Prefabricated construction assembly sequence optimization method

Abstract

The invention provides a fabricated construction assembly sequence optimization method, and belongs to the technical field of computer-aided process optimization. Firstly, construction site radar three-dimensional point cloud data are acquired, an RGB-D sensor acquires an image containing color and depth information, and an inertial measurement unit acquires pose data of the two sensors; then, through the processes of point cloud data conversion, point cloud data coordinate conversion, point cloud data time alignment, point cloud data fusion, construction site 3D point cloud data modeling, assembly sequence optimization based on reinforcement learning and online verification feedback of the assembly sequence, all the steps are connected with one another, and dynamic optimization of the optimal assembly strategy is jointly achieved. Through multi-source sensor fusion, the point cloud drift is effectively reduced, the modeling precision is improved, the assembly sequence is autonomously optimized by adopting reinforcement learning, and compared with a traditional genetic algorithm or manual rules, the flexibility and adaptability of an assembly strategy are improved, the assembly feasibility is fed back in real time, and the construction reliability is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer-aided process optimization, and in particular relates to a method for optimizing an assembly sequence in prefabricated construction. Background Art

[0002] Prefabricated buildings, due to their high efficiency, environmental friendliness, and energy-saving characteristics, are becoming a key development trend in the global construction industry. Compared to traditional cast-in-place construction methods, prefabricated buildings utilize factory-produced prefabricated components and on-site assembly, improving construction efficiency, reducing construction waste, and optimizing construction quality. However, due to the large number of prefabricated components, machinery, and the coordination of construction personnel, the complexity and dynamic nature of the construction process pose significant challenges to assembly management. In actual construction, the rationality of the assembly sequence directly impacts construction efficiency and costs. An improper assembly sequence can lead to site congestion, low machinery utilization, increased worker wait times, and even operational conflicts that can compromise construction safety. Therefore, scientifically and rationally optimizing the assembly sequence to ensure a smoother and more efficient construction process is a core issue in prefabricated building management.

[0003] Currently, the level of intelligence in prefabricated construction still needs to be improved. Although BIM (Building Information Modeling) has been widely used in construction planning, it is primarily used for design and management, rather than for intelligently optimizing assembly sequences. Construction site environments are complex and ever-changing. Existing methods often rely on manual experience or static planning, making them difficult to adapt to dynamic construction needs. On the other hand, with the rapid development of the Internet of Things, sensor technology, and building digital twins, the information perception capabilities of construction sites have been significantly enhanced. By acquiring real-time data from the construction site through multi-source sensors and combining it with intelligent decision-making methods, assembly processes can be effectively optimized and construction efficiency improved. However, how to utilize this multi-source data to construct an accurate construction site model in a dynamic environment and optimize the assembly sequence based on this real-time data remains a difficult and challenging issue in current research. Furthermore, uncertainties in the construction site, such as weather changes, material supply delays, and equipment failures, can affect the execution of the assembly sequence. Therefore, building an intelligent optimization system that can adapt to environmental changes to make prefabricated construction more efficient and safer is a key issue that needs to be addressed.

[0004] Currently, the optimization methods for assembly sequences in prefabricated construction mainly rely on two types of technologies:

[0005] Rule-based manual optimization and mathematical model-based optimization algorithms. The former mainly relies on the experience of construction experts to formulate assembly rules and make manual adjustments. Although this method can guarantee the feasibility of construction to a certain extent, it has the following disadvantages: (1) Poor adaptability: Rules are usually based on historical experience and are difficult to adapt to the complex and changing construction site environment; (2) Difficulty in processing large-scale data: As the scale of construction expands, manual optimization methods cannot efficiently handle the arrangement and combination of a large number of assembly components; (3) Limited optimization effect: The optimization ability of manual rules is limited, and it is difficult to find the truly optimal assembly sequence.

[0006] Optimization methods based on mathematical models, such as linear programming, integer programming, or genetic algorithms, can improve the optimization effect of assembly sequences to a certain extent, but they still have the following limitations: (1) Static nature: Many mathematical optimization methods are based on pre-set construction conditions, while construction sites are usually highly dynamic, which limits the optimization effect of these methods; (2) High computational complexity: When the assembly sequence involves a large number of components, the computational complexity of traditional optimization methods increases sharply, making it difficult to apply them to the construction site in real time; (3) Lack of real-time feedback mechanism: Most existing methods are based on offline optimization and do not fully consider unexpected situations that may occur during the construction process, such as component errors, equipment scheduling conflicts, etc., and lack a flexible adjustment mechanism.

[0007] Overall, existing assembly sequence optimization methods are difficult to simultaneously take into account real-time performance, adaptability, and optimization efficiency. Therefore, there is an urgent need for an intelligent method that combines construction site perception capabilities and can dynamically adjust optimization strategies to improve the efficiency and intelligence level of prefabricated construction. Summary of the Invention

[0008] In response to the above problems, the present invention provides a method for optimizing assembly sequence of prefabricated construction, which includes the following steps:

[0009] S1: The radar sensor acquires 3D point cloud data from the construction site, the RGB-D sensor collects images containing color and depth information, and the inertial measurement unit collects pose data from the above two sensors;

[0010] S2, converts the collected image containing color and depth information into RGB three-dimensional point cloud data;

[0011] S3, time-aligns the radar 3D point cloud data, RGB 3D point cloud data, and pose data;

[0012] Use the pose data to construct a transformation matrix, and based on the constructed transformation matrix, perform coordinate transformation on the two time-aligned 3D point cloud data and map them to the same coordinate system;

[0013] S4, performing point cloud registration on the two 3D point cloud data processed in S3, obtaining fused point cloud data, and constructing 3D point cloud data of the construction site;

[0014] S5, build a reinforcement learning framework, input the construction site 3D point cloud data, the set of components to be assembled, and the construction status data into the built reinforcement learning framework for iterative optimization to obtain the assembly sequence;

[0015] S6, defines the feedback scoring function, optimizes the assembly sequence based on assembly error analysis and collision detection, and obtains the maximum scoring function. If the score does not meet the standard, it returns to S6 for optimization and adjustment, and finally outputs the optimal assembly sequence.

[0016] Preferably, in S3, a linear time interpolation formula is used to perform time synchronization on the radar three-dimensional point cloud data to the RGB three-dimensional point cloud data;

[0017] At the same time, the posture data is interpolated in time, and the posture data is calculated at t D The transformation matrix T at the moment IMU (t D ), and temporally aligned with the RGB 3D point cloud data.

[0018] Preferably, the specific process of obtaining the fused point cloud data in S4 is:

[0019] First, use the external parameter transformation matrix T L→D Perform coordinate transformation; according to the installation position of the radar and RGB-D sensor, measure the coordinate rotation relationship of the two sensors to obtain the 3*3 rotation matrix R, and measure the spatial displacement of the two sensors to obtain the 3*1 translation vector T; therefore, the external parameter transformation matrix T L→D Including rotation matrix R and translation vector T; T L→D The coordinate transformation is specifically defined as: first, matrix multiplication of the radar point cloud data using the rotation matrix R, and then adding the translation vector T to convert the radar point cloud data into the RGB-D coordinate system;

[0020] After converting the radar point cloud data into the RGB-D coordinate system, it is directly aligned with the RGB three-dimensional point cloud data obtained by the RGB-D sensor to obtain a coarse registration result. Then, the iterative closest point (ICP) method is used for precise registration. Specifically, based on the coarse registration result, the closest corresponding points in the two point cloud data are matched by Euclidean distance, and the external parameter transformation matrix T is calculated based on the error of each set of corresponding points. L→D The rotation matrix R and translation vector T are optimized and adjusted, and the optimized external parameter transformation matrix T is finally obtained. opt ; and use T opt Achieve more accurate coordinate transformation and data alignment;

[0021] Then, the nearest RGB-D color information is found for the radar point. The nearest point index is found by using the method of finding the nearest RGB-D point. The closest RGB-D sampling point is found by calculating the Euclidean distance. After matching the color information of the nearest neighbor point, the RGB-D color information (R, G, B) is assigned to the radar point cloud to generate a colored point cloud.

[0022] Preferably, the specific process of constructing the 3D point cloud data of the construction site in S4 is:

[0023] First, the point cloud is voxelized and converted into a regular grid structure to reduce the computational complexity of the point cloud.

[0024] Secondly, voxel downsampling is performed. After voxelization, each voxel contains multiple points. A punctuated point is selected to reduce the amount of data. The amount of data is reduced by voxelization and downsampling. The final 3D construction site model is composed of P V (i) The constructed point cloud data is obtained.

[0025] Preferably, the reinforcement learning framework constructed in S5 is to enable the intelligent agent to V (i) The set of components to be assembled C = {c1, c2, ..., c N}、Under the 3D point cloud data corresponding to each component, learn the assembly sequence S * ; Determine the next optimal installation component c through reinforcement learning strategy t , the reinforcement learning framework is defined as follows:

[0026] Status t :The reinforcement learning agent needs to perceive the state of the construction scene at time t, including the point cloud data of the current assembly; create a component set C, and record the currently installed component sequence C t =(c1, c2, ..., c t ), then the remaining components that can be assembled are: C r =(c t+1 ,...,c N );

[0027] Action a t :The agent selects the next component c to install t+1 ;

[0028] Strategy π θ (a t |s t ): Given the current state s t When , take action a t The probability distribution of

[0029] Reward R t: Evaluate the pros and cons of the current assembly sequence and select the construction efficiency, safety and stability of the component;

[0030] Goal: Learn the optimal policy π θ To maximize the cumulative reward; that is, to solve the assembly order: Indicates the i-th installed component, whose index order is t i Need to be optimized, N is the total number of components that need to be assembled, at each step t, the agent is based on the current state s t Select the optimal assembly action a t , that is, select the next component c to be installed t , the optimization goal is:

[0031]

[0032] Where γ is the discount factor, which controls the impact of future rewards on current decisions, and argmax S It means selecting the sequence with the largest cumulative reward among all assembly sequences.

[0033] Preferably, the reinforcement learning framework built on S5 uses the nearest neighbor policy optimization (PPO) for policy optimization:

[0034] The agent passes the value function V(s t ) Assess the current construction status t The long-term benefits that may be obtained in the future, if the construction sequence is reasonable, the future benefits V(s t ) is higher; and, using the advantage function A t Evaluate the currently selected action a t Is it better than the average strategy?

[0035] Perform PPO strategy update. PPO uses a clipped objective function to stabilize training:

[0036] L(θ)=E π [min(r t (θ)A t ,clip(r t (θ), 1-∈, 1+∈)A t )]

[0037] Where L(θ) represents the policy optimization objective function, represents the probability ratio of the new and old strategies, ∈ is the PPO clipping range, which is used to limit the update amplitude of the strategy, and clip(·) is clipping the strategy ratio to ensure that the strategy update does not deviate too much.

[0038] Preferably, the geometric information of the point cloud is introduced to optimize the assembly sequence, and the agent learns to optimize the assembly sequence through the reward function, and the reward function Rt include:

[0039] Construction time optimization: component c t Installation time T(c t ) Construction equipment efficiency v w Calculate and set the reward item R time (t)=-T(c t ), the goal is to minimize the construction time, so the reward takes a negative value;

[0040] Structural stability constraint: Introduce point cloud center of gravity calculation to determine the support relationship of components. If the constraint G(c i ) z ≥G(c j ) z , a negative reward is given:

[0041]

[0042] Among them, λ s represents the penalty factor, G(c i ) z For component c i The height of the center of gravity, G(c j ) z Indicates the supporting member c j The height of the center of gravity;

[0043] Construction safety: Define the minimum safe distance D(c i , c j ), if D(c i , c j )<d min , then the penalty is:

[0044]

[0045] Among them, d min Indicates the minimum allowable safety distance, λ d is the collision penalty factor;

[0046] The final reward function is expressed as follows:

[0047] R t =ω1R time (t)+ω2R stability (t)+ω3R safety (t)

[0048] Among them, ω1, ω2 and ω3 are reward weighting coefficients.

[0049] Preferably, in S6, a feedback verification process is performed based on the obtained assembly sequence S*, combined with real-time construction point cloud data and assembly constraints as input. Whenever a component is installed, the construction scene is synchronously updated and used as actual environment information. Dynamic environment modeling is used to provide interactive feedback to the next component to be placed:

[0050] First, error analysis and position adjustment are performed. For each component c in the assembly sequence, t , its theoretical installation position is P C (c t ), and the actual installation location comes from the on-site point cloud P V (i), the deviation between the two is defined as ΔP(c t ), if ||ΔP(c t )||>δ, the deviation exceeds the limit and the component position is adjusted. If assembly deviation is detected, the compensation transformation matrix T needs to be calculated. comp ; Get the new position P' of the assembly component C (c t );

[0051] After that, real-time collision detection is performed; collision detection is performed using nearest neighbor search:

[0052]

[0053] Where, d min (c t , c j ) is component c t With assembled component c j The minimum distance between them. C (c t ) and P′ C (c j ) represent the components to be assembled c t With assembled component c j Point cloud data, and p and q represent one of the point clouds respectively. If d min (c t , c j )<d safe , then a collision occurs, then let II(C collision )=1,d safe Indicates the safety distance threshold;

[0054] Finally, a comprehensive evaluation of assembly feedback, integrated error analysis and collision detection is performed, and a feedback scoring function is defined. The feedback scoring function is maximized to achieve the final optimization goal, which is R.

[0055] Determine the maximization feedback scoring function R of the optimization goal to minimize assembly errors, avoid collisions and ensure stability. When R exceeds the set threshold, feedback information is sent to S5, and the subsequent component installation sequence is re-optimized based on the current construction environment. If it does not exceed the threshold, the optimal assembly sequence S obtained in S5 is used. * Continue to assemble components and finally output an optimal assembly sequence

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. Multi-source sensor fusion 3D construction site modeling method: This paper proposes a 3D construction site modeling method based on the fusion of LiDAR, RGB-D cameras, and IMU sensors. Compared with existing single-sensor modeling methods, this method fully utilizes the high precision of laser point clouds, the rich texture features of RGB-D visual information, and the posture compensation of IMU inertial data. Through time synchronization, spatial registration, and data fusion techniques, it achieves higher-precision 3D reconstruction of construction sites.

[0058] 2. Assembly Sequence Optimization Method Based on Proximal Policy Optimization: This paper proposes an assembly sequence optimization method based on PPO reinforcement learning. Unlike traditional genetic algorithms or manual rules, this method can autonomously learn the optimal assembly sequence and adapt to dynamic environmental changes. Through iterative updates of reinforcement learning strategies, value function optimization, and exploration and utilization mechanisms, it continuously improves assembly optimization strategies, enhancing assembly efficiency and stability.

[0059] 3. Assembly Sequence Verification and Feedback Method Based on 3D Modeling: This paper proposes a real-time assembly sequence verification and feedback method based on 3D modeling of the construction site. Unlike existing offline simulation strategies, this method can verify the feasibility of assembly sequences in real time during construction. This method uses collision detection, assembly error calculation, and other methods to optimize the assembly sequence through a feedback scoring mechanism, improving construction accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of the overall technical route of the present invention.

[0061] Figure 2 Model a flow chart for multi-source perception data collection and processing.

[0062] Figure 3 Schematic diagram of construction assembly components after data fusion.

[0063] Figure 4 Optimizing flow charts for assembly sequences based on PPO reinforcement learning.

[0064] Figure 5Verify the feedback flow chart for the assembly sequence.

[0065] Figure 6 This is a comparison chart of the assembly error between the PPO reinforcement learning reward change and the greedy algorithm in the embodiment.

[0066] Figure 7 This is an assembly sequence diagram of the 10 components in the embodiment. DETAILED DESCRIPTION

[0067] This paper proposes a method for optimizing assembly sequence in prefabricated construction based on multi-source perception and reinforcement learning to improve the efficiency and accuracy of prefabricated construction. This method mainly includes four core steps: 3D modeling of the construction site, multi-source data fusion, assembly sequence optimization based on reinforcement learning, and online verification and feedback of the assembly sequence. Each step is interconnected to achieve dynamic optimization of the optimal assembly strategy. The overall technical route is as follows: Figure 1 shown.

[0068] The implementation process of the present invention is further described below with reference to specific embodiments.

[0069] 1. Multi-source data perception and collection

[0070] The environment of an assembled construction site is complex, involving the dynamic interaction of a large number of components, equipment, and personnel. In order to achieve an efficient and accurate assembly process, it is first necessary to perform high-precision modeling of the construction site to provide real, real-time environmental information. At present, a single sensor method is difficult to meet the complex and changing needs of a construction site. Therefore, the present invention adopts a multi-source perception sensor of LiDAR (laser radar), RGB-D camera (color-depth camera) and IMU (inertial measurement unit) to improve the comprehensiveness, accuracy and robustness of the data.

[0071] (1) The point cloud data P obtained by the LiDAR used in the present invention L =(x i ,y i , z i ), which represents the high-precision three-dimensional coordinate information of the construction site (x i ,y i , z i ), i represents the i-th point in the point cloud data, x i Indicates the coordinates of the point cloud data in the x-axis direction, that is, the left and right position of the point relative to the LiDAR device, and y i Indicates the coordinate of the point cloud data in the y-axis direction, that is, the front and back position of the point relative to the LiDAR device, z iRepresents the z-axis coordinate of the point cloud data, i.e., the height of the point relative to the LiDAR device. 3D point cloud data offers long-range, high-precision performance, making it suitable for scanning large-scale environments. The sampling frequency is 20Hz, and the number of laser lines is 64. It generates a high-density 3D point cloud at approximately 1,300,000 points per second.

[0072] (2) RGB-D camera simultaneously captures color I RGB With depth data D RGB , providing color images and depth information, which can supplement details that LiDAR cannot capture, such as surface textures and small objects. Depth resolution is 1280*720, RGB resolution is 1920*1080, and the frame rate is 30fps; the specific data collected is:

[0073] I RGB ={(R u，v , G u，v , B u，v )} (u，v)∈Ω

[0074] D RGB (u, v) = {d u，v} (u，v)∈Ω

[0075] Where (u, v) represents a pixel point. (R u，v , G u，v , B u，v ) is the RGB pixel value at (u, v). Ω represents the set of all pixels in the image; d u，v Indicates the depth information corresponding to the pixel point (u, v).

[0076] (3) IMU pose data I pose =(p x , p y , p z ,q x ,q y ,q z ,q w ). (p x , p y , p z ) represents the spatial position of the sensor, (q x ,q y ,q z ,q w ) is the rotation angle expressed by quaternion. pose Provides the spatial position and attitude information of the sensor, ensures the spatial alignment of multiple sensor data, and improves the accuracy of data fusion. The data update frequency is 400Hz and uses UART / CAN transmission.

[0077] The three sensors collect initial data information about all components N and the construction scene. The data then needs to be processed to make it more suitable for the component assembly task.

[0078] 2. Conversion of 3D point cloud data

[0079] If the initially collected data contains noise, in order to improve data quality, data processing is performed on the collected data, including data preprocessing and data fusion. Before data fusion, the data collected by each sensor needs to be denoised, coordinate transformed, and aligned. After data processing, the environmental point cloud information of the construction site is obtained. The specific steps include:

[0080] (1) First, denoise the LiDAR point cloud data. Since LiDAR point cloud data may contain noise, mismeasured points, and outliers, a statistical filtering method is used for denoising:

[0081] P′ L ={p∈P L |||p-μ||<kσ}

[0082] Among them, P′ L represents the denoised point cloud dataset, μ and σ are the mean and standard deviation of the local area respectively, k is the filtering parameter, and p represents each point in the point cloud dataset.

[0083] Use bilateral filtering to denoise the depth map and get D′ RGB (u, v):

[0084]

[0085] Where W represents the normalization factor, (i, j) represents the index of the adjacent pixel, G s (iu, jv) is the weight calculated based on the spatial distance between pixels, using Gaussian function. The closer the pixel position, the greater the weight. r (D RGB (i, j)-D RGB (u, v)) represents the weight of similarity calculation between depth values, using Gaussian function. The closer the depth values are, the greater the weight.

[0086] (2) Secondly, the color image and depth map collected by the RGB-D camera are converted into RGB three-dimensional point cloud. The calculation formula is:

[0087]

[0088] Among them, i is the index of the point cloud data point, as well as Represents the transformed 3D point cloud coordinates; f x , fy represents the focal length of the camera, c x , c y The main point coordinates. Therefore, the processed RGB-D point cloud data P D Expressed as:

[0089]

[0090] Where i is the index of the point cloud data point. is the 3D point coordinate after RGB-D camera conversion, Indicates the RGB pixel value corresponding to the i-th point cloud data index.

[0091] 3. Temporal interpolation of 3D point cloud data

[0092] Since the sampling times of LiDAR and RGB-D cameras are different, the sensor data needs to be time synchronized. Specifically, LiDAR scans point by point, while RGB-D is collected frame by frame. Therefore, the point cloud collected by LiDAR needs to be temporally interpolated and aligned to the timestamp t of the RGB-D acquisition frame. D The present invention uses a linear time interpolation formula to synchronize LiDAR to RGB-D:

[0093]

[0094] in, Indicates that at time t D The LiDAR point cloud coordinates under; Indicates t D The coordinates of the LiDAR sampling point at the most recent time point; Indicates t D The coordinates of the LiDAR sampling point at the nearest time point; Indicates a point timestamp; for point timestamp.

[0095] Secondly, perform time synchronization from IMU to RGB-D. The sampling frequency of IMU is usually higher than that of RGB-D and LiDAR, so the IMU data (pose data) is time interpolated to calculate the IMU at t D The transformation matrix T at the moment IMU (t D ). The IMU transformation matrix T of the present invention IMU (t D )The linear interpolation formula is as follows:

[0096]

[0097] in, tD The IMU transformation matrix at the most recent time point; t D The IMU transformation matrix at the latest time point; and The corresponding IMU sampling timestamp.

[0098] Next, the pose data I collected by IMU pose Used to construct the transformation matrix T IMU .

[0099]

[0100] Among them, R IMU =(q x ,q y ,q z ,q w ) T The rotation matrix calculated by IMU. IMU =(p x , p y , p z ) T The position offset vector provided by the IMU. Use the IMU transformation matrix T IMU (t D )Transform the LiDAR point cloud to the IMU coordinate system:

[0101]

[0102] The transformation from RGB-D to IMU coordinate system also uses the formula P′ D =T IMU ·P D accomplish.

[0103] After the above process, the time-synchronized LiDAR data P″ is obtained L and RGB-D data P′ D .

[0104] 4. Data Fusion

[0105] After time synchronization is completed, LiDAR and RGB-D point clouds still have problems such as coordinate system differences, accuracy differences, and data redundancy. Therefore, data fusion is required to form a complete, high-precision, colored three-dimensional point cloud model.

[0106] First, coordinate conversion is performed. Since the LiDAR and RGB-D sensors are installed in different locations, their data are located in different coordinate systems, and the external parameter transformation matrix T is required. L→D Convert the LiDAR point cloud to RGB-D coordinate system.

[0107]

[0108]

[0109] in, is the LiDAR point cloud converted to RGB-D coordinate system. L→D is the rotation matrix, t L→D is the translation vector.

[0110] Secondly, after the coordinates are converted, there may still be slight deviations due to the sensor calibration error, and further point cloud registration is required to align the two sets of point clouds more accurately.

[0111] Optimizing transformation matrices is the i-th point of the LiDAR point cloud in the RGB-D coordinate system. T is the coordinate transformation matrix of the current iteration. N is the total number of point clouds. After the registration is completed, the optimized T opt Transform the LiDAR point cloud to align it with the RGB-D point cloud:

[0112] Since LiDAR sampling points are sparse and RGB-D sampling points are dense, it is necessary to find the nearest RGB-D color information for the LiDAR point. The nearest point index is found by finding the nearest RGB-D point:

[0113]

[0114] Among them, j * Is the LiDAR point The nearest RGB-D point index. The closest RGB-D sampling point is found by calculating the Euclidean distance. After matching the color information of the nearest neighbor point, the RGB-D color information (R, G, B) is assigned to the LiDAR point to generate a color point cloud:

[0115]

[0116] Among them, P F (i) represents the fused point cloud data; is the LiDAR point cloud coordinate; (R j , G j , B j ) is the matching RGB-D color value.

[0117] 5. 3D construction site modeling with point cloud integration

[0118] After completing the fusion of LiDAR and RGB-D point cloud data, a complete point cloud dataset with geometric structure information and color information is obtained. This dataset contains spatial coordinate and color information, which is then used for 3D construction site modeling to construct a structured 3D environment, supporting subsequent assembly sequence optimization.

[0119] First, in order to reduce the computational complexity of the point cloud, the point cloud is voxelized and converted into a regular grid structure. Voxel index calculation:

[0120]

[0121] Among them, s v Represents the voxel size, which controls the fineness of the voxel network. V(i) represents the point P F (i) The voxel index to which the points belong. The value is rounded down to ensure that adjacent points belong to the same voxel.

[0122] Secondly, voxel downsampling. After voxelization, each voxel may contain multiple points. It is necessary to select a marked point to reduce the amount of data to improve computational efficiency. Voxel center point calculation:

[0123]

[0124] Where, P V (i) represents the representative point of volume V(i). By voxelization and downsampling, the amount of data is reduced and the computational efficiency is improved. The final 3D construction site model is composed of P V (i) The constructed point cloud data is obtained and used for subsequent construction planning and optimization.

[0125] The overall process of data collection, data preprocessing, data fusion, and building a 3D point cloud data model of the construction site is as follows: Figure 2 shown.

[0126] The point cloud and color data are collected by sensors according to the original construction environment, and then the three-dimensional point cloud data diagram is obtained after data fusion for assembly sequence optimization. For example Figure 3 shown.

[0127] 6. Design of assembly sequence optimization module based on PPO reinforcement learning

[0128] In prefabricated construction, different components need to be installed in a reasonable order to ensure the stability, safety and efficiency of the construction. Due to the complex and dynamic environment of the construction site, traditional methods based on artificial rules or genetic algorithms are difficult to adapt. Therefore, the present invention uses an assembly sequence optimization method based on PPO reinforcement learning to enable the intelligent agent to optimize the assembly sequence in the 3D construction environment point cloud data P V (i) The set of components to be assembled C = {c1, c2, ..., c N}、3D point cloud data corresponding to each component, learning assembly sequence S *.

[0129] 1. Build a reinforcement learning framework design

[0130] The assembly sequence optimization is modeled as a reinforcement learning framework based on a Markov decision process. The agent uses a reinforcement learning strategy to decide the next optimal component to be installed c. t , so that the assembly sequence can minimize the construction time. Ensure construction safety and meet the structural stability constraints. The reinforcement learning framework is defined as follows:

[0131] Status t :The reinforcement learning agent needs to perceive the state of the construction scene at time t, including the point cloud information P of the current assembly V (i). Create component set C and record the currently installed component sequence C t =(c1, c2, ..., c t ), then the remaining components that can be assembled are: C r =(c t+1 ,...,c N ).

[0132] Action a t :The agent selects the next component c to install t+1 .

[0133] Strategy π θ (a t |s t ): Given the current state s t When , take action a t The probability distribution of .

[0134] Reward R t :Evaluate the pros and cons of the current assembly sequence. Select the construction efficiency, safety and stability of the component.

[0135] Goal: Learn the optimal policy π θ To maximize the cumulative reward. That is, to solve the assembly order: Indicates the i-th installed component, whose index order is t i Need to be optimized. N is the total number of components that need to be assembled. At each step t, the agent is based on the current state s t Select the optimal assembly action a t , that is, select the next component c to be installed t The optimization goal is:

[0136]

[0137] Where γ is the discount factor that controls the impact of future rewards on current decisions. SIt means selecting the sequence with the largest cumulative reward among all possible assembly sequences.

[0138] 2. Value Function and Policy Update

[0139] In reinforcement learning, the value function is used to evaluate the quality of the current state, while the policy update determines how to optimize assembly decisions. To adapt to the dynamic environment and complex constraints of prefabricated construction, this paper designs policy optimization within the framework of Proximal Field Policy Optimization (PPO).

[0140] The agent passes the value function V(s t ) Assess the current construction status t Possible long-term benefits in the future:

[0141]

[0142] Where γ is the discount factor that controls the impact of future rewards on current decisions. k represents the index of the future time step. If the construction sequence is reasonable, the future benefit V(s t ) is higher.

[0143] And, using the advantage function A t Evaluate the currently selected action a t Is it better than the average strategy?

[0144] A t =R t +γV(s t+1 )-V(s t )

[0145] Furthermore, the PPO strategy is updated. PPO uses a clipped objective function to stabilize training:

[0146] L(θ)=E π [min(r t (θ)A t ,clip(r t (θ), 1-∈, 1+∈)A t )]

[0147] Where L(θ) represents the policy optimization objective function, represents the probability ratio of the new and old strategies. ∈ is the PPO clipping range, which is used to limit the update amplitude of the strategy. clip(·) is to clip the strategy ratio to ensure that the strategy update does not deviate too much.

[0148] 3. Assembly strategy optimization based on 3D point cloud data

[0149] In the construction environment, assembly optimization needs to be based on point cloud data P V(i) Calculate the spatial relationship of components and influence the strategic decision of the agent. The present invention introduces the geometric information of point clouds to optimize the assembly sequence. The agent learns to optimize the assembly sequence through the reward function. The reward function R t It consists of the following parts:

[0150] 1) Construction time optimization. Component c t Installation time T(c t ) Construction equipment efficiency v w calculate:

[0151]

[0152] Among them, V(c t ) represents component c t And set the reward term R time (t)=-T(c t ). The goal is to minimize the construction time, so the reward is negative.

[0153] 2) Structural stability constraints. During the assembly process, some components need to be assembled after certain supporting components are installed. Therefore, the present invention introduces point cloud centroid calculation to determine the component support relationship:

[0154]

[0155] G(c i ) z ≥G(c j ) z

[0156] In the formula, G(c t ) represents component c t The center of gravity, G(c i ) z For component c i The height of the center of gravity, G(c j ) z Indicates the supporting member c j The height of the center of gravity. If the constraint G(c i ) z ≥G(c j ) z , a negative reward is given:

[0157]

[0158] Among them, λ s Represents the penalty factor.

[0159] 3) Construction safety. Define the minimum safe distance D(c i , c j ):

[0160]

[0161] Among them, P V (i) and P V (j) represent the voxel centroid of the i-th and j-th components respectively. If D(c i , c j )<d min , then the penalty is:

[0162]

[0163] Among them, d min Indicates the minimum allowable safety distance, λ d is the collision penalty factor.

[0164] Therefore, the final reward function is expressed as follows:

[0165] R t =ω1R time (t)+ω2R stability (t)+ω3R safety (t)

[0166] Among them, ω1, ω2 and ω3 are reward weighting coefficients.

[0167] 4. Assembly sequence S * The specific solution process

[0168] (1) Data input. Input the 3D point cloud data of the construction site, the set of components to be assembled, and the construction status data information (the set of components to be assembled refers to the dictionary type data consisting of the types and quantities of prefabricated building components that the construction unit plans to install. The key of the dictionary represents the name of the prefabricated building component, and the value corresponding to each key is the number of the component. The construction status data is the area of the current building where component installation has been completed and the area where component installation has not been completed, determined based on the 3D construction environment point cloud data. Specifically, it is dictionary type data, the key is the area in the building where components should be installed, and the value corresponding to each key is whether the component is installed in the area, with 1 indicating installation and 0 indicating non-installation).

[0169] (2) Through strategy π θ Select the next component. At each time step t, the policy π θ Select an assembly component P θ (c|s t ) represents the probability distribution of the policy network output, that is, the probability of selecting a component.

[0170] (3) Construction status update. Assemble components c t After that, update the construction status:t+1 =f(s t , a t ). Update the assembled component set, update the point cloud data, and check whether the construction constraints are met. If not, the strategy is penalized.

[0171] (4) Calculate the immediate reward. Calculate the immediate reward and impose a penalty if the construction constraints are not met.

[0172] (5) Calculate the advantage estimate. The formula for calculating the advantage estimate measures the degree of improvement of the current strategy relative to the old strategy. If A t If it is positive, it means that the assembly action c selected by the current strategy t is valid. Otherwise, it means the strategy needs to be adjusted to optimize the assembly sequence.

[0173] (6) Strategy update. Use the PPO objective function to update the strategy. The root strategy update method calculates the L(θ) optimization target. Make the strategy more inclined to choose the assembly sequence with high reward. Then update the strategy parameters, α represents the learning rate, which controls the step size of the policy update. PPO ensures the stability of the optimization process of the assembly sequence by limiting the amplitude of the policy update.

[0174] (7) Output assembly sequence S * Output the assembly sequence calculated based on the optimization goal in S601 This assembly sequence meets the requirements of optimal construction time, optimal stability and optimal safety.

[0175] The assembly sequence optimization process based on PPO reinforcement learning is as follows Figure 4 As shown, after the neighboring field reinforcement learning module, an assembly sequence S* can be obtained. This sequence is taken as the current optimal assembly sequence result. t Assemble and adjust the assembly sequence in real time through the verification feedback module according to the environmental conditions generated by the currently assembled components.

[0176] 7. Assembly Sequence Verification Feedback Module Design

[0177] In prefabricated construction, traditional offline simulation feedback methods usually rely on preset rules or limited simulation data for evaluation, and cannot adapt to changes in the construction site in real time, resulting in the possibility that the optimized assembly sequence S* may deviate in actual construction, thereby affecting construction efficiency and safety. Therefore, the present invention designs an assembly sequence verification and feedback method for 3D modeling. Combined with online point cloud data updates for real-time collision detection, dynamic stability analysis and sequence adjustment strategies, efficient and fast assembly feedback is achieved. The assembly sequence verification feedback process is as follows: Figure 5 shown.

[0178] According to the obtained assembly sequence S * , combined with real-time construction point cloud data P V (i) and assembly constraints as input. A feedback verification process is performed. Whenever the assembly sequence optimization module places a component, the construction scene is synchronously updated and used as actual environment information. Dynamic environment modeling is used to provide interactive feedback to the next component to be placed.

[0179] (1) First, error analysis and position adjustment are performed. For each component c in the assembly sequence, t , its theoretical installation position is P C (c t ), and the actual installation location comes from the on-site point cloud P V (i), the deviation between the two is defined as:

[0180] ΔP(c t )=P C (c t )-P V (i)

[0181] Among them, ΔP(c t ) represents the deviation vector between the theoretical model and the construction site point cloud. If ||ΔP(c t )||>δ, the deviation exceeds the limit and the component position needs to be adjusted. Furthermore, if the assembly deviation is detected, the compensation transformation matrix T needs to be calculated. comp :

[0182]

[0183] Among them, P C (c t ,j) is represented by component c t The coordinates of the j-th point cloud in the theoretical model, P V (i, j) represents the coordinates of the jth point in the construction point cloud. T is the rigid transformation matrix, which includes rotation and translation. The new position of the assembly component is calculated as follows:

[0184] P′ C (c t )=T comp P C (c t )

[0185] (2) After that, real-time collision detection is performed. In order to ensure that no collisions between components occur during the assembly process, the present invention uses nearest neighbor search for collision detection:

[0186]

[0187] Where, d min (ct , c j ) is component c t With assembled component c j The minimum distance between them. C (c t ) and P′ C (c j ) represent the components to be assembled c t With assembled component c j Point cloud data, and p and q represent one of the point clouds respectively. min (c t , c j )<d safe , then a collision occurs, then let II(C collision )=1,d safe Indicates the safety distance threshold.

[0188] (3) Finally, a comprehensive evaluation of assembly feedback is performed. Combining error, collision detection, and stability evaluation, the feedback scoring function is defined as:

[0189]

[0190] Where η1 and η2 are weight parameters used to balance the impact of assembly error, collision, and stability in the score. The final optimization goal is to maximize the feedback score function:

[0191]

[0192] The optimization objective is to maximize the feedback scoring function R to minimize assembly errors, avoid collisions, and ensure stability. When R exceeds the set threshold, the information is fed back to the assembly sequence optimization module, and the subsequent component installation sequence is re-optimized based on the current construction environment. If it does not exceed the threshold, the optimal assembly sequence S obtained in the assembly sequence optimization module is used. * Continue to assemble components, and this method will eventually output an optimal assembly sequence

[0193] 8. Deployment and Application

[0194] This method is actually deployed on the construction site or in the intelligent assembly system. The specific deployment process is as follows:

[0195] Hardware: LiDAR sensors, RGB-D cameras, and IMU sensors are deployed to collect construction site data, and high-performance computing servers are used for data processing and optimization calculations. Industrial robots or automated assembly equipment are used to perform assembly tasks, and PLCs (programmable logic controllers) are integrated for on-site control.

[0196] Software: The 3D modeling and point cloud processing module implements sensor fusion, time synchronization, point cloud stitching, and optimization algorithms to generate a high-precision 3D model of the construction site. The reinforcement learning assembly optimization module uses PPO-based training and optimization, leveraging the computing platform to accelerate strategy training and assembly sequence optimization. The real-time feedback and adjustment module monitors assembly status in real time during the assembly process and dynamically adjusts the assembly sequence using feedback scoring functions.

[0197] System integration and execution. In automated assembly systems at construction sites, the optimal assembly sequence is used as input and combined with 3D modeling data to precisely control the assembly process. Edge computing and wireless communications are used to connect construction robots or intelligent assembly systems, enabling remote optimization and execution of assembly sequences.

[0198] In order to verify the superiority of the assembly sequence optimization method of the present invention, a comparative experiment was conducted using a greedy algorithm to evaluate its optimization effect, computational efficiency, and adaptability. The comparison results and the assembly sequence sorting diagram generated with ten components are shown in Figure 2. Figure 6 and Figure 7 shown.

[0199] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

[0200] Although the above describes the specific implementation methods of the present invention, it does not limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for optimizing assembly sequence of prefabricated construction, characterized in that: The following processes are included: S1: The radar sensor acquires 3D point cloud data from the construction site, the RGB-D sensor collects images containing color and depth information, and the inertial measurement unit collects pose data from the above two sensors; S2, converts the collected image containing color and depth information into RGB three-dimensional point cloud data; S3, time-aligns the radar 3D point cloud data, RGB 3D point cloud data, and pose data; Use the pose data to construct a transformation matrix, and based on the constructed transformation matrix, perform coordinate transformation on the two time-aligned 3D point cloud data and map them to the same coordinate system; S4, performing point cloud registration on the two 3D point cloud data processed in S3, obtaining fused point cloud data, and constructing 3D point cloud data of the construction site; S5, build a reinforcement learning framework, input the construction site 3D point cloud data, the set of components to be assembled, and the construction status data into the built reinforcement learning framework for iterative optimization to obtain the assembly sequence; S6, defines the feedback scoring function, optimizes the assembly sequence based on assembly error analysis and collision detection, and obtains the maximum scoring function. If the score does not meet the standard, it returns to S6 for optimization and adjustment, and finally outputs the optimal assembly sequence.

2. The method for optimizing assembly sequence of prefabricated construction according to claim 1, wherein: In S3, a linear time interpolation formula is used to synchronize the radar 3D point cloud data with the RGB 3D point cloud data; At the same time, the posture data is interpolated in time, and the posture data is calculated at t D The transformation matrix T at the moment IMU (t D ), and temporally aligned with the RGB 3D point cloud data.

3. The method for optimizing assembly sequence of prefabricated construction according to claim 1, wherein: The specific process of obtaining the fused point cloud data in S4 is as follows: First, use the external parameter transformation matrix T L→D Perform coordinate conversion. Based on the installation positions of the radar and RGB-D sensor, measure the coordinate rotation relationship of the two sensors to obtain a 3*3 rotation matrix R, and measure the spatial displacement of the two sensors to obtain a 3*1 translation vector T. Use the rotation matrix R to perform matrix multiplication on the radar point cloud data, and then add the translation vector T to convert the radar point cloud data to the RGB-D coordinate system. After converting the radar point cloud data into the RGB-D coordinate system, it is directly aligned with the RGB three-dimensional point cloud data obtained by the RGB-D sensor to obtain a coarse registration result; Afterwards, the iterative closest point ICP method is used for accurate registration; Based on the coarse registration results, the nearest corresponding points in the two point cloud data are matched by Euclidean distance, and the external parameter transformation matrix T is calculated based on the error of each set of corresponding points. L→D The rotation matrix R and translation vector T are optimized and adjusted to obtain the optimized external parameter transformation matrix T opt ; and use T opt Achieve more accurate coordinate transformation and data alignment; Then, the nearest RGB-D color information is found for the radar point. The nearest point index is found by using the method of finding the nearest RGB-D point. The closest RGB-D sampling point is found by calculating the Euclidean distance. After matching the color information of the nearest neighbor point, the RGB-D color information (R, G, B) is assigned to the radar point cloud to generate a colored point cloud.

4. The method for optimizing assembly sequence of prefabricated construction according to claim 3, wherein: The specific process of constructing the construction site 3D point cloud data in S4 is as follows: First, the point cloud is voxelized and converted into a regular grid structure to reduce the computational complexity of the point cloud. Secondly, voxel downsampling is performed. After voxelization, each voxel contains multiple points. A punctuated point is selected to reduce the amount of data. The amount of data is reduced by voxelization and downsampling. The final 3D construction site model is composed of P V (i) The constructed point cloud data is obtained.

5. The method for optimizing assembly sequence of prefabricated construction according to claim 1, wherein: The reinforcement learning framework constructed in S5 is to enable the intelligent agent to V (i) The set of components to be assembled C = {c1, c2, ..., c N }、Under the 3D point cloud data corresponding to each component, learn the assembly sequence S * ; Determine the next optimal installation component c through reinforcement learning strategy t , the reinforcement learning framework is defined as follows: Status t :The reinforcement learning agent needs to perceive the state of the construction scene at time t, including the point cloud data of the current assembly; create a component set C, and record the currently installed component sequence C t =(c1, c2, ..., c t ), then the remaining components that can be assembled are: C r =(c t+1 ,...,c N ); Action a t :The agent selects the next component c to install t+1 ; Strategy π θ (a t |s t ): Given the current state s t When , take action a t The probability distribution of Reward R t : Evaluate the pros and cons of the current assembly sequence and select the construction efficiency, safety and stability of the component; Goal: Learn the optimal policy π θ To maximize cumulative rewards; That is, solve the assembly sequence: Indicates the i-th installed component, whose index order is t i Need to be optimized, N is the total number of components that need to be assembled, at each step t, the agent is based on the current state s t Select the optimal assembly action a t , that is, select the next component c to be installed t , the optimization goal is: Where γ is the discount factor, which controls the impact of future rewards on current decisions, and argmax S It means selecting the sequence with the largest cumulative reward among all assembly sequences.

6. The method for optimizing assembly sequence of prefabricated construction according to claim 5, characterized in that: Based on the reinforcement learning framework built on S5, the nearest neighbor policy optimization (PPO) is used for policy optimization: The agent passes the value function V(s t ) Assess the current construction status t The long-term benefits that may be obtained in the future, if the construction sequence is reasonable, the future benefits V(s t ) is higher; and, using the advantage function A t Evaluate the currently selected action a t Is it better than the average strategy? Perform PPO strategy update. PPO uses a clipped objective function to stabilize training: L(θ)=E π [min(r t (i)A t ,clip(r t (θ), 1-∈, 1+∈)A t )] Where L(θ) represents the policy optimization objective function, represents the probability ratio of the new and old strategies, ∈ is the PPO clipping range, which is used to limit the update amplitude of the strategy, and clip(·) is clipping the strategy ratio to ensure that the strategy update does not deviate too much.

7. The method for optimizing assembly sequence of prefabricated construction according to claim 5, wherein: The geometric information of the point cloud is introduced to optimize the assembly sequence. The agent learns to optimize the assembly sequence through the reward function. The reward function R t include: Construction time optimization: component c t Installation time T(c t ) Construction equipment efficiency v w Calculate and set the reward item R time (t)=-T(c t ), the goal is to minimize the construction time, so the reward takes a negative value; Structural stability constraint: Introduce point cloud center of gravity calculation to determine the support relationship of components. If the constraint G(c i ) z ≥G(c j ) z , a negative reward is given: Among them, λ s represents the penalty factor, G(c i ) z For component c i The height of the center of gravity, G(c j ) z Indicates the supporting member c j The height of the center of gravity; Construction safety: Define the minimum safe distance D(c i , c j ), if D(c i , c j )<d min , then the penalty is: Among them, d min Indicates the minimum allowable safety distance, λ d is the collision penalty factor; The final reward function is expressed as follows: R t =ω1R time (t)+ω2R stability (t)+ω3R safety (t) Among them, ω1, ω2 and ω3 are reward weighting coefficients.

8. The method for optimizing assembly sequence of prefabricated construction according to claim 1, wherein: In S6, the feedback verification process is performed based on the obtained assembly sequence S*, combined with real-time construction point cloud data and assembly constraints as input. Every time a component is installed, the construction scene is synchronously updated and used as actual environment information. Dynamic environment modeling is used to provide interactive feedback to the next component to be placed: First, error analysis and position adjustment are performed. For each component c in the assembly sequence, t , its theoretical installation position is P C (c t ), and the actual installation location comes from the on-site point cloud data P V (i), the deviation between the two is defined as ΔP(c t ), if ||ΔP(c t )||>δ, the deviation exceeds the limit and the component position is adjusted. If assembly deviation is detected, the compensation transformation matrix T needs to be calculated. comp ; Get the new position P' of the assembly component C (c t ); After that, real-time collision detection is performed; collision detection is performed using nearest neighbor search: Where, d min (c t , c j ) is component c t With assembled component c j The minimum distance between C (c t ) and P′ C (c j ) represent the components to be assembled c t With assembled component c j Point cloud data, and p and q represent one of the point clouds respectively. If d min (c t , c j )<d safe , then a collision occurs, then let d safe Indicates the safety distance threshold; Finally, a comprehensive evaluation of assembly feedback, integrated error analysis and collision detection is performed, and a feedback scoring function is defined. The feedback scoring function is maximized to achieve the final optimization goal, which is R. Determine the maximization feedback scoring function R of the optimization goal to minimize assembly errors, avoid collisions and ensure stability. When R exceeds the set threshold, feedback information is sent to S5, and the subsequent component installation sequence is re-optimized based on the current construction environment. If it does not exceed the threshold, the optimal assembly sequence S obtained in S5 is used. * Continue to assemble components and finally output an optimal assembly sequence

Images