Intelligent positioning system for precise butt joint of fabricated building components

Abstract

The invention provides an intelligent positioning system for precise butt joint of fabricated building components, and belongs to the technical field of fabricated building construction. The system comprises a multi-source heterogeneous sensing module, an intelligent positioning host, an execution adjustment module, a BIM cooperation module and an environment adaptive calibration module. The multi-source heterogeneous sensing module is integrated with a laser radar, a visual sense, a UWB and an attitude sensing unit, and data synchronous acquisition is realized through a self-created time synchronization mechanism; the intelligent positioning host adopts a three-layer architecture of preprocessing, feature fusion and decision output, and processes data in combination with improved Kalman filtering and a dynamic deviation correction model; the execution adjustment module realizes six-degree-of-freedom posture adjustment based on a cross type hydraulic driving structure, and is matched with force feedback compliant control; the BIM collaboration module constructs twin mapping and generates a deviation thermodynamic diagram; and the environment self-adaptive calibration module realizes quick calibration of environment abrupt change. The problems that traditional positioning is low in precision, poor in efficiency, weak in environment adaptability and the like are solved, and the method is suitable for precise butt joint of various assembly type components.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of prefabricated building construction, more specifically, relates to an intelligent positioning system for precise butt joint of prefabricated building components. BACKGROUND

[0002] The butt joint of prefabricated building components is a key link in construction, and its precision and efficiency directly determine the engineering quality and progress. The current industry mainstream adopts manual observation combined with simple instruments such as level and theodolite for positioning, which has many technical bottlenecks:

[0003] Firstly, the positioning accuracy is low, and manual judgment is easily affected by subjective factors. The actual measurement accuracy is usually ≥5mm, which is difficult to meet the high-precision requirements of prefabricated building joint tightness and bolt precise alignment;

[0004] Secondly, the environmental adaptability is poor. Changes in temperature and humidity (such as thermal expansion and contraction of components due to high temperature in summer), light interference (strong light or rainy weather), and hoisting shaking will all cause significant increase in positioning deviation, and the accuracy may even decrease to ≥8mm in extreme environments;

[0005] Thirdly, the butt joint efficiency is low. Manual repeated measurement and adjustment require multiple people to work together, and the time consumption for single butt joint is generally ≥30min, which restricts the construction progress;

[0006] Fourthly, the collaboration is seriously insufficient. The positioning process is disconnected from the BIM model, and digital tracing and multi-device linkage early warning cannot be achieved.

[0007] Although a few existing automatic positioning systems try to introduce a single sensor (such as only laser radar or UWB), there are still inherent defects:

[0008] Single laser radar is easily disturbed by shielding, and the positioning accuracy of UWB is only ±10cm, which cannot meet the requirements. Visual recognition is easily disabled under complex lighting;

[0009] The data fusion algorithm lacks a dynamic deviation compensation mechanism and cannot adapt to the changing environment of the construction site;

[0010] The actuator mostly adopts simple mechanical posture adjustment without soft control design, and the component collision damage rate is as high as 12%.

[0011] In summary, the existing technology cannot balance positioning accuracy, environmental adaptability, work efficiency, and construction safety, and an intelligent positioning technology that is multi-source fusion, dynamic correction, and collaborative closed loop is needed to break through the above bottlenecks. SUMMARY

[0012] In order to solve the above technical problems, the application provides an intelligent positioning system for precise butt joint of fabricated building components, which solves the technical problems of low positioning accuracy, poor environmental adaptability, low butt joint efficiency, insufficient collaboration and easy component collision damage caused by manual butt joint or single sensor of traditional fabricated building components.

[0013] The intelligent positioning system for precise butt joint of fabricated building components comprises a multi-source heterogeneous sensing module, an intelligent positioning host, an execution adjustment module and a BIM collaboration module.

[0014] The multi-source heterogeneous sensing module comprises a laser radar positioning unit, a visual feature recognition unit, an UWB (ultra-wideband) positioning unit and a component attitude sensing unit, each unit synchronously collecting component three-dimensional coordinate (x, y, z), surface feature point, spatial position and attitude angle (α, β, γ) data.

[0015] The intelligent positioning host is internally provided with a multi-source data fusion engine and a dynamic deviation correction model, the fusion engine adopts a hybrid strategy based on Kalman filtering and improved RANSAC algorithm to process sensing data, wherein the state update formula of Kalman filtering is as follows:

[0016] State equation: X k =A k X k-1 +B k U k +W k ;

[0017] Observation equation: Z k =H k X k +V t In the formula, X h is a system state vector (including three-dimensional coordinates and attitude angles) at time k, A ij is a state transition matrix, B max is a control input matrix, U ij is a posture adjustment control amount, W max is a process noise;

[0018] Z ij is an observation vector, H max is an observation matrix, and V ij is an observation noise;

[0019] The correction model dynamically outputs a positioning compensation amount in combination with environmental temperature and humidity and hoisting swing frequency, and the formula is as follows:

[0020] In the formula, ΔP is a positioning compensation amount vector, k max , k k , Temperature, humidity, frequency correction coefficient, T, RH, f respectively are real-time temperature and humidity, shaking frequency, T0, RH0, f0 are standard environmental parameters respectively;

[0021] The execution adjustment module receives host instructions to drive the hydraulic posture adjustment mechanism to adjust the component posture in real time;

[0022] The BIM coordination module compares the positioning data with the component BIM model in real time, generates a docking deviation thermal map, and feeds back to the host to form a closed-loop control.

[0023] Preferably, the visual feature recognition unit adopts a binocular industrial camera with a polarization filter lens, and the lens surface is integrated with a self-created multi-layer anti-dust coating; the unit is built-in with a component feature template library, the template library adopts a lightweight point cloud + texture feature composite storage structure, and the component surface is identified at a sub-pixel level through a self-created "feature point pyramid matching algorithm" to realize the matching similarity calculation of more than 32 key feature points.

[0024] S(i,j) = ω1 x [1- |d ij | / d max ]+ω2 x [1- |g ij | / g max ] formula, S(i,j) is the matching similarity of the i th template feature point and the j th real-time feature point, ω1, ω2 are the point cloud distance weight (0.6) and the texture gradient weight (0.4) respectively, d ij is the spatial distance of two point clouds, d max is the maximum effective distance threshold, g ij is the texture gradient difference value, g max is the maximum gradient difference threshold; when S(i,j) ≥ 0.85, it is determined that the matching is successful.

[0025] Preferably, the multi-source data fusion engine of the intelligent positioning host includes a data preprocessing layer, a feature fusion layer, and a decision output layer; the preprocessing layer adopts a self-created "abnormal data adaptive elimination algorithm" to filter sensor noise, and the abnormality calculation formula is as follows:

[0026] In the formula, D k is the abnormality of the k th observation data, Z k is the real-time observation value, is the predicted value based on the previous 5 frames of data, σ k is the observation standard deviation;

[0027] When D kData with a value ≥3 is automatically removed; the feature fusion layer constructs a correlation matrix of 3D coordinates, attitude angles, and feature points, and achieves data fusion through matrix weighted iteration; the decision output layer embeds an edge computing unit, supporting the output of positioning results within 100ms.

[0028] Preferably, the execution adjustment module includes a six-degree-of-freedom hydraulic attitude adjustment platform and a force feedback sensing unit; the attitude adjustment platform adopts a self-developed "cross-shaped hydraulic drive structure" to achieve independent control of the component's X / Y / Z axis translation and rotation around the three axes, and the attitude adjustment amount is calculated using the following formula:

[0029] Δθ=K p ×e k +K i ×∫e k dt+K_d×(e k -e k-1 In the formula ) / Δt, Δθ is the attitude angle adjustment amount, K p K i K and K_d are the proportional, integral, and differential coefficients, respectively, and e k Δt is the attitude deviation value at time k, and Δt is the control period; the force feedback unit collects docking contact force data in real time, and triggers the host to start the compliant adjustment mode when the force value exceeds the preset threshold (5kN).

[0030] Preferably, the BIM collaboration module has a built-in self-developed "component twin mapping engine". Through this engine, the real-time data collected by the positioning system is dynamically associated with the component parameters in the BIM model to generate a twin dataset containing positioning deviation, attitude error and environmental interference.

[0031] The formula for the deviation evaluation index is:

[0032] E = √[(Δx)] 2 +(Δy) 2 +(Δz) 2 +(Δα) 2 +(Δβ) 2 +(Δγ) 2 In the formula, E is the comprehensive deviation value, Δx, Δy, and Δz are the three-dimensional coordinate deviations, and Δα, Δβ, and Δγ are the attitude angle deviations. When E ≤ 0.5 mm, it is determined that the docking accuracy requirements are met. The module supports data interaction with the tower crane monitoring system and construction elevator system at the construction site to achieve multi-device collaborative positioning.

[0033] Preferably, it also includes a self-developed environment adaptive calibration module, which includes a temperature and humidity sensor, a wind speed sensor, and a light sensor. By constructing an "environment-positioning error" mapping database, a BP neural network algorithm is used to compensate for environmental errors in the positioning results. The formula for calculating the compensated positioning value is as follows:

[0034] P' = P + W n ×σ n (...)+...+W1×σ1(W0×X+b0)+b n In the formula, P' is the compensated positioning value, P is the original positioning value, and W0~W n Here are the weight matrices for each layer, b0~b n Let σ1 be the bias vector, and σ2 be the bias vector. n X is the activation function, and X is the environmental parameter vector (temperature, humidity, wind speed, light intensity). When the environmental parameters change abruptly (change rate ≥ 10% / s), the fast calibration process is automatically started, and the calibration time is ≤ 3s.

[0035] Preferably, the intelligent positioning host also has a built-in data security encryption unit and an edge storage module; the encryption unit adopts a self-developed "dynamic key + data fragmentation" encryption strategy to perform end-to-end encryption of positioning data and control commands, and the dynamic key generation formula is: K t =Hash(DevID+T) t +Rand k In the formula, K t The key at time t is the dynamic key, Hash is the SHA-256 hash function, DevID is the unique identifier of the device, and T is the dynamic key at time t. t For timestamps, Rand k The value is a random number; the edge storage module adopts a dual storage mode of SSD + cloud backup, which supports local real-time storage and cloud traceability of location data.

[0036] Preferably, the visual feature recognition unit also includes a self-developed "feature point dynamic update mechanism." When stains or damage on the component surface cause some feature points to fail, it automatically calls a backup feature point set from the template library. Recognition accuracy is ensured through redundant feature point matching. The formula for calculating the redundancy matching accuracy is:

[0037] Acc = (N_valid / N_total) × [1 - (N_fail / N_max)] where Acc is the recognition accuracy, N_valid is the number of valid matching feature points, N_total is the total number of feature points, N_fail is the number of failed feature points, and N_max is the maximum allowed number of failed features. When the proportion of failed feature points is ≤40%, Acc ≥99.5%, and sub-pixel-level positioning can still be achieved.

[0038] Preferably, the multi-source heterogeneous sensing module adopts a self-developed "time synchronization triggering mechanism," which combines GPS timing with local crystal oscillator calibration to achieve a time synchronization accuracy of ≤50μs for data acquisition from each sensor. The time synchronization error is calculated using the formula: ΔT=|T_GPS-T_osc| where ΔT is the time synchronization error, T_GPS is the GPS timing, and T_osc is the local crystal oscillator time. Calibration is performed every 10s to ensure that ΔT is always ≤50μs. The module shell adopts a one-piece magnesium-aluminum alloy structure and integrates an IP67 protection design that is waterproof, dustproof, and impact-resistant.

[0039] Preferably, the execution adjustment module further includes a self-developed "pre-docking posture planning unit." This unit, based on the component docking node parameters in the BIM model and combined with the hoisting path data, uses the A* algorithm to plan the optimal pre-docking posture, reducing the posture adjustment amount during the docking process by ≥30%. The formula for calculating the adjustment amount optimization rate is as follows:

[0040] η=(Δθ orig9 -Δθ Opt ) / Δθ Ori9 In the formula ×100%, η is the optimization rate of the adjustment amount, and Δθ Ori9 The total attitude adjustment, Δθ, is the value before planning. Opt This represents the total attitude adjustment amount after planning; η≥30%.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] It pioneered a multi-source heterogeneous fusion strategy combining LiDAR, vision, UWB, and attitude sensing, coupled with a self-developed time synchronization triggering mechanism and an adaptive algorithm for eliminating abnormal data. After improved Kalman filtering, the positioning accuracy is stable at ≤0.5mm, which is an improvement over traditional manual positioning and completely solves high-precision construction problems such as bolt alignment and joint sealing.

[0043] An environmental adaptive calibration module is added, which constructs an environment-positioning error mapping model based on a BP neural network. When environmental parameters change abruptly, rapid calibration can be completed within ≤2.5s. It still maintains an accuracy of ≤0.5mm in harsh environments such as -10℃ / 85%RH, while the accuracy of traditional methods drops to ≥8mm in such environments, thus achieving stable operation in complex construction environments.

[0044] The system adopts a cross-shaped hydraulic attitude adjustment structure, combined with a pre-dock attitude planning unit to reduce the adjustment amount by ≥35%. With force feedback smooth control, the time for a single docking is ≤10 minutes, which improves efficiency compared to traditional methods. No component collision damage was found in 50 actual tests, and the damage rate was reduced from 12% in the traditional method to 0%, reducing construction costs and rework risks.

[0045] By constructing digital twins of components through the BIM collaboration module, deviation heat maps are generated in real time and interconnected with systems such as tower cranes and construction elevators to achieve multi-device collaborative early warning. When the tower crane is overloaded or personnel approach, it automatically reduces its speed. At the same time, the data is encrypted and stored in dual storage (local + cloud), providing digital support for construction quality traceability and promoting the transformation of prefabricated building construction towards intelligence and refinement. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the workflow of the present invention. Detailed Implementation

[0047] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0048] Please see Figure 1 This invention provides an intelligent positioning system for precise docking of prefabricated building components. The core architecture includes a multi-source heterogeneous sensing module, an intelligent positioning host, an execution adjustment module, a BIM collaboration module, and an environmental adaptive calibration module. Each module interacts with data via industrial Ethernet (Profinet protocol) and a 5G private network, with an overall response latency of ≤150ms. The system's physical deployment consists of three parts:

[0049] Component end (installation of multi-source heterogeneous sensing modules and force feedback units);

[0050] Construction machinery end (equipped with intelligent positioning host and execution adjustment module, integrated on the lifting device of tower crane or truck crane);

[0051] The monitoring center (deploys a BIM collaboration module, which is interconnected with the tower crane monitoring system and construction elevator system at the construction site via the OPCUA interface).

[0052] Multi-source heterogeneous sensing module:

[0053] This module integrates a lidar positioning unit, a visual feature recognition unit, a UWB positioning unit, and a component attitude sensing unit. It features a unibody magnesium-aluminum alloy casing with an IP67 protection rating, suitable for construction environments ranging from -20℃ to 60℃. The specific design of each unit is as follows:

[0054] The lidar positioning unit uses a Velodyne 16-line lidar with a ranging range of 0.5m to 100m, a ranging accuracy of ±2cm, and a sampling frequency of 10Hz. Driven by ROS (Robot Operating System), it outputs real-time point cloud data of the component surface. After removing ground interference through pass-through filtering (Z-axis range 0.1m to 5m), the three-dimensional coordinates (x1, y1, z1), (x2, y2, z2), and (x3, y3, z3) of the three positioning reference points on the top of the component are extracted.

[0055] Visual Feature Recognition Unit: Employs a Baslerac A2040-180uc binocular industrial camera (2048×1536 resolution, 15fps), equipped with a 25mm polarization filter lens (polarization direction can be remotely adjusted via the host). The lens surface is coated with three layers of a self-developed anti-reflective and dustproof coating (first layer: SiO2, 100nm thickness; second layer: Al2O3, 80nm thickness; third layer: fluoride, 50nm thickness), achieving a light transmittance ≥95% and an IP65 dustproof rating. The unit has a built-in feature template library (stored in 8GB eMMC flash memory), employing lightweight point cloud + texture feature composite storage: point cloud data is compressed using voxel downsampling (5mm voxel size), and texture features are extracted using the SIFT algorithm and reduced to 64 dimensions. The self-developed feature point pyramid matching algorithm is implemented as a 4-layer pyramid structure (layer 1: original size, layer 2: 1 / 2 size, layer 3: 1 / 4 size, layer 4: 1 / 8 size). Matching is performed layer by layer from the top to the bottom, with a matching threshold S(i,j) ≥ 0.85 for each layer. The final output is the sub-pixel coordinates of 32 key feature points (precision 0.1 pixels), corresponding to an actual physical precision of ±0.1mm. A dynamic feature point update mechanism is also integrated.

[0056] The system presets 48 feature points (32 primary and 16 backup). It detects the confidence level of feature point matching in real time (a score below 0.7 is considered invalid). When the proportion of invalid points is ≤40%, it automatically calls up 32 valid points from the backup set to ensure recognition accuracy Acc≥99.5%.

[0057] UWB positioning unit: Employs Decawave DW1000 chip, operating in the 3.1GHz~10.6GHz frequency band, with a positioning range of 0m~50m, positioning accuracy of ±10cm, and a communication rate of 6.8Mbps. The system deploys 4 UWB anchor points (installed at the four corners of the construction site, 10m high) and 1 UWB tag (integrated within the module). The tag's 3D coordinates (x_U, y_U, z_U) are calculated using the TDOA (Time Difference of Arrival) algorithm, with a sampling frequency of 20Hz.

[0058] Component attitude sensing unit: The MPU9250 nine-axis sensor (gyroscope range ±2000° / s, accelerometer range ±8g, magnetometer range ±4800μT) is selected with a sampling frequency of 100Hz. The component attitude angles (roll angle α, pitch angle β, yaw angle γ) are calculated by Kalman filtering with a calculation accuracy of ±0.1°.

[0059] Time synchronization triggering mechanism: A combination of GPS time synchronization (UTC time synchronization accuracy ±1μs) and a local crystal oscillator (temperature compensated crystal oscillator, frequency deviation ±5ppm) is used. Calibration is performed every 10s: the standard time T_GPS is obtained via GPS, and the error ΔT = |T_GPS - T_osc| between the local crystal oscillator time T_osc and T_GPS is calculated. If ΔT > 30μs, the crystal oscillator frequency is adjusted to ensure that the time synchronization accuracy of data acquisition for each unit is ≤50μs. The synchronization trigger signal is generated by the FPGA (Xilinx Artix-7) with a trigger interval of 100ms.

[0060] Intelligent positioning host:

[0061] As the system core, the host uses an Intel Core i7-12700H processor (14 cores, 20 threads, 2.7GHz), equipped with 16GB DDR5 memory and a 512GB NVMe SSD, running the Ubuntu 20.04 operating system, and supporting multiple network interfaces including industrial Ethernet, 5G, and WiFi 6. It incorporates a multi-source data fusion engine, a dynamic deviation correction model, a data security encryption unit, and an edge storage module, with the specific design as follows:

[0062] Multi-source data fusion engine: Adopts a three-layer architecture of preprocessing, feature fusion, and decision output. Specific process:

[0063] Preprocessing layer: Timestamps are aligned for data from each sensor (based on synchronization trigger signals), and an adaptive outlier removal algorithm is used: the mean μ and standard deviation σ are calculated for 5 consecutive frames of data from each sensor. k When the current frame data Z k Satisfy D k =|Z k -μ| / σ k When the value is ≥3, it is judged as abnormal and removed, and at the same time, linear interpolation of the first two frames of data is used to complete it.

[0064] Feature fusion layer: A 3×4 correlation matrix is ​​constructed (rows: LiDAR, vision, UWB; columns: x-coordinate, y-coordinate, z-coordinate, attitude angle). The coordinate data from LiDAR and vision are assigned weights of 0.4 (LiDAR) and 0.4 (vision), respectively. UWB data is assigned a weight of 0.2 (due to lower accuracy). Attitude angle data is directly taken from the sensor unit output (weight 1.0). The fused value is calculated through matrix weighted iteration: X_fuse=Σ(W_i×X_i), where W_i is the weight and X_i is the data from each unit. The iteration count is 3, with a convergence accuracy of ±0.05mm.

[0065] Decision output layer: Embedded with NVIDIA Jetson Xavier NX edge computing units, running an improved Kalman filter algorithm.

[0066] State vector (Including position, attitude, and velocity), state transition matrix A k The observation matrix H is a 12×12 identity matrix (ignoring short-term velocity changes). k For a 6×12 matrix (observation position and attitude only), the process noise covariance is:

[0067] Q = diag([1e-4,1e-4,1e-4,1e-6,1e-6,1e-6,1e-3,1e-3,1e-3,1e-5,1e-5,1e-5]), and the observation noise covariance R = diag([1e-3,1e-3,1e-3,1e-5,1e-5,1e-5]). The location result is output within 100ms after filtering.

[0068] Dynamic deviation correction model: Based on the mapping relationship between training environment and positioning error of BP neural network, the network input is temperature and humidity (T,RH) and hoisting sway frequency (f), and the output is the positioning compensation amount ΔP=[Δx,Δy,Δz,Δα,Δβ,Δγ].

[0069] The network structure consists of 3 input layers → 16 hidden layers → 6 output layers. ReLU (for hidden layers) and Linear (for output layers) activation functions are used. The network was trained using 1000 sets of construction site data (-20℃~60℃, 30%RH~90%RH, 0.5Hz~5Hz oscillation frequency). The fitting accuracy R0 was [value missing]. 2 ≥0.92.

[0070] Real-time correction formula In the middle, the coefficient k t =0.005mm / ℃, k h =0.002mm / %RH Standard environmental parameters: T0 = 25℃, RH0 = 60%RH, f0 = 1Hz.

[0071] Data security and storage module: The encryption unit adopts a dynamic key + data fragmentation strategy.

[0072] Every 500ms, via formula K t =Hash(DevID+T) t +Rand k Generate a key (DevID is the module's unique MAC address, T) t Rand provides timestamps in milliseconds. k The location data and control commands are divided into 8 pieces, each using a K-series random number (0-1e9, hash SHA-256). tData is transmitted after encryption. Edge storage uses a 512GB SSD + Alibaba Cloud OSS cloud backup: local real-time storage of nearly 7 days of data (sampling interval 100ms), and automatic uploading of the previous day's data (ZIP format) to the cloud at 3 am every day, with a cloud storage period of 1 year.

[0073] Execute the adjustment module:

[0074] This module integrates a six-degree-of-freedom hydraulic attitude adjustment platform, a force feedback sensing unit, and a pre-docking attitude planning unit. It is installed below the lifting device (load capacity ≤ 10t). The attitude adjustment range is: X / Y axis ±200mm (translation), Z axis ±100mm (lifting), and rotation around the X / Y / Z axes ±10° (attitude adjustment). The adjustment accuracy is ±0.1mm (translation) and ±0.05° (rotation). The specific design is as follows:

[0075] Six-DOF hydraulic attitude adjustment platform: Utilizing a self-developed cross-shaped hydraulic drive structure, it consists of four horizontal hydraulic cylinders (model: HOB40×100, working pressure 16MPa, thrust 50kN) and two vertical hydraulic cylinders (model: HOB50×50, working pressure 16MPa, thrust 80kN). The horizontal cylinders are arranged in a cross shape (two on the X-axis and two on the Y-axis) to control translation and rotation within the control plane; the vertical cylinders are symmetrically arranged to control Z-axis lifting and pitch. The hydraulic system is equipped with electro-hydraulic proportional valves (response time ≤20ms), driven by a DSP controller (TITMS320F28335). After receiving the adjustment amount Δθ output from the host, it controls the cylinder extension and retraction through a PID algorithm. PID parameter: K... p =0.8, K i =0.1, K_d=0.05, control period 10ms.

[0076] Force feedback sensing unit: Four spoke-type tension sensors (model: H3-C3, range 0~50kN, accuracy ±0.1%FS) are installed at the flange connecting the attitude adjustment platform and the component. These sensors collect the contact forces F1, F2, F3, and F4 in real time. When any force value ≥5kN (preset threshold), a signal is immediately sent to the host to trigger the compliant adjustment mode: reducing the hydraulic cylinder thrust to 10kN and increasing the integral time of the PID control (K). i Adjust to 0.3) to avoid component collisions.

[0077] Pre-docking posture planning unit: Based on Qt-developed planning software, after importing the BIM model (format: IFC4.0), the unit automatically extracts component docking node parameters (such as reserved hole diameter, bolt spacing, etc.), combines them with hoisting path data (obtained from the tower crane monitoring system, including coordinates of the lifting point, via points, and target point), and uses the A* algorithm to plan the optimal pre-docking posture. The algorithm's heuristic function h(n) = √[(x n -x_goal)2 +(y n - y_goal) 2 +(z n - z_goal) 2 (Euclidean distance). The cost function g(n) is the energy consumption for attitude adjustment (proportional to the square of the adjustment angle). The search step is 0.5 mm / 0.1°, and the planning time is ≤ 2 s, reducing the total attitude adjustment amount during the docking process by ≥ 35% (the measured optimization rate η = 35% - 40%).

[0078] BIM Collaboration Module:

[0079] This module is deployed on the industrial computer in the monitoring center (Intel Core i9-13900K, 32 GB of memory, 2 TB SSD), uses Autodesk Revit 2024 as the BIM basic platform, and realizes the "Component Twin Mapping Engine" through C# secondary development. The specific functions are as follows:

[0080] Twin Mapping and Deviation Calculation: The engine receives the fused positioning data (x, y, z, α, β, γ) from the intelligent positioning host in real time, compares it with the theoretical coordinates (x0, y0, z0, α0, β0, γ0) of the components in the BIM model, and through the formula:

[0081] E = √[(Δx) 2 +(Δy) 2 +(Δz) 2 +(Δα) 2 +(Δβ) 2 +(Δγ) 2 to calculate the comprehensive deviation E, where the unit of coordinate deviation is mm, and the unit of attitude angle deviation is ° (conversion coefficient: 1° = 0.017 mm / m).

[0082] Deviation Heat Map Generation: Use VTK (Visualization Toolkit) to render the component model, map the deviation value E to a color gradient (E ≤ 0.3 mm: green; 0.3 mm < E ≤ 0.5 mm: yellow; E > 0.5 mm: red), generate the docking deviation heat map (update frequency 1 Hz), and display it in real time on the monitor screen in the monitoring center (4K resolution), while marking the position and value of the maximum deviation point.

[0083] Multi-device Collaboration: Interconnect with the tower crane monitoring system (obtain the lifting capacity and boom angle) and the construction elevator system (obtain the personnel position) through the OPC UA interface. When the lifting capacity of the tower crane ≥ 80% of the rated value or the construction elevator approaches the docking area (distance ≤ 5 m), automatically mark the warning information on the heat map and send a "deceleration adjustment" instruction to the intelligent positioning host, reducing the attitude adjustment speed to 50% of the original speed.

[0084] Environmental Adaptive Calibration Module:

[0085] This module integrates an SHT30 temperature and humidity sensor (accuracy ±0.3℃, ±2%RH), an FS4000 wind speed sensor (range 0~30m / s, accuracy ±0.1m / s), and a BH1750 light sensor (range 0~65535lux, accuracy ±1lux), and is integrated with a multi-source heterogeneous sensor module. When environmental parameters change abruptly (rate of change ≥10% / s, such as a sudden increase in humidity due to summer rainstorms), a rapid calibration process is automatically initiated.

[0086] 1) Collect three sets of environmental parameters and location data after the mutation;

[0087] 2) Call the pre-trained BP neural network to output the compensation amount ΔP;

[0088] 3) Inject ΔP into the fusion engine of the smart positioning host, with a calibration time of ≤2.5s (average measured time of 2.3s), to ensure that the positioning accuracy remains ≤0.5mm after sudden environmental changes.

[0089] System workflow:

[0090] The application process of this system in the docking construction of prefabricated concrete shear wall components is as follows, with a total docking time of ≤10 minutes (including hoisting and positioning adjustment):

[0091] Step 1: System Initialization (30s) The monitoring center starts the BIM collaboration module, imports the BIM model of the component to be docked (e.g., a 3m×2.4m×0.2m shear wall, weighing 5t), and sets the docking target coordinates (x0, y0, z0) and attitude angles (α0=0°, β0=0°, γ0=0°); the intelligent positioning host starts the multi-source heterogeneous sensing module and performs time synchronization calibration (ΔT≤50μs); the execution adjustment module initializes the hydraulic system, and the pre-docking attitude planning unit outputs the optimal pre-docking attitude (e.g., X-axis translation +50mm, Y-axis rotation -2°).

[0092] Step 2: Component Lifting and Data Acquisition (5 min) The tower crane lifts the component, and the multi-source heterogeneous sensing modules work synchronously: the lidar collects point cloud data, the vision unit identifies feature points, the UWB outputs tag coordinates, and the attitude sensing unit calculates the attitude angle. All data are transmitted to the intelligent positioning host after time synchronization, with a transmission delay ≤50ms. The environmental adaptive calibration module collects temperature and humidity (e.g., T=32℃, RH=75%), wind speed (e.g., 1.2m / s), and illumination (e.g., 8000 lux) data in real time. Calibration is not initiated if there are no sudden environmental changes.

[0093] Step 3: Data Fusion and Deviation Correction (Time taken 100ms / time) The intelligent positioning host preprocessing layer removes one set of abnormal point cloud data from the LiDAR (D k=3.2>3), the feature fusion layer calculates the weighted fusion coordinates and attitude angles, and outputs the preliminary positioning results after the improved Kalman filter (e.g., x=1000.2mm, y=500.3mm, z=300.1mm, α=0.05°, β=-0.03°, γ=0.02°);

[0094] The dynamic deviation correction model outputs compensation based on environmental parameters:

[0095] ΔP=[-0.02mm,-0.01mm,0mm,0°,0°,0.01°];

[0096] The corrected positioning results are (1000.18mm, 500.29mm, 300.1mm, 0.05°, 0.03°, 0.03°).

[0097] Step 4: Perform adjustment and force feedback (3 min). The adjustment module receives instructions from the host and drives the hydraulic cylinders to adjust according to the pre-dating posture planning path: the X-axis cylinder extends and retracts -50 mm (from +50 mm back to 0 mm), the Y-axis cylinder fine-tunes -0.01 mm, and the rotation cylinder around the Y-axis extends and retracts to make β return from -2° to 0.03°; when the component approaches the docking surface (distance ≤10 mm), the force feedback unit detects the contact force F1 = 2.3 kN (<5 kN) and maintains the normal attitude adjustment speed; when the distance is ≤1 mm, F1 increases to 4.8 kN, triggering the compliant adjustment mode, and slowly advances to the docking surface.

[0098] Step 5: BIM Collaborative Feedback and Closed-Loop Optimization (Time taken: 1.5 min) The BIM collaboration module receives the corrected positioning data in real time, calculates the comprehensive deviation E = 0.42 mm (yellow heat map), and marks the maximum deviation point as the Y-axis direction (0.29 mm). After the monitoring center personnel confirm the heat map, the host issues a "fine adjustment" command, and the execution module fine-tunes the Y-axis cylinder by +0.01 mm, reducing the deviation to E = 0.38 mm (green heat map). At this time, the tower crane monitoring system reports a lifting capacity of 4.5 t (<5 t, 80% of the rated value), with no warning information, and the docking is complete.

[0099] Step 6: Data storage and traceability. The docking process data (location data, environmental parameters, adjustment instructions) is encrypted with a dynamic key and stored in real time on the local SSD. It is automatically uploaded to the cloud at 3:00 AM on the same day to generate a docking report (including deviation curves and adjustment logs) for later quality traceability.

[0100] Key performance testing and verification:

[0101] At the construction site of a prefabricated residential project (10 18-story residential buildings), 50 shear wall components were selected for docking tests. The performance indicators of this system and the traditional manual positioning method were compared. The results are shown in the table below:

[0102]

[0103] Test results show that this system is significantly superior to traditional methods in terms of positioning accuracy, efficiency, environmental adaptability, and safety, and fully meets the high-precision docking construction requirements of prefabricated buildings.

[0104] This invention achieves three key innovations through core designs such as multi-source heterogeneous sensor fusion, dynamic deviation correction, and BIM collaborative closed-loop:

[0105] 1) A self-developed multi-source fusion strategy combining LiDAR, vision, and UWB, combined with a time synchronization mechanism, achieves sub-millimeter positioning accuracy, solving the problem of weak anti-interference of single sensors;

[0106] 2) An environmental adaptive calibration and dynamic deviation correction model is proposed to adapt to complex construction environments and maintain high accuracy even after sudden environmental changes;

[0107] 3) The design incorporates a cross-shaped hydraulic attitude adjustment and pre-docking plan, combined with force feedback compliant control, significantly reducing collision risks and adjustment time. This system can improve component docking efficiency by over 75% and reduce the damage rate to zero, providing core technical support for intelligent construction of prefabricated buildings.

[0108] The embodiments of the present invention are given for illustrative and descriptive purposes only, and are not intended to be exhaustive or to limit the invention to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of the invention, and to enable those skilled in the art to understand the invention and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. An intelligent positioning system for precise docking of prefabricated building components, characterized by: It includes a multi-source heterogeneous sensing module, an intelligent positioning host, an execution adjustment module, and a BIM collaboration module; The multi-source heterogeneous sensing module includes a lidar positioning unit, a visual feature recognition unit, a UWB positioning unit, and a component attitude sensing unit. Each unit synchronously collects the component's three-dimensional coordinates x, y, z, surface feature points, spatial position, and attitude angles α, β, γ. The intelligent positioning host has a built-in multi-source data fusion engine and a dynamic deviation correction model. The fusion engine uses a hybrid strategy based on Kalman filtering and an improved RANSAC algorithm to process the sensor data. The Kalman filtering state update formula is as follows: State equation: X k =A k X k-1 +B k U k +W k ; Observation equation: Z k =H k X k +V k In the formula, X k Let A be the system state vector at time k. k Let B be the state transition matrix. k To control the input matrix, U k For attitude control, W k This is process noise; Z k H is the observation vector. k V is the observation matrix. k To observe noise; The correction model dynamically outputs the positioning compensation amount by combining ambient temperature and humidity with the hoisting sway frequency. The formula is as follows: In the formula, ΔP is the positioning compensation vector, and k t k h , These are the correction coefficients for temperature, humidity, and frequency, respectively; T, RH, and f are the real-time temperature, humidity, and swaying frequency, respectively; and T0, RH0, and f0 are the standard environmental parameters, respectively. The execution adjustment module receives instructions from the host to drive the hydraulic attitude adjustment mechanism to adjust the posture of the component in real time. The BIM collaboration module compares the positioning data with the component BIM model in real time, generates a docking deviation heat map, and feeds it back to the host to form a closed-loop control.

2. The system according to claim 1, characterized in that, The visual feature recognition unit uses a binocular industrial camera equipped with a polarizing filter lens, and the lens surface is integrated with a multi-layer anti-reflective and dustproof coating. The unit has a built-in component feature template library. The template library adopts a lightweight point cloud + texture feature composite storage structure. It achieves sub-pixel level recognition of more than 32 key feature points on the component surface through a feature point pyramid matching algorithm. The formula for calculating the matching similarity is as follows: S(i,j)=ω1×[1-|d ij | / d max ]+ω2×[1-|g ij | / g max ]; In the formula, S(i,j) represents the matching similarity between the i-th template feature point and the j-th real-time feature point, ω1 and ω2 are the point cloud distance weight and texture gradient weight, respectively, and d \ Let d be the spatial distance between two points in the cloud. max g is the maximum effective distance threshold. ij g represents the texture gradient difference value. max The maximum gradient difference threshold; A successful match is determined when S(i,j)≥0.

85.

3. The system according to claim 1, characterized in that, The multi-source data fusion engine of the intelligent positioning host includes a data preprocessing layer, a feature fusion layer, and a decision output layer; The preprocessing layer uses an adaptive outlier removal algorithm to filter sensor noise. The outlier calculation formula is as follows: In the formula, D k Z represents the outlier of the k-th observation. k For real-time observations, σ is the predicted value based on the first 5 frames of data. k The standard deviation of observations; When D k The data will be automatically removed if the value is ≥3. feature The fusion layer constructs a correlation matrix of 3D coordinates, attitude angles, and feature points, and achieves data fusion through matrix weighted iteration. The decision output layer embeds an edge computing unit, supporting the output of positioning results within 100ms.

4. The system according to claim 1, characterized in that, The execution adjustment module includes a six-degree-of-freedom hydraulic attitude adjustment platform and a force feedback sensing unit; The attitude adjustment platform adopts a cross-shaped hydraulic drive structure to achieve independent control of the component's X / Y / Z axis translation and rotation around the three axes. The formula for calculating the attitude adjustment amount is: Δθ=K p ×e k +K i ×∫e k dt+K_d×(e k -e k-1 ) / Δt; In the formula, Δθ is the attitude angle adjustment amount, and K p K i K and Kd are the proportional, integral, and differential coefficients, respectively, and e k Let Δt be the attitude deviation value at time k, and Δt be the control period. The force feedback unit collects docking contact force data in real time, and triggers the host to start the compliance adjustment mode when the force value exceeds the preset threshold.

5. The system according to claim 1, characterized in that, The BIM collaboration module has a built-in component twin mapping engine. Through this engine, the real-time data collected by the positioning system is dynamically associated with the component parameters in the BIM model to generate a twin dataset containing positioning deviation, attitude error and environmental interference. The formula for the deviation evaluation index is: E=√[(Δx) 2 +(Δy) 2 +(Δz) 2 +(Da) 2 +(Δβ) 2 +(Δγ) 2 ]; In the formula, E is the comprehensive deviation value, Δx, Δy, and Δz are the three-dimensional coordinate deviations, and Δα, Δβ, and Δγ are the attitude angle deviations; When E≤0.5mm, it is determined that the docking accuracy requirement is met; The module supports data interaction with the tower crane monitoring system and construction elevator system at the construction site, enabling collaborative positioning of multiple devices.

6. The system according to claim 1, characterized in that, It also includes an environment adaptive calibration module, which contains temperature and humidity sensors, wind speed sensors, and light sensors. By constructing an environment-positioning error mapping database, a BP neural network algorithm is used to compensate for environmental errors in the positioning results. The formula for calculating the compensated positioning value is as follows: P'=P+W n ×σ n (...)+...+W1×σ1(W0×X+b0)+b n ; In the formula, P' is the compensated positioning value, P is the original positioning value, and W0~W n Here are the weight matrices for each layer, b0~b n Let σ1 be the bias vector, and σ2 be the bias vector. n Let X be the activation function, and let X be the environment parameter vector. When environmental parameters change abruptly, a rapid calibration process is automatically initiated, with a calibration time of ≤3 seconds.

7. The system according to claim 1, characterized in that, The intelligent positioning host also has a built-in data security encryption unit and an edge storage module; The encryption unit employs a dynamic key + data fragmentation encryption strategy to perform end-to-end encryption of positioning data and control commands. The dynamic key generation formula is as follows: K t =Hash(DevID+T t +Rand k ); In the formula, K t The key at time t is the dynamic key, Hash is the SHA-256 hash function, DevID is the unique identifier of the device, and T is the dynamic key at time t. t For timestamps, Rand k It is a random number; The edge storage module adopts a dual storage mode of SSD + cloud backup, which supports local real-time storage and cloud-based tracking of location data.

8. The system according to claim 2, characterized in that, The visual feature recognition unit also includes a dynamic feature point update mechanism. When stains or damage on the component surface cause some feature points to fail, it automatically calls a backup feature point set from the template library. The recognition accuracy is ensured through feature point redundancy matching. The formula for calculating the redundancy matching accuracy is as follows: Acc=(N_valid / N_total)×[1-(N_fai l / N_max)]; In the formula, Acc is the recognition accuracy, N_valid is the number of valid matching feature points, N_total is the total number of feature points, N_fail is the number of failed feature points, and N_max is the maximum number of allowed failures. When the proportion of failed feature points is ≤40%, Acc ≥99.5%, sub-pixel-level positioning can still be achieved.

9. The system according to claim 1, characterized in that, The multi-source heterogeneous sensing module adopts a time synchronization triggering mechanism, combining GPS time synchronization with local crystal oscillator calibration to achieve a time synchronization accuracy of ≤50μs for data acquisition from each sensor. The time synchronization error calculation formula is as follows: ΔT = |T_GPS - T_osc|; In the formula, ΔT is the time synchronization error, T_GPS is the GPS timing time, and T_osc is the local crystal oscillator time; Perform calibration every 10 seconds to ensure that ΔT is always ≤50μs; The module housing adopts a one-piece magnesium-aluminum alloy structure, integrating IP67 protection design for waterproofing, dustproofing, and impact resistance.

10. The system according to claim 4, characterized in that, The execution adjustment module also includes a pre-docking posture planning unit. This unit, based on the component docking node parameters in the BIM model and combined with the hoisting path data, uses the A* algorithm to plan the optimal pre-docking posture, reducing the posture adjustment amount during the docking process by ≥30%. The formula for calculating the adjustment amount optimization rate is as follows: In the formula, η is the optimization rate of the adjustment amount, and Δθ ori9 This represents the total attitude adjustment amount before planning. This represents the total attitude adjustment amount after planning; η≥30%.

Images