Steel structure construction precision control system based on multi-data fusion

Abstract

The invention discloses a steel structure construction precision control system based on multi-data fusion, and particularly relates to the technical field of sensor data fusion. The system comprehensively collects initial processing size, real-time spatial displacement and installation environment data of a component through an integrated sensing network; respectively analyzing an initial deviation rate, a process deviation rate and an environment deviation rate; the intelligent prediction module predicts the primary pose deviation of the component in the installation stage; the intelligent decision-making module constructs a deviation quantification model according to the prediction result, and generates an operable pose correction decision; the execution and control module receives the decision instruction, drives an execution mechanism to complete actual adjustment operation and feeds back an execution state in real time; and the closed-loop verification and self-learning module carries out secondary pose measurement on the adjusted component and feeds back data to the model for incremental training, and finally, autonomous evolution and continuous improvement of a construction precision control system are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of multi-sensor data fusion, and more particularly to a steel structure construction precision control system based on multi-data fusion. BACKGROUND

[0002] Steel structure construction often uses automatic total station and digital level, and the two instruments usually work together to form a high-precision measurement control system for the project; the total station is responsible for the measurement and setting out of three-dimensional space coordinates, while the digital level is responsible for the transmission and settlement monitoring of precise elevation; taking steel column installation as an example to introduce the installation process, hoist the steel column, align the bottom hole with the anchor bolt, preliminarily position and temporarily fasten the nut, and paste or install the reflecting prism in two directions at 90 degrees on the upper part of the steel column. The prism position should be calculated so that its coordinates can accurately reflect the center and attitude of the steel column; set up the total station on the control point and direct the other control point; the total station will compare the measured prism coordinates with the design coordinates of the point in the BIM model in real time, and the software will automatically calculate the deviation of the steel column in X and Y directions and the perpendicularity deviation; the construction personnel adjust the steel column according to the displayed deviation data, use tools to accurately adjust and review and fix the steel column, and then measure the prism coordinates with the total station again to confirm that the deviation value meets the specification requirements.

[0003] However, it still has some shortcomings in actual use, (1) data chain fragmentation, lack of whole-process closed-loop management, the shortcomings are: data collection is scattered and discontinuous, and it is difficult to accurately locate the deviation in which link or introduced by which process; (2) data utilization is shallow, lack of intelligent fusion and traceability ability, data processing stays at the level of simple recording and comparison, multi-source data exists in isolation, and cannot be deeply fused and correlated; (3) passive control mode, lack of prediction and self-optimization ability, the control system lacks foresight and cannot predict future trends, can only passively correct deviation after deviation occurs, cannot dynamically optimize and adjust according to real-time working conditions on site, and has no ability to learn and evolve from historical data. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a steel structure construction precision control system based on multi-data fusion, which solves the problems in the above background technology by the following scheme.

[0005] Full state perception module: use an integrated sensing network to comprehensively collect component construction precision data including component initial processing size data, component real-time spatial displacement data and component installation environment influence data; Deviation calculation module: based on the component initial processing size data, component real-time spatial displacement data and component installation environment influence data, the initial deviation rate, process deviation rate and environment deviation rate of the component installation process are calculated respectively; Intelligent prediction module: based on the initial deviation rate, process deviation rate and environment deviation rate, the one-time pose deviation and total deviation rate of the component are predicted; Intelligent decision module: according to the predicted one-time pose deviation result of the component, a deviation quantization model is constructed, and a pose correction decision is generated; Execution and control module: receiving the pose correction decision, driving the execution device to complete the specific operation of the decision, and adjusting the completion of the decision operation through the built-in sensor feedback; Closed-loop verification and self-learning module: for the installed component after executing the decision operation, the pose information is collected again, the secondary pose deviation is identified, and the secondary pose deviation data is fed back to the model for incremental learning, continuously optimizing the pose deviation prediction algorithm and the pose deviation optimization decision algorithm.

[0006] Technical effects and advantages of the present application: 1. Whole-process, high-precision data closed loop: the present application constructs a precision data chain covering the whole life cycle of component installation. The data chain systematically defines and collects key precision data from before, during and after installation, and quantizes and correlates through unified standards, forming a complete, consistent and high-precision data closed loop; 2. Multi-source data fusion and intelligent deviation tracing: the present application realizes effective integration and deep analysis of various sources and different types of data, identifies various key factors affecting installation precision and their relative importance, and realizes the transition from simple data measurement to intelligent cause tracing; 3. Predictive regulation and continuous self-optimization: an intelligent closed-loop control system is established, which can actively predict future possible deviations based on the current state, and generate and execute optimized adjustment decisions in advance. The system will feed back the actual effect of each adjustment to itself for continuous learning and improvement. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 The figure is a schematic diagram of the overall structure of the present application.

[0008] Figure 2 The figure is a schematic diagram of the execution process of the present application. DETAILED DESCRIPTION

[0009] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some 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.

[0010] like Figure 1 The steel structure construction accuracy control system shown includes a full-state perception module, a deviation calculation module, an intelligent prediction module, an intelligent decision-making module, an execution and control module, and a closed-loop verification and self-learning module.

[0011] like Figure 2 The invention illustrates a steel structure construction accuracy control system based on multi-data fusion, which collects data during component installation, calculates the deviation rate, makes a posture correction decision based on the deviation rate, and collects data on the component again after the decision is executed.

[0012] Full-state perception module: Using an integrated sensor network, it comprehensively collects component construction accuracy data, including initial processing dimension data, real-time spatial displacement data, and environmental impact data of component installation.

[0013] The initial dimensional data of the components are obtained by scanning with a 3D laser scanner. Before the components arrive on site or are hoisted, the dimensional data of the components to be installed are acquired and compared with the dimensional data of the intended installation location to calculate the initial deviation rate δ. i .

[0014] A 3D laser scanner is used to perform a comprehensive scan of the component and the proposed installation location. It is important to note that the scan points must be sufficiently dense to cover all feature dimensions, and there should be no significant obstruction of these dimensions. In this example, the selected scan point spacing is 1mm. Feature dimensions refer to the digital representation of the geometric attributes of the component and the proposed installation location and their spatial relationship within the overall structure. A unified field control network coordinate system is established. Based on the 3D laser scanner data, precise point cloud data of the component and the proposed installation location are obtained. Through point cloud data processing and coordinate system registration, the point cloud data of the component and the proposed installation location are unified to the control network coordinate system, achieving feature dimension alignment. After outputting the unified coordinate system of the feature dimension coordinates of the component and the proposed installation location, initial deviation calculation is performed. For each feature dimension i to be checked, the initial deviation rate calculation formula is: δ i L represents the initial deviation rate of the i-th feature size; a L represents the dimension of the i-th component feature extracted from the component point cloud data, in mm. fδ i > 0, the actual size of the component is larger than the installation position size; if δ i < 0, the actual size of the component is smaller than the installation position size.

[0015] Judge the initial deviation rate eligibility: compare the calculated δ i with the deviation tolerance δ0 specified in the specification or design requirement, if δ i < δ0, the size is qualified, if δ i > δ0, the size is out of tolerance, which needs to be recorded; If all feature sizes have initial deviation rates within the deviation tolerance, the component installation proceeds normally, if some feature sizes have initial deviation rates exceeding the deviation tolerance and have on-site correction conditions, the feature sizes are corrected on site, and the component installation proceeds normally after marking; if there are feature sizes with initial deviation rates exceeding the deviation tolerance and without on-site correction conditions, the component is returned to the factory for repair.

[0016] The real-time spatial displacement data of the component is obtained through GNSS receiver, inclination sensor and force sensor, which monitors the spatial position, attitude and force state of the component in the installation process, and calculates the process deviation rate δ p .

[0017] The GNSS receiver calculates the component's x-axis, y-axis and z-axis component velocities and three-dimensional combined velocity in the field control network coordinate system in real time, and tracks the component's motion trajectory; the inclination sensor monitors the component's pitch angle in real time, which is used to judge the component's motion stability, identify shaking, and review whether the installation angle meets the design requirements after being in place; the force sensor collects load data of each lifting hook in real time, calculates the force difference percentage between lifting points, and ensures that the load deviation from the average value is not more than ±10%, so as to avoid single-point overload or underload.

[0018] The GNSS receiver provides the component installation three-dimensional motion velocity in the field control network coordinate system: including the component's component velocities V x (t), V y (t), V z (t) in X, Y, Z axis direction at time t, and the combined velocity V(t) representing the instantaneous motion rate, compare whether the velocities in each direction exceed the safety speed limit, V xa , V ya , V za and V a are the maximum allowable velocities of x direction, y direction, z direction and combined velocity respectively; calculate the velocity deviation rate, taking the x-axis direction as an example, the x-axis velocity deviation rate is The same method is used to calculate the speed deviation rates of the y-axis and z-axis and the total speed deviation rate respectively; if any speed deviation rate exceeds the maximum allowable speed in that direction, the installation is suspended, otherwise the component installation continues.

[0019] The tilt sensor is directly mounted on the component, and its measurements are based on the field control network coordinate system, reflecting the component's attitude relative to the ground horizontal reference, regardless of how the component moves within the coordinate system. The tilt sensor obtains the pitch angle θ, which is compared with the maximum allowable tilt angle θ0. If θ < θ0, the angular deviation rate δ θ =1, not included in the process deviation rate calculation; if θ>θ0, then .

[0020] Force sensor monitors F i (t), that is, the force on the i-th support point during transportation at time t during the component installation process, i=1,2,3,...,n, then the average load is: Force deviation at each support point Overall force deviation rate δF is required for all points i Both δF(t) and δF(t) are less than or equal to 10%.

[0021] Process deviation rate δ p The product of the resultant velocity deviation rate, angular deviation rate, and overall force deviation rate is worth noting. It is important to note that the deviation rate calculation requires the resultant velocity deviation rate, angular deviation rate, and overall force deviation rate to all meet their respective limits.

[0022] Environmental impact data is collected by vibration sensors, which monitor the vibration frequency of components in real time. The vibration of components is caused by the instantaneous kinetic energy generated by adjacent installed machines. The environmental deviation rate δ is calculated based on the frequency. e .

[0023] When f m >f a hour, When f m <f a At that time, δ e =0; where f m To measure the vibration frequency of the component, a high-sensitivity vibration sensor collects data in real time, transmitting instantaneous kinetic energy intensity during the operation of adjacent equipment, thus determining the vibration frequency generated by the component; f a The safety threshold is set according to the component. Exceeding this limit may cause component displacement, loose connection or installation deviation.

[0024] The intelligent prediction module: based on the initial deviation rate, process deviation rate and environmental deviation rate, the component one-time pose deviation and total deviation rate are predicted; the one-time pose deviation is expressed by a state vector, the state vector calculation adopts Kalman filtering algorithm, based on the initial deviation rate, process deviation rate and environmental deviation rate, the translation deviation of x-axis, y-axis and z-axis is obtained by analyzing, and θ x , θ y , θ z are respectively the rotation angle deviation of the component around x-axis, y-axis and z-axis, which are six degrees of freedom.

[0025] Based on Kalman filtering, a pose deviation prediction model is designed, and through the fusion of multi-source deviation rate data, the dynamic estimation of the comprehensive pose deviation of the component is realized.

[0026] The state vector is the real comprehensive pose deviation of the component to be estimated, which is composed of six degrees of freedom, wherein Δx, Δy and Δz are respectively the translation deviation of x-axis, y-axis and z-axis of the component based on the coordinates of the installation position in the control network coordinate system; θ x , θ y , θ z are respectively the rotation angle deviation of the component around x-axis, y-axis and z-axis.

[0027] The construction method of the pose deviation prediction model based on Kalman filtering is to calculate the initial deviation rate δ i by using the three-dimensional laser scanning result.

[0028] The guess value of X k is obtained based on the point cloud data in the full state perception module, and the initial state vector X0 is obtained by multiplying the guess value and the initial deviation rate δ i ; X0 and X k are state vectors of the same dimension; the state prediction equation is constructed as follows: F is the state transition matrix, for a simple model, F can be set as the unit matrix I; the deviation at the next time is the same as the current time, and the covariance prediction equation is constructed as follows: wherein F is the state transition matrix, P k-1 is the covariance matrix at the last time, F T is the transpose of the state transition matrix, and Q k is the process noise covariance matrix; the dynamic adjustment equation of the process noise covariance matrix is Q k =Q base +γ*|δ ek |*I; wherein Q base is the basic process noise; δ ekThe environmental deviation rate calculated at time k, γ is a scaling factor, maps the vibration intensity derived environmental deviation rate to the noise level, I is an identity matrix, Q base and γ are calibrated through offline simulation or historical data analysis to balance the response speed and stability of the system. The observation vector equation is constructed as follows: where H is an observation matrix, which is an identity matrix I in this example, v k is an observation noise, which is R k = | δ pk | * I in this example, where δ pk is the process deviation rate calculated at time k, I is an identity matrix, and R k represents the uncertainty of the measuring machine itself; the Kalman gain is calculated as follows: , Finally, the state prediction equation is as follows: .

[0029] Based on the results of the Kalman filter algorithm, a state vector ΔX k containing six degrees of freedom is obtained, which contains information about the first-time pose deviation. From the design specifications or process requirements, the maximum allowed deviation value corresponding to each degree of freedom of the state vector is read, and each degree of freedom of the state vector is divided by the corresponding allowed deviation value to obtain six dimensionless deviation ratios. The total deviation rate is the sum of the six deviation ratios.

[0030] Intelligent decision module: according to the predicted first-time pose deviation of the component, a deviation quantification model is constructed, and a pose correction decision is generated. The deviation quantification model is used to analyze and define the factor that has the greatest impact on the total deviation, and the total deviation rate is approximately expressed as a linear combination of δ i , δ p , and δ e : ΔX k ≈ β i * δ i + β p * δ p + β e * δ e + ε, where β i , β p , and β e are regression coefficients to be solved, δ i , δ p , and δ e are the initial deviation rate, the process deviation rate, and the environmental deviation rate, respectively, and ε is a residual term representing the part that the model cannot explain; using the collected data, the coefficients β i , β p , and β e are fitted by the least squares method.The size of these coefficients directly reflects the average influence of the corresponding deviation rate on the total deviation. The conditions for the factors with the greatest influence are adjusted first.

[0031] The deviation vector is deeply analyzed to convert abstract numerical values into explicit adjustment requirements. The system sensitivity model is called, which is obtained through pre-stored or external interface, and is formed through previous fine mechanics analysis or a large number of experimental data calibration. The model details the influence of each actuator on the pose of the component. A set of actuator action instructions is calculated, which makes the pose change of the component offset the six-degree-of-freedom deviation after the actuators act together, realizing accurate deviation correction. The action instruction is the pose correction decision this time.

[0032] It is worth noting that the action instruction setting must be combined with the actual scene to set strict constraint conditions, including stroke constraints, the operating range of each actuator cannot exceed the physical limit; action amplitude constraints, the single adjustment amplitude cannot be too large to avoid sudden stress increase or loss of control of the component; and coordination constraints, some actions need to be executed in a specific order to ensure that the optimization result is executable in reality.

[0033] Execution and control module: receives the pose correction decision, drives the actuators to complete the specific operation of the decision, and feeds back the adjustment operation completion through the built-in sensor.

[0034] The pose correction decision is received through Ethernet, and the received pose correction decision is format-verified to check whether the data field is complete, the data type is correct, and the numerical range is within a reasonable interval. If a format error is found, an error message is immediately fed back to request retransmission of the correct pose correction decision. If the verification is passed, the next step of the analysis process is entered. The physical limitations of the actuators, the requirements for motion smoothness, the requirements for task efficiency, and the environmental obstacle information are considered to determine the optimal motion path. The planned motion path is smoothed to eliminate the inflection points and mutations in the path, reduce the motion speed and direction changes of the actuators, reduce the impact on the mechanical structure, and improve the motion precision and stability. According to the length and complexity of the motion path, the control instructions are divided into batches and sent to the actuators. For short paths and simple movements, all instructions can be sent at once. For long paths and complex movements, the instructions are sent in batches. Real-time feedback information from the actuators is received to ensure that the next batch of instructions is sent after the previous batch of instructions is executed, avoiding instruction accumulation leading to execution confusion. During the operation of the execution device, the sensor installed on the execution device collects the motion parameters of the execution mechanism in real time, and feeds back the collected data in real time; the actual motion parameters collected in real time are compared with the target motion parameters in the instruction, the real-time motion error is calculated, the error is corrected according to the size and change trend of the error, and the error correction instruction is generated by using the PID control algorithm, which is sent to the execution device to adjust the motion state of the execution mechanism in real time, so that the actual motion parameters are constantly close to the target motion parameters, and the execution precision is ensured.

[0035] If the execution result meets the condition that all degrees of freedom offsets are less than the predicted degrees of freedom offsets, the execution success result is fed back to the decision system, and the next adjustment instruction is prepared to be received; if the execution result does not meet the preset index, all information including abnormal execution time, execution position, execution fault code and execution parameters is recorded in detail, and the execution failure result and information are fed back to the decision system; at the same time, according to the type of abnormal situation, the corresponding fault diagnosis process is started, the fault reason is analyzed, and the corresponding repair measures are taken, and after the fault is eliminated, the adjustment instruction is executed again.

[0036] The closed-loop verification and self-learning module: the installed component after executing the decision operation is collected again, the secondary pose deviation is identified, and the secondary pose deviation data is fed back for incremental learning to continuously optimize the pose deviation prediction algorithm and the pose deviation optimization decision algorithm.

[0037] After the adjustment operation of the execution and control module is completed, a three-dimensional laser scanner is used to scan the component and the installation position comprehensively, and the sampling rules consistent with the first collection are used, that is, the same number of sampling points, the same distribution position of sampling points and the same sampling frequency. During the collection process, the sensor maps the acquired pose data to the unified coordinate system to obtain the pose data of the target object in the unified coordinate system, checks whether there are missing sampling points or abnormal data values, and if there are, re-performs data collection of the corresponding feature size until complete and effective secondary pose data is acquired, the secondary pose deviation is calculated, and if the absolute values of the secondary pose deviations of all dimensions are less than the absolute values of the first pose deviations, it indicates that the adjustment decision is effective and the pose deviation is reduced; if the absolute values of the secondary pose deviations of some dimensions are greater than the absolute values of the first pose deviations, the abnormal situation needs to be marked to provide key analysis data for subsequent model learning. Validity determination of secondary pose deviation data: according to the task accuracy requirement, the allowed threshold of each dimension deviation in the unified coordinate system is set, if all the secondary pose deviations of all dimensions are within the allowed threshold range, it is determined that this adjustment meets the accuracy requirement; if any dimension deviation exceeds the allowed threshold, it is determined that the adjustment is not up to standard, and the secondary pose deviation and the pose correction decision need to be monitored as key samples. Each time the pose correction decision is converted into a learning opportunity, the pose deviation in operation is continuously reduced through incremental learning, and finally the full-automatic high-precision installation close to zero error is realized; the self-learning model is input with a pose deviation, a pose correction decision and a second pose deviation, the dimensional data of the second pose deviation greater than the first pose deviation is marked as abnormal data, and vice versa.

[0038] The self-learning model is trained with normal data and abnormal data, and the data chain composed of the current state, i.e. the second pose deviation, the decision instruction, i.e. the pose correction decision, and the predicted deviation, i.e. the first pose deviation, is used to train the model, so that the prediction is closer and closer to the physical response; the system gives feedback to the reduction of the deviation, the size of the feedback is positively correlated with the absolute value difference between the first pose deviation and the absolute value of the second pose deviation, the pose correction decision of abnormal data is given negative feedback, and the pose correction decision of normal data is given positive feedback; the model is iterated continuously, and for a given first pose deviation, the model selects the pose correction decision that can bring the maximum positive feedback; the existence of abnormal data helps the model quickly exclude invalid or harmful strategies.

[0039] Secondly, in the drawings of the disclosed embodiments, only the structures related to the disclosed embodiments are involved, other structures can refer to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the present application can be combined with each other; Finally, the above only describes the preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A steel structure construction accuracy control system based on multi-data fusion, characterized in that, include: Full-state perception module: Using an integrated sensor network, it comprehensively collects component construction accuracy data, including initial processing dimension data, real-time spatial displacement data, and environmental impact data of component installation. Deviation calculation module: Based on the initial processing dimension data of the component, the real-time spatial displacement data of the component, and the environmental impact data of the component installation process, the initial deviation rate, process deviation rate, and environmental deviation rate of the component installation process are calculated respectively. Intelligent prediction module: Based on the initial deviation rate, process deviation rate, and environmental deviation rate, predict the first pose deviation and total deviation rate of the component; Intelligent decision-making module: Based on the predicted pose deviation results of the components, constructs a deviation quantification model and generates pose correction decisions; Execution and Control Module: Receives pose correction decisions, drives the execution device to complete the specific operations of the decisions, and uses built-in sensors to provide feedback on the completion status of the decision operations; Closed-loop verification and self-learning module: For installed components that have completed the decision-making operation, pose information is collected again, secondary pose deviations are identified, and the secondary pose deviation data is fed back for incremental learning to continuously optimize the pose deviation prediction algorithm and the pose deviation optimization decision algorithm.

2. The steel structure construction accuracy control system based on multi-data fusion according to claim 1, characterized in that: The initial machining dimensions of the component are obtained by scanning with a 3D laser scanner. Before the component arrives on site or is hoisted, the dimensions of the component to be installed are acquired and compared with the dimensions of the intended installation location to calculate the initial deviation rate δ. i .

3. The steel structure construction accuracy control system based on multi-data fusion according to claim 2, characterized in that: The real-time spatial displacement data of the components is obtained through a GNSS receiver, tilt sensor, and force sensor to monitor the spatial position, attitude, and stress state of the components during installation in real time, and to calculate the process deviation rate δ. p .

4. The steel structure construction accuracy control system based on multi-data fusion according to claim 2, characterized in that: The environmental impact data is collected by vibration sensors, which monitor the vibration frequency of the components in real time. The vibration of the components is caused by the instantaneous kinetic energy generated by adjacent installed machines. The environmental deviation rate δ is calculated based on the frequency. e .

5. A steel structure construction accuracy control system based on multi-data fusion according to claim 1, characterized in that: The pose deviation is represented by a state vector, which is calculated using a Kalman filter algorithm. Based on the initial deviation rate, process deviation rate, and environmental deviation rate, the translational deviations along the x, y, and z axes, θ, are obtained. x θ y θ z It consists of six degrees of freedom: the rotational angle deviation of the component around the x-axis, y-axis, and z-axis.

6. The steel structure construction accuracy control system based on multi-data fusion according to claim 1, characterized in that: The total deviation rate is calculated based on the state vector. The maximum permissible deviation value corresponding to each degree of freedom of the state vector is divided by the corresponding permissible deviation value to obtain six dimensionless deviation ratios. The total deviation rate is the sum of the six deviation ratios.

7. A steel structure construction accuracy control system based on multi-data fusion according to claim 1, characterized in that: The aforementioned deviation quantification model is used to analyze and define the factors that have the greatest impact on the total deviation, and approximates the total deviation rate as δ. i δ p δ e Linear combination: ΔX k ≈β i *δ i +β p *δ p +β e *δ e +ε, where β i ,β p ,β e Let δ be the regression coefficient to be solved. i δ p δ e These represent the initial deviation rate, process deviation rate, and environmental deviation rate, respectively, with ε being the residual term, representing the portion not explained by the model. Using the collected data, the coefficients β are fitted using the least squares method. i ,β p ,β e .

8. The steel structure construction accuracy control system based on multi-data fusion according to claim 1, characterized in that: The adjustment decision involves issuing control commands to the actuator in batches based on the length and complexity of the motion path. For short paths and simple motions, all commands can be issued at once. For long paths and complex motions, a batch issuance method is adopted, with a certain number of commands issued in each batch. At the same time, feedback information from the actuator is received in real time to ensure that the next batch of commands is issued only after the previous batch of commands has been executed.

Images