based on the state estimation, wherein the error matrix represents the deviation between the component posture and the target posture; S4. Establishing a matrix differential equation model based on the error matrix, and solving the optimal compensation control quantity in the future time domain by a model predictive control algorithm; S5, decomposing the optimal compensation control amount into a spatial deformation compensation component and a temporal drift compensation component, and driving the actuators to perform error correction respectively; S6. Based on historical error data and control variable feedback, update the parameters of the matrix differential equation model online to form closed-loop adaptive control.

[0007] Preferably, in step S1: The multi-source sensor array includes a laser tracker, an inclinometer and a strain gauge, and each sensor achieves clock synchronization through the IEEE1588 protocol, wherein the sampling rate of the laser tracker is not less than 100 Hz.

[0008] Preferably, the step S2 implements multi-rate data fusion and state estimation through the following process: Construct a discrete-time state space model and define the six-dimensional state vector as: ; in, For the system at time The state vector of , which contains position and attitude errors; 、 and are the three-dimensional translation error components, respectively. 、 and Displacement deviation in the axial direction; 、 and are the three-dimensional rotation error components, representing the rotation around 、 and Axis attitude angle deviation; is the transposition symbol, which means converting a row vector into a column vector; For each sensor observation data, local state estimation is performed through an independent Kalman filter, and the calculation process satisfies: ; in, For the moment No. The Kalman gain matrix of each sensor; For the moment The state covariance matrix prediction value of ; For the The observation matrix of the sensors; For the The measurement noise covariance matrix of the sensors; is the matrix inversion operator; For the moment No. The updated state estimate of each sensor; For the moment The estimated value of the predicted state; For the moment No. The actual measurement value of each sensor; is the residual between the predicted state and the measured value; The estimation results of each sensor are fused and the global optimal estimate is generated through the covariance intersection method.

[0009] Preferably, in step 2, the estimation results of each sensor are fused by a covariance intersection method, which specifically includes: calculating the inverse covariance weights of the local estimates of each sensor, constructing a weighted synthesis of the global optimal estimate, and updating the global covariance matrix.

[0010] Preferably, the step 3 of constructing the error matrix in the Lie group SE(3) space specifically includes: Map the rotation components in the six-dimensional state estimation to Lie algebra rotation vectors, and satisfy: ; in, is the attitude deviation expressed by quaternion; is the Lie algebra rotation vector; Extract the translation component as a Lie algebraic translation vector; Construct an antisymmetric representation of the error matrix: ; in, 、 and For a rigid body 、 and Angular velocity components of the axis; 、 and For rigid body along 、 and The linear velocity component of the shaft; is the Lie algebra of the special Euclidean group SE(3), and is the tangent space of SE(3) at the identity element.

[0011] Preferably, the step S4 obtains the optimal compensation control amount through the following process: Formulate the continuous-time differential equation for the error matrix: ; in, is the system state matrix; is the control input matrix; is the state vector; is the control input vector; Discretize the differential equation and obtain the recursive formula: ; in, is the discrete state vector; is the discrete system matrix; is the discrete control matrix; is the discrete control input; Construct a rolling horizon optimization problem: ; in, is the control sequence; is the prediction time domain, a positive integer representing the number of steps considered in the optimization problem; is the predicted state; is the Frobenius norm; is the weighted norm; To predict the time domain The control input vector of the step; The optimal control sequence is calculated in real time through the quadratic programming solver, and the first control variable is taken as the compensation instruction at the current moment.

[0012] Preferably, step S5 performs error correction through the following process: The spatial component The antisymmetry of characterizes the non-rigid error caused by the elastic deformation of the component, and then the error matrix corresponding to the optimal compensation control quantity is obtained. Extract the spatial deformation component from: ; in, is the error matrix in Lie algebra, which represents the deviation between the component pose and the target pose; is a matrix The transpose of is the antisymmetric matrix component; The symmetry of the time component characterizes the rigid pose deviation caused by the cumulative assembly error, and the time drift component is then calculated through the matrix residual: ; in, is a symmetric matrix component; Design independent control channels for error components with different physical characteristics: ; in, is the spatial deformation compensation control amount; is the control amount for time drift compensation; is the control gain matrix; is the control gain matrix; This is a matrix vectorized operation.

[0013] Preferably, the online updating of the parameters of the matrix differential equation model in step 6 includes: Maintenance includes recent Sliding window of group history data: ; in, is the historical error matrix; is the corresponding control quantity; For the dataset; is the sliding window length; For the current moment; is the time index; Construct parameter identification equation: ; in, is the observation output matrix; is the regression matrix; is the parameter vector to be identified; is the noise vector; Update the system matrix using the recursive least squares method with forgetting factor: ; in, is the Kalman gain matrix; is the error covariance matrix; is the transpose of the design matrix; For the forgetting factor; is the state vector; is the state vector at the previous moment; is the observation vector; is the identity matrix; is the updated error covariance matrix; is the time step index; is the error covariance matrix before updating; The updated parameters are fed back to the model predictive controller.

[0014] Preferably, forming a closed-loop adaptive control in step 6 includes: Collect the posture feedback data of the actuator after compensation in real time and calculate the residual; Update the model confidence factor based on the residual sequence; Dynamically adjust the prediction model weights; The updated system matrix is ​​input into the model predictive controller for rolling optimization.

[0015] An assembly system for a steel structure prefabricated building, comprising: Data acquisition module, which acquires component posture data and environmental parameters through a multi-source sensor array; Data fusion module for multi-rate Kalman filtering and spatiotemporal alignment of heterogeneous sensor data; Error modeling module, used to construct the dynamic error matrix in Lie group SE(3) space; Predictive control module, used to execute the rolling horizon optimization algorithm to generate the optimal compensation strategy; An execution drive module, used to decouple spatial deformation and temporal drift components and drive the six-degree-of-freedom platform; The parameter updating module is used to modify the system matrix parameters online through the recursive least squares method.

[0016] The present invention provides a method for assembling a steel structure prefabricated building. The method has the following beneficial effects: 1. The present invention combines Lie group spatial error modeling with rolling time domain optimization, enabling the system to perceive component posture deviations in real time and dynamically adjust the control strategy, thereby breaking through the limitations of traditional static error compensation and achieving online parameter identification and disturbance suppression under complex working conditions, effectively coping with the uncertainties brought about by dynamic changes in the construction environment.

[0017] 2. The present invention adopts a spatiotemporal error decoupling and compensation mechanism to independently process spatial deformation and temporal drift components. By designing a feedforward-feedback composite control structure, it simultaneously optimizes geometric accuracy and cumulative error, solving the technical problem of balancing rigid positioning and elastic deformation in traditional methods.

[0018] 3. The present invention constructs a closed-loop adaptive system including neural network compensation, which effectively integrates the advantages of model-driven and data-driven, improves the tolerance to unmodeled dynamics and sensor noise, and ensures system stability when some sensors fail or parameters suddenly change.

[0019] 4. The present invention adopts an optimization solution set generation method based on the Nash equilibrium strategy to achieve an intelligent trade-off between control accuracy and energy consumption, breaking through the traditional empirical parameter adjustment mode. It automatically adjusts the control intensity of each degree of freedom through a dynamic weight distribution mechanism, reducing actuator wear while ensuring accuracy.

[0020] 5. This invention forms a full-process digital control system of "perception-decision-execution", provides standardized intelligent solutions for prefabricated buildings, promotes the transformation of traditional construction methods to data-driven ones, and provides core technical support for the modernization of the construction industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A diagram showing the steps of the method of the present invention; Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, a method for assembling a steel structure prefabricated building may include the following steps: S1, collect real-time position data of steel structure components through a multi-source sensor array; S2. Perform multi-rate fusion processing on the pose data to generate a six-dimensional state estimate including position and pose errors; S3, constructing an error matrix in the Lie group SE(3) space based on the state estimation, wherein the error matrix represents the deviation between the component posture and the target posture; S4. Establishing a matrix differential equation model based on the error matrix, and solving the optimal compensation control quantity in the future time domain by a model predictive control algorithm; S5, decomposing the optimal compensation control amount into a spatial deformation compensation component and a temporal drift compensation component, and driving the actuators to perform error correction respectively; S6. Based on historical error data and control variable feedback, update the parameters of the matrix differential equation model online to form closed-loop adaptive control.

[0024] The following is a detailed description of each step in the method of the present invention, which comprehensively explains the specific implementation principles, technical details and processes of each step.

[0025] In step S1, in this embodiment, the multi-source sensor data acquisition process achieves full-dimensional perception of the steel structure component's position and posture by constructing a heterogeneous sensor network. Specifically, multimodal sensing units are deployed at key deformation monitoring points on the steel structure to be assembled. These sensing units include non-contact optical measurement devices, inertial measurement units, and strain sensors, forming a spatially distributed data acquisition topology.

[0026] Preferably, the non-contact optical measurement device utilizes a laser tracker array. Its optical path arrangement must adhere to the principle of three-point spatial positioning, establishing a component's spatial position measurement benchmark using at least three non-coplanar laser emitters. The incident angles of the laser trackers' beams are pre-calibrated to ensure that the transformation matrix between the measurement coordinate system and the assembly reference coordinate system can be analytically calculated. The measurement data output interface is configured in real-time streaming mode, with the sampling interval dynamically adjusted based on the deformation characteristics of the steel structure.

[0027] To measure attitude angles, this embodiment utilizes a measurement array of high-precision MEMS inclinometers. Each inclinometer is mounted at a characteristic stress point on the component, with its sensitive axis oriented according to the component's theoretical deformation mode. Measurement data is transmitted to the central processing unit via the CAN bus. Device identification codes are embedded in the transmission protocol to facilitate data source traceability.

[0028] For strain monitoring, this embodiment utilizes fiber Bragg grating (FBG) sensors to form a distributed measurement network. The sensors are spaced evenly along the component's primary stress direction. The mapping between the grating's central wavelength and the strain is determined through pre-conducted tensile calibration tests. The sensor data acquisition module incorporates a built-in temperature compensation algorithm to eliminate measurement drift caused by ambient temperature changes.

[0029] To achieve spatiotemporal consistency for multi-source data, this embodiment deploys a precise clock synchronization mechanism within the sensor network. Each sensor node incorporates an IEEE 1588 Precision Time Protocol slave clock, which achieves microsecond-level time synchronization via synchronization messages sent by a master clock node. A precise timestamp field is added to the data packet encapsulation, containing a pulse-per-second count and a nanosecond-level time offset.

[0030] Preferably, the data packet adopts a layered encapsulation structure: the physical layer complies with the Industrial Ethernet protocol specification; the transport layer includes a data checksum and sequence number; and the application layer payload includes a device ID, timestamp, measurement value, and status flag. The checksum is generated using a cyclic redundancy check algorithm to ensure the integrity of the data transmission.

[0031] In terms of data transmission channel design, this embodiment builds a dual-redundant communication network: the primary channel uses real-time Ethernet to transmit raw sampled data, while the backup channel transmits compressed key parameters via a wireless mesh network. A data consistency check mechanism is implemented between the two channels, automatically switching the transmission path when the packet loss rate of the primary channel exceeds a preset threshold.

[0032] The layout of the multi-source sensor array was pre-optimized through finite element simulation. The sensor placement met the following requirements: covering the component's maximum stress area, avoiding areas affected by welding, and ensuring spatial correlation between measurement data. The mounting bracket utilizes a magnetic quick-release structure, and its resonant frequency, as tested, is higher than the dominant vibration frequency of the steel structure, thus minimizing the introduction of measurement noise.

[0033] Regarding step S2, in this embodiment, the multi-rate data fusion process achieves spatiotemporal consistency fusion of heterogeneous sensor data by constructing a hierarchical filtering architecture. In specific implementation, a six-dimensional state space model containing translation and rotation error components is first established, and its state vector is defined as: ; in, For the system at time The state vector of , which contains position and attitude errors; , , are the three-dimensional translation error components, respectively. , , Displacement deviation in the axial direction; , , are the three-dimensional rotation error components, representing the rotation around , , Axis attitude angle deviation; is the transpose symbol, which means converting a row vector into a column vector.

[0034] The dimension setting of the state vector takes into account the degree of freedom characteristics of rigid body motion, ensuring that the posture error of the component can be fully described.

[0035] Preferably, in view of the differences in characteristics of various sensor types, this embodiment designs a multi-stage Kalman filter structure. For high-sampling-rate laser tracker data, a strong tracking filter is used for preprocessing. The covariance matrix is ​​adjusted by a fading factor to enhance the ability to track sudden changes. The prediction equation is expressed as: ; in, For the moment No. The Kalman gain matrix of each sensor is used to weight the measured and predicted values; For the moment The state covariance matrix prediction value of represents the uncertainty of the state estimation; For the The measurement matrix of each sensor maps the state space to the measurement space; For the The measurement noise covariance matrix of each sensor characterizes the sensor noise characteristics; is the matrix inversion operator; For the moment No. The updated state estimate of each sensor; For the moment The estimated value of the predicted state; For the moment No. The actual measurement value of each sensor; is the residual between the predicted state and the measured value, which is used to correct the state estimation.

[0036] For low-frequency strain data, a sliding window filtering algorithm is used to smooth the slow variables in the time domain.

[0037] In terms of data spatiotemporal alignment, this embodiment proposes a compensation method based on Lie group interpolation. When receiving asynchronous sensor data packets, a state transition matrix is ​​constructed based on the nearest neighbor timestamps: ; in, From the moment arrive The state transition matrix of is the time-varying system matrix; , is the endpoint of the time interval; is the matrix exponential operation.

[0038] This matrix is ​​used to uniformly map the observations of each sensor to the same reference time, eliminating the time delay error caused by sampling rate differences. The manifold properties of the SE(3) group are maintained during the interpolation process, avoiding the distortion of the posture representation caused by linear interpolation.

[0039] Preferably, the global state estimation uses an improved covariance cross-fusion algorithm. This method constructs an unbiased optimal estimate by calculating the inverse covariance weights of each local estimate: ; in, It is the inverse matrix of the covariance matrix of the fused global state estimate; For the The covariance matrix of the local estimates of the sensors; For the The fusion weight coefficient of each sensor; is the total number of sensors involved in the fusion.

[0040] The information entropy theory is introduced into the specific calculation, so that the sensor data with high confidence will receive a greater fusion weight. This process effectively suppresses the impact of a single sensor failure on the global estimation.

[0041] For heterogeneous data with different dimensions, this embodiment implements standardization preprocessing. The position error component is converted into a dimensionless form: ; in, is the dimensionless position error after normalization; is the original position error measurement value; is the reference characteristic length.

[0042] This process eliminates the influence of physical dimensions on the fusion results, making sensor data of different magnitudes comparable. The rotation component remains in radians, and quaternion normalization ensures the compactness of the attitude representation.

[0043] In the residual processing link, this embodiment designs an adaptive threshold detection mechanism. When the prediction residual of a certain sensor is: ; in, For the The sensors in The prediction residual vector at time t; For the moment No. The actual measurement value of each sensor; For the The observation matrix of the sensors; For the The sensors in Moment based on The prior state estimate of the data at time -1.

[0044] When the dynamically calculated confidence interval is exceeded, the sensor's fusion weight is automatically reduced, and the fault diagnosis procedure is triggered. This mechanism uses a sliding window to calculate the mean and variance of the residual sequence, updating the detection threshold in real time to improve system robustness.

[0045] Preferably, the fused global estimate is converted to an SE(3) group element via a Lie group exponential mapping, providing a mathematical representation consistent with the rigid body motion characteristics for subsequent error modeling. This conversion process maintains geometric consistency and avoids singularities in the Euler angle representation.

[0046] Regarding step S3, in this embodiment, the Lie group SE(3) space error modeling described in step S3 implements the canonical representation of the rigid body posture error through geometric algebra methods. In specific implementation, the rotation component in the six-dimensional state estimation is first mapped to the Lie algebra space, and its conversion relationship follows the exponential mapping law: ; in, is the actual quaternion obtained by solving the sensor data; is the theoretical target quaternion; is quaternion multiplication; It is the logarithmic mapping operation of Lie algebra.

[0047] This mapping process maintains the shortest path property of the rotation quantity and avoids the angle winding problem in the posture representation.

[0048] Preferably, the translation error component is directly derived from the 3D position deviation output by multi-source data fusion and, after normalization, serves as the translation vector of the Lie algebra. The normalization coefficient is dynamically adjusted based on the component size to ensure comparability of the error magnitudes for components of different scales.

[0049] When constructing the SE(3) error matrix, this embodiment uses an antisymmetric matrix to represent the spatial posture deviation: ; in, , , For a rigid body 、 、 Angular velocity components of the axis; , , For rigid body along 、 、 The linear velocity component of the shaft; is the Lie algebra of the special Euclidean group SE(3), and is the tangent space of SE(3) at the identity element.

[0050] This matrix form strictly satisfies the closure property of Lie algebra, ensuring the geometric consistency of subsequent differential operations.

[0051] When establishing the error dynamics model, this embodiment introduces a matrix differential equation with time-varying parameters: ; in, is the time-varying system matrix; is the control input matrix; is the disturbance term, used to characterize the unmodeled dynamics; is the derivative of the error matrix with respect to time, representing the error change rate; is the continuous-time control input vector.

[0052] The update frequency of the matrix elements is synchronized with the parameter identification module to achieve adaptive adjustment of the model.

[0053] Preferably, the discretization of the differential equation adopts the Lie group structure-preserving algorithm. When converting the continuous-time differential equation into a discrete form, the manifold properties of the SE(3) group are maintained: ; in, To control the cycle; is the matrix exponential operation; for The discretization error matrix at time +1; for The time-varying system matrix at moment ; for The control input matrix at time t; for The control input vector at time t; is a discrete time index, indicating the control cycle.

[0054] This discretization method avoids the deviation of group elements caused by the traditional Euler method and ensures the geometric correctness of the numerical calculation.

[0055] For the decomposition of the error matrix, this embodiment designs a decoupling method based on the symmetry of the matrix. The error matrix is ​​decomposed into antisymmetric components and symmetric components: ; in, is the complete SE(3) Lie algebra error matrix; This is a matrix transpose operation that swaps the row and column indices of the original matrix.

[0056] The antisymmetric component represents the non-rigid error caused by the elastic deformation of the component, and its physical meaning corresponds to the strain tensor in classical mechanics; the symmetric component reflects the rigid posture deviation caused by the cumulative assembly error, and has a corresponding relationship with the kinematic error in classical rigid body dynamics.

[0057] During the parameter initialization phase, the initial values ​​of the system matrix are obtained through finite element modal analysis, including information about the first six vibration modes of the component. The initial configuration of the control matrix is ​​determined based on the thrust-posture transfer function of the six-degree-of-freedom platform, ensuring a linear relationship between the control input and the posture adjustment.

[0058] Preferably, the estimation of the disturbance term adopts the sliding window least square method. Sliding window of group history data: ; in, is the historical error matrix; is the corresponding control quantity; For the dataset; is the sliding window length; For the current moment; is the time index.

[0059] The disturbance estimate is updated online by solving the optimization problem. This process effectively suppresses the impact of environmental disturbances on model accuracy and improves the robustness of the control system.

[0060] Regarding step S4, in this embodiment, the rolling horizon optimization control described in step S4 achieves accurate error compensation by building a framework that combines a prediction model with real-time optimization. In specific implementation, the continuous-time error dynamics equation is first discretized into a form suitable for digital control: ; in, is the discrete state vector; is the discrete system matrix; is the discrete control matrix; is the discrete control input; The discretization process maintains the Lie group structure characteristics and ensures the geometric consistency between the discrete model and the continuous model.

[0061] Preferably, the prediction model is built with time-varying parameters in mind. At the beginning of each control cycle, the latest system and control matrices are retrieved from the parameter update module, dynamically reconstructing the prediction model parameters. The model update frequency is synchronized with the sensor data acquisition rate to ensure prediction accuracy.

[0062] When constructing a rolling optimization problem, this embodiment designs a multi-objective cost function: ; in, For the prediction time domain; To control the time domain; , The weighted matrix adjusts the relative importance of the state error and the control quantity respectively; To optimize the objective function value, it needs to be minimized; For Predicted at all times Time state error matrix; For Calculated at all times Control the input vector at all times; for weighted quadratic norm; for Weighted quadratic norm.

[0063] Weighting Matrix , Its structural design meets: ; ; in, , , , , , is the weight coefficient corresponding to the six-dimensional pose error, is the weight coefficient corresponding to the control input of each actuator; is an operator that constructs a diagonal matrix.

[0064] The initialization value of the weight coefficient is determined by controllability Gram matrix analysis to ensure system stability.

[0065] In terms of solution strategy, this embodiment uses the active set method to solve the constrained quadratic programming problem. The constraints include: ; ; in, is the lower limit of the control input vector; is the upper limit of the control input vector; is the lower limit of the control variable change rate; is the upper limit of the control variable change rate; for The actual control input vector at time t; for The historical control input vector at time -1.

[0066] The first constraint reflects the physical limits of the actuator, while the second limits the rate of change of the controlled variable to avoid mechanical shock. The constraint boundaries are set based on the dynamic response characteristics of the actuator and are determined during the initialization phase through step response testing.

[0067] Preferably, real-time optimization calculations utilize hot start technology to accelerate the solution. The optimal solution from the previous control cycle is used as the initial guess for the current cycle, leveraging the path-following properties of the sequential quadratic programming algorithm to significantly reduce the number of iterations. If system parameters suddenly change, the system automatically switches to cold start mode to reinitialize the optimization variables.

[0068] To solve the model mismatch problem, this embodiment introduces a robustness compensation term. A slack variable is added to the cost function: ; in, is the cumulative amount of prediction error; is a weighting matrix that is adaptively adjusted according to the statistical characteristics of historical forecast errors; It is the matrix transpose operator, which converts a column vector into a row vector.

[0069] This mechanism enhances the controller's tolerance to model uncertainty and improves the reliability of the control system.

[0070] In the control quantity output link, this embodiment designs a feedforward-feedback composite structure. As the basic control quantity, the compensation amount of the disturbance observer output is superimposed: ; in, is the disturbance compensation gain matrix; is the disturbance term for online estimation.

[0071] This structure effectively suppresses the influence of unmodeled dynamics on the control effect and improves tracking accuracy.

[0072] Preferably, the control quantity distribution module shapes the command based on the dynamic characteristics of the actuator. The original control command is low-pass filtered, with a cutoff frequency set below the actuator's resonant frequency to avoid excitation of mechanical resonance. The filtered command is output to the drive circuit via a D / A conversion module, completing the digital-to-analog conversion.

[0073] Regarding step S5, in this embodiment, the temporal and spatial error decoupling compensation described in step S5 is implemented by Lie algebra decomposition and independent control channel design to achieve precise posture adjustment. Map from Lie group space to Lie algebra space and use the symmetry of the matrix to extract components: ; ; in, is the spatial deformation compensation, corresponding to the antisymmetric component of the error matrix; is the time drift compensation, corresponding to the symmetrical component; To expand the matrix into vectors by columns; To calculate the matrix trace; is the identity matrix; Transpose the matrix.

[0074] The decomposition process maintains the physical meaning of the control quantity and compensates for elastic deformation and accumulated error respectively.

[0075] Preferably, the spatial component control channel adopts a feedforward compensation structure. Input a pre-calibrated inverse dynamics model: ; in, is the inverse matrix of the mass matrix of the component; is the Coriolis force term; is the gravity compensation term.

[0076] This feedforward structure effectively offsets the nonlinear characteristics of the system and improves the response speed.

[0077] For time component compensation, this embodiment designs a PID control law based on integral separation: ; in, is the pose tracking error; is the sliding integration window length; , , is the gain matrix; is the error integral term within the sliding window.

[0078] The integral term uses a finite time window integration to avoid integral saturation.

[0079] In the control quantity synthesis phase, this embodiment proposes a dynamic weight allocation strategy: ; in, is the mixing coefficient, according to the error matrix The symmetry index is dynamically adjusted, and the adjustment rules are as follows: ; in, is the transposed matrix of the error matrix; is the Frobenius norm.

[0080] This metric reflects the contribution of spatial deformation to the total error in real time, enabling adaptive allocation of control resources. Frobenius norm calculation ensures dimensional consistency.

[0081] Preferably, the actuator drive module adopts a decoupling control architecture. The control instructions of the six-degree-of-freedom platform are decomposed into: ; in, is the control force in the translation direction; Control torque for the direction of rotation; , is the translation stiffness coefficient; , is the rotation compensation gain; is the control input in the translation direction; is the spatial deformation compensation Directional gradient; is the double integral of the spatial deformation compensation.

[0082] This structure achieves dynamic decoupling of translation and rotation, avoiding coupled vibration.

[0083] This embodiment employs a double-buffering mechanism for instruction transmission. Calculated control instructions are written to a ring buffer, and the driver thread periodically reads the latest instructions. The buffer size is dynamically adjusted based on network transmission latency to ensure accurate control timing.

[0084] Preferably, the state monitoring module of the actuator collects the driving current and position feedback in real time, and estimates the equivalent friction coefficient of each axis through the observer: ; in, is the command torque; is the moment of inertia; is the damping coefficient; is the torque constant.

[0085] The estimated friction coefficient is used for online compensation to improve the positioning accuracy of the actuator; For the executive body degrees of freedom; For the Angular acceleration of the axis; For the Angular velocity of the axis.

[0086] Regarding step S6, in this embodiment, the online parameter identification and dynamic compensation realizes real-time tracking of system characteristics by building a closed-loop adaptive mechanism. In specific implementation, firstly, an extended state space model including time-varying parameters is established: ; in, is the parameter vector to be identified, including physical parameters such as mass distribution characteristics and damping coefficient; is the measurement noise term; is the time derivative of the state vector; is the parameter-dependent system matrix; is the parameter-dependent control input matrix; is the control input vector; is the unmodeled dynamic disturbance; is the system observation output vector; is the observation matrix; is the system state vector.

[0087] The model incorporates the system parameter changes into the state equation and provides a mathematical basis for parameter estimation.

[0088] Preferably, the parameter identification adopts the recursive maximum likelihood estimation algorithm. Construct the augmented state vector containing the parameter sensitivity: ; in, is the system state vector; is the sensitivity vector of the state to the first parameter; is the total number of parameters to be identified; For the parameters to be identified.

[0089] The parameter sensitivity matrix is ​​obtained by solving the adjoint differential equation and the Fisher information matrix is ​​updated in real time. The convergence rate of this method is directly related to the identifiability of the parameters, ensuring unbiased estimation under observable system conditions.

[0090] This embodiment employs a parallel processing architecture to implement recursive computation. A main thread performs state prediction and parameter updates, while auxiliary threads calculate the sensitivity matrix and update the covariance. A double-buffering mechanism streamlines the computation process. Data synchronization utilizes a lock-free circular queue to avoid thread blocking.

[0091] Preferably, the covariance matrix is ​​updated using a square root filtering algorithm. The covariance matrix is ​​decomposed into: ; in, is a unit lower triangular matrix; is a diagonal matrix; is the transpose symbol.

[0092] This decomposition form ensures the stability of numerical calculations and prevents divergence problems caused by matrix pathology. The matrix elements impose lower bound constraints to avoid over-compression of the parameter space.

[0093] For tracking time-varying parameters, this embodiment introduces a fading factor adjustment mechanism. An exponential weighting factor is embedded in the covariance update equation: ; in, For the forgetting factor; for The posterior estimated covariance matrix at time t; for The prior covariance matrix of the time instant; for The Kalman gain matrix at time t; for The observation Jacobian matrix at time t.

[0094] The forgetting factor Adaptive adjustment based on parameter change rate. When a sudden change in parameter is detected, the value to speed up the tracking speed; in the steady state stage, increase Improve estimation accuracy.

[0095] In the disturbance compensation link, this embodiment designs a residual learner based on a neural network. The parameter estimation residual: ; in, for The residual vector of parameter estimates at time t; for The sensor measurement output vector at time t; is the observation matrix; for The estimated state at the moment.

[0096] The input is a recurrent neural network with a time delay link, and the network output is used as the compensation for the disturbance estimation: ; in, is the neural network mapping function, which represents the forward calculation process of the network; For Time has come The residual sequence at each moment constitutes the input feature vector.

[0097] This structure can learn the nonlinear characteristics of unmodeled dynamics and improve the accuracy of disturbance estimation. The network weights are updated online using stochastic gradient descent with momentum, balancing convergence speed and stability.

[0098] Preferably, the parameter confidence assessment module monitors the estimation quality in real time. Calculate the condition number of the parameter covariance matrix: ; in, is the covariance matrix of parameter estimates; is the matrix condition number operator; , are the maximum singular value of the matrix and the minimum singular value of the matrix, respectively.

[0099] When the condition number exceeds a preset threshold, parameter identifiability enhancement measures are triggered, including injecting auxiliary excitation signals and adjusting sensor sampling strategies. This mechanism ensures that the parameter estimation process is always in a good condition number state.

[0100] For parameter transfer, this embodiment employs a shared memory area with integrity checking. The latest parameter estimates are written to a dual-port memory with parity bits. The control module performs a CRC check when reading the data, preventing system instability caused by data transmission errors. A semaphore mechanism is used for access to the memory area to ensure atomicity of multi-threaded access.

[0101] For key safety parameters, this embodiment implements rate-of-change limit protection. When the sensor fails, it automatically switches to the safe parameter set and triggers an abnormality alarm. This protection mechanism prevents parameter estimation divergence caused by sensor failure and improves system reliability.

[0102] In general, the present invention combines Lie group spatial error modeling with rolling time domain optimization control to achieve high-precision dynamic adjustment of building steel structure assembly. The method includes: constructing a six-dimensional posture error model based on SE (3) Lie group, separating spatial deformation and time drift components through matrix decomposition; designing a multi-objective rolling optimization strategy, integrating the prediction model with real-time feedback to generate an optimal control sequence; establishing a parameter online identification mechanism, combining recursive estimation with neural network compensation for unmodeled dynamics; implementing a spatiotemporal decoupling compensation strategy, and performing feedforward-feedback composite control for elastic deformation and cumulative error respectively; finally, achieving real-time monitoring, dynamic compensation and parameter self-tuning of assembly errors through a closed-loop adaptive system, forming a full-process intelligent control system of "perception-decision-execution", and improving the precision stability and environmental adaptability of steel structure assembly under complex working conditions.

[0103] The assembly system of a steel structure prefabricated building described below and the assembly method of a steel structure prefabricated building described above can correspond to each other.

[0104] Please see the attached Figure 2 The present invention also provides an assembly system for a steel structure prefabricated building, comprising: Data acquisition module, which acquires component posture data and environmental parameters through a multi-source sensor array; Data fusion module for multi-rate Kalman filtering and spatiotemporal alignment of heterogeneous sensor data; Error modeling module, used to construct the dynamic error matrix in Lie group SE(3) space; Predictive control module, used to execute the rolling horizon optimization algorithm to generate the optimal compensation strategy; An execution drive module, used to decouple spatial deformation and temporal drift components and drive the six-degree-of-freedom platform; The parameter updating module is used to modify the system matrix parameters online through the recursive least squares method.

[0105] The system of this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, so they will not be repeated here.

[0106] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for assembling a steel structure prefabricated building, characterized in that: The following steps are involved: S1, collect real-time position data of steel structure components through a multi-source sensor array; S2. Perform multi-rate fusion processing on the pose data to generate a six-dimensional state estimate including position and pose errors; S3, constructing an error matrix in the Lie group SE(3) space based on the state estimation, wherein the error matrix represents the deviation between the component posture and the target posture; S4. Establishing a matrix differential equation model based on the error matrix, and solving the optimal compensation control quantity in the future time domain by a model predictive control algorithm; S5, decomposing the optimal compensation control amount into a spatial deformation compensation component and a temporal drift compensation component, and driving the actuators to perform error correction respectively; S6. Based on historical error data and control variable feedback, update the parameters of the matrix differential equation model online to form closed-loop adaptive control.

2. The method for assembling a steel structure prefabricated building according to claim 1, characterized in that: In the step S1: The multi-source sensor array includes a laser tracker, an inclinometer and a strain gauge, and each sensor achieves clock synchronization through the IEEE1588 protocol, wherein the sampling rate of the laser tracker is not less than 100 Hz.

3. The method for assembling a steel structure prefabricated building according to claim 1, wherein: Step S2 implements multi-rate data fusion and state estimation through the following process: Construct a discrete-time state space model and define the six-dimensional state vector as: ； in, For the system at time The state vector of , which contains position and attitude errors; 、 and are the three-dimensional translation error components, respectively. 、 and Displacement deviation in the axial direction; 、 and are the three-dimensional rotation error components, representing the rotation around 、 and Axis attitude angle deviation; is the transposition symbol, which means converting a row vector into a column vector; For each sensor observation data, local state estimation is performed through an independent Kalman filter, and the calculation process satisfies: ； in, For the moment No. The Kalman gain matrix of each sensor; For the moment The state covariance matrix prediction value of ; For the The observation matrix of the sensors; For the The measurement noise covariance matrix of the sensors; is the matrix inversion operator; For the moment No. The updated state estimate of each sensor; For the moment The estimated value of the predicted state; For the moment No. The actual measurement value of each sensor; is the residual between the predicted state and the measured value; The estimation results of each sensor are fused and the global optimal estimate is generated through the covariance intersection method.

4. The method for assembling a steel structure prefabricated building according to claim 3, characterized in that: In step 2, the estimation results of each sensor are fused by the covariance intersection method, which specifically includes: calculating the inverse covariance weights of the local estimates of each sensor, constructing a weighted synthesis of the global optimal estimate, and updating the global covariance matrix.

5. The method for assembling a steel structure prefabricated building according to claim 1, characterized in that: The step 3 of constructing the error matrix in the Lie group SE(3) space specifically includes: Map the rotation components in the six-dimensional state estimation to Lie algebra rotation vectors, and satisfy: ； in, is the attitude deviation expressed by quaternion; is the Lie algebra rotation vector; Extract the translation component as a Lie algebraic translation vector; Construct an antisymmetric representation of the error matrix: ； in, 、 and For a rigid body 、 and Angular velocity components of the axis; 、 and For rigid body 、 and The linear velocity component of the shaft; is the Lie algebra of the special Euclidean group SE(3), and is the tangent space of SE(3) at the identity element.

6. The method for assembling a steel structure prefabricated building according to claim 1, characterized in that: The step S4 obtains the optimal compensation control amount through the following process: Formulate the continuous-time differential equation for the error matrix: ； in, is the system state matrix; is the control input matrix; is the state vector; is the control input vector; Discretize the differential equation and obtain the recursive formula: ； in, is the discrete state vector; is the discrete system matrix; is the discrete control matrix; is the discrete control input; Construct a rolling horizon optimization problem: ； in, is the control sequence; is the prediction time domain, a positive integer representing the number of steps considered in the optimization problem; is the predicted state; is the Frobenius norm; is the weighted norm; To predict the time domain The control input vector of the step; The optimal control sequence is calculated in real time through the quadratic programming solver, and the first control variable is taken as the compensation instruction at the current moment.

7. The method for assembling a steel structure prefabricated building according to claim 1, characterized in that: Step S5 performs error correction through the following process: The spatial component The antisymmetry of characterizes the non-rigid error caused by the elastic deformation of the component, and then the error matrix corresponding to the optimal compensation control quantity is obtained. Extract the spatial deformation component from: ； in, is the error matrix in Lie algebra, which represents the deviation between the component pose and the target pose; is a matrix The transpose of is the antisymmetric matrix component; The symmetry of the time component characterizes the rigid pose deviation caused by the cumulative assembly error, and the time drift component is then calculated through the matrix residual: ； in, is a symmetric matrix component; Design independent control channels for error components with different physical characteristics: ； in, is the spatial deformation compensation control amount; is the control amount for time drift compensation; is the control gain matrix; is the control gain matrix; This is a matrix vectorized operation.

8. The method for assembling a prefabricated steel structure building according to claim 1, wherein: The online updating of the parameters of the matrix differential equation model in step 6 includes: Maintenance includes recent Sliding window of group history data: ； in, is the historical error matrix; is the corresponding control quantity; For the dataset; is the sliding window length; For the current moment; is the time index; Construct parameter identification equation: ； in, is the observation output matrix; is the regression matrix; is the parameter vector to be identified; is the noise vector; Update the system matrix using the recursive least squares method with forgetting factor: ； in, is the Kalman gain matrix; is the error covariance matrix; is the transpose of the design matrix; For the forgetting factor; is the state vector; is the state vector at the previous moment; is the observation vector; is the identity matrix; is the updated error covariance matrix; is the time step index; is the error covariance matrix before updating; The updated parameters are fed back to the model predictive controller.

9. The method for assembling a steel structure prefabricated building according to claim 1, characterized in that: The closed-loop adaptive control formed in step 6 includes: Collect the posture feedback data of the actuator after compensation in real time and calculate the residual; Update the model confidence factor based on the residual sequence; Dynamically adjust the prediction model weights; The updated system matrix is ​​input into the model predictive controller for rolling optimization.

10. An assembly system for a prefabricated steel structure building, used for executing an assembly method for a prefabricated steel structure building according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, which acquires component posture data and environmental parameters through a multi-source sensor array; Data fusion module for multi-rate Kalman filtering and spatiotemporal alignment of heterogeneous sensor data; Error modeling module, used to construct the dynamic error matrix in Lie group SE(3) space; Predictive control module, used to execute the rolling horizon optimization algorithm to generate the optimal compensation strategy; An execution drive module, used to decouple spatial deformation and temporal drift components and drive the six-degree-of-freedom platform; The parameter updating module is used to modify the system matrix parameters online through the recursive least squares method.

Images