An orbital cloth spreading system based on cloud factory collaboration

Abstract

The invention discloses a track-type material placing system based on cloud-based factory collaboration, and relates to the technical field of material placing systems. The system comprises: a material placing unit, comprising a plurality of material placing machine units sliding along double guide rails, and an embedded controller for real-time collection of slump, pumping pressure, boom inclination and ambient wind speed data; an edge gateway, deployed inside each unit, configured with a sliding time window filtering algorithm and a feature extraction module, for compressing sensor data into a standardized construction feature stream containing a timestamp; a cloud-based federated learning platform, integrating a digital twin engine, for receiving the feature streams of each edge gateway, constructing a multi-objective optimization model, and generating a collaborative control strategy containing path planning, hydraulic parameters and obstacle avoidance instructions; a human-machine collaborative terminal, for parsing the collaborative control strategy into boom motion trajectories and valve control parameters, and sending them to the material placing unit, so as to achieve multi-machine posture synchronization and stress balance control during the pouring process.

Description

Technical Field

[0001] This application relates to the technical field of fabric systems, and more particularly, to an orbital fabric system based on cloud factory collaboration. Background Art

[0002] The orbital fabric system is an essential key device in modern construction. Its main function is to evenly transport concrete to the designated construction area through rail sliding, ensuring the efficiency and quality stability of the pouring process. As an important part of construction mechanization, the orbital fabric system is widely used in complex engineering scenarios such as high-rise buildings, bridges, and tunnels.

[0003] However, existing orbital fabric systems mostly adopt a single-machine operation mode and lack intelligent collaboration capabilities. This mode mainly relies on manual experience for path planning and construction parameter adjustment, resulting in significant limitations in construction efficiency and quality. Especially in complex construction environments, traditional fabric systems are difficult to optimize construction parameters in real time, thus affecting the stability of pouring quality.

[0004] To address the above problems, no effective solutions have been proposed yet. Summary of the Invention

[0005] Embodiments of this application provide an orbital fabric system based on cloud factory collaboration to solve the above technical problems.

[0006] This application provides an orbital fabric system based on cloud factory collaboration, including:

[0007] A fabric unit, including multiple fabric machine monomers that slide along double rails. Each monomer integrates a three-section hydraulic folding arm, a slewing drive assembly, and an embedded controller, and the embedded controller is used to collect data on slump, pumping pressure, boom inclination, and ambient wind speed in real time;

[0008] An edge gateway, deployed inside each monomer, configured with a sliding time window filtering algorithm and a feature extraction module, for compressing sensor data into a standardized construction feature stream containing timestamps;

[0009] A cloud federated learning platform, integrated with a digital twin engine, for receiving the feature streams from each edge gateway, constructing a multi-objective optimization model, and generating a collaborative control strategy containing path planning, hydraulic parameters, and obstacle avoidance instructions;

[0010] A human-machine collaboration terminal, for parsing the collaborative control strategy into boom motion trajectories and valve control parameters, and sending them to the fabric unit to achieve multi-position synchronization and stress balance control during the pouring process.

[0011] Further, the double rails are configured with:

[0012] RFID positioning tag array, with a spacing of less than or equal to 1 meter, is used for the position calibration of the fabric placing machine;

[0013] The anti-collision warning module calculates the safe distance between adjacent fabric placing machine units in real time through UWB ranging;

[0014] Dynamically leveling hydraulic legs are used to automatically adjust the inclination of the support plate according to the optical fiber strain data to ensure balanced support reaction force.

[0015] Furthermore, the three-section hydraulic folding arm comprises:

[0016] The first boom is hinged to the base frame through a pin shaft, with built-in dual redundant balance valve and pressure compensator;

[0017] The second boom is equipped with distributed strain sensors to monitor the dynamic load distribution on the guide rail support surface in real time;

[0018] A microwave slump detector and a MEMS inertial navigation unit are integrated at the end of the third boom; wherein the microwave slump detector is used to measure the rheological properties of concrete based on the principle of microwave resonance, and the MEMS inertial navigation unit is used to solve the position and posture of the boom end in real time.

[0019] Further, the edge gateway executes:

[0020] Based on wavelet packet decomposition, hydraulic shock noise is filtered out and pumping pressure spectrum characteristics are extracted;

[0021] When the slump fluctuation is detected to exceed the set threshold, the local PID controller is triggered to adjust the hydraulic flow;

[0022] An event-driven transmission mode is adopted, and data is uploaded to the cloud-based federated learning platform when the characteristic value change rate is greater than the dynamically adjusted threshold interval; wherein the dynamically adjusted threshold interval is optimized in real time in combination with historical construction data and environmental parameters.

[0023] Furthermore, the event-driven transmission mode includes:

[0024] Based on the spatiotemporal attention mechanism, key features are screened and differential transmission weights are assigned to high-frequency boom vibration data and low-frequency pumping pressure data.

[0025] A lightweight federated filtering algorithm is used to pre-aggregate local data of multiple fabric machine units at the edge gateway, generate compressed gradient tensors, and then upload them to the cloud-based federated learning platform.

[0026] Furthermore, the construction of the digital twin engine includes:

[0027] Discretize the imported BIM model into voxel grids and map them to the fabric placement unit workspace;

[0028] Fuse lidar point cloud data to reconstruct the 3D construction scene and generate a dynamic obstacle topology map;

[0029] Adopt a heterogeneous federated learning framework to aggregate the local gradient updates of multiple concrete placing booms. Among them, differential privacy encryption is used for the hydraulic parameter gradient, and adaptive Laplace noise is injected; homomorphic encryption transmission is used for the path planning gradient, and after decryption in the cloud, the path weight matrix is optimized through Monte Carlo tree search.

[0030] Furthermore, the cloud federated learning platform receives the feature streams from each edge gateway, constructs a multi-objective optimization model, and generates a collaborative control strategy including path planning, hydraulic parameters, and obstacle avoidance instructions, including:

[0031] Define a multi-objective loss function according to the pouring efficiency, energy consumption cost, and structural stress;

[0032] Iteratively update the model parameters of the multi-objective optimization model through an adaptive momentum estimation optimizer;

[0033] Based on the optimized multi-objective optimization model, generate a collaborative control strategy including path planning, hydraulic parameter adjustment, and obstacle avoidance instructions;

[0034] Introduce a hybrid optimization strategy of reinforcement learning and federated learning, use the output of the multi-objective optimization model as a reward signal to feedback to the edge gateway, and dynamically adjust the local model training priority to further optimize the collaborative control strategy.

[0035] Furthermore, the hybrid optimization strategy of reinforcement learning and federated learning specifically includes:

[0036] Construct a virtual construction sandbox to simulate the multi-machine collaborative pouring process and generate adversarial training samples;

[0037] Optimize the path planning action space through the deep deterministic policy gradient algorithm;

[0038] Fuse the virtual training results with the real construction data to generate an enhanced dataset for federated learning.

[0039] Furthermore, the human-machine collaborative terminal includes:

[0040] AR virtual-real registration error compensation algorithm, used to control the virtual path projection deviation by combining MEMS pose data;

[0041] Multi-modal sensor fusion positioning module, used to fuse UWB positioning data, visual SLAM point cloud, and inertial navigation information in real time;

[0042] A dynamic priority arbitration module is used to automatically select an execution instruction based on a security level weight when a gesture instruction conflicts with the collaborative control strategy.

[0043] Further, the AR virtual-real registration error compensation algorithm includes:

[0044] A construction scene understanding module based on semantic segmentation identifies the semantic labels of the pouring area boundary and obstacles;

[0045] Adopt a neural radiance field strategy to render the lighting consistency between the virtual path and the real environment in real time;

[0046] Embed a feedback correction loop to adjust the AR projection parameters in reverse through the pose error at the end of the boom.

[0047] Based on the embodiments provided in this application, multi-dimensional construction parameters such as slump and pumping pressure are collected in real time through an embedded controller, combined with the standardized feature stream processing of the edge gateway, to solve the problem of data islands in traditional systems and improve the integrity and timeliness of parameter collection; the cloud federated learning platform constructs a multi-objective optimization model based on the digital twin engine, dynamically generates a collaborative control strategy, and realizes the global optimal matching of the pouring path and hydraulic parameters, overcoming the lag of manual experience decision-making, and can help effectively cope with concrete slump fluctuations or pumping pressure imbalances; through the AR visualization interface and the multi-modal interaction module, complex cloud strategies are converted into executable boom movement instructions, reducing the operation threshold and improving construction accuracy. Description of the Drawings

[0048] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and the illustrative embodiments and descriptions thereof are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0049] Figure 1 It is a structural diagram of an optional rail-type concrete placing boom system based on cloud factory collaboration according to an embodiment of this application;

[0050] Figure 2 It is a flowchart executed by an optional edge gateway according to an embodiment of this application;

[0051] Figure 3 It is a flowchart for constructing an optional digital twin engine according to an embodiment of this application.

[0052] The realization, functional characteristics, and advantages of the object of the present invention will be further described with reference to the embodiments and the drawings. Detailed Embodiments

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0054] Optionally, as Figure 1 shown, the present application provides an orbital concrete placing system based on cloud factory collaboration, including:

[0055] A concrete placing unit 101, including a plurality of concrete placing machine monomers sliding along a double guide rail. Each monomer integrates a three-section hydraulic folding arm, a slewing drive assembly, and an embedded controller. The embedded controller is used to collect data on slump, pumping pressure, boom inclination, and ambient wind speed in real time;

[0056] Among them, the slewing drive assembly includes: a cycloidal hydraulic motor with a maximum output torque greater than or equal to 149 Newton meters; a harmonic reducer with a transmission accuracy less than or equal to 0.1 arc minute; a slewing limiter that triggers a micro switch through a cam mechanism to limit the slewing angle within ±180°;

[0057] Among them, the embedded controller is connected to the hydraulic proportional valve group through a CAN bus;

[0058] An edge gateway 102, deployed inside each monomer, configured with a sliding time window filtering algorithm and a feature extraction module, for compressing sensor data into a standardized construction feature stream including timestamps;

[0059] A cloud federated learning platform 103, integrating a digital twin engine, for receiving the feature streams of each edge gateway, constructing a multi-objective optimization model, and generating a collaborative control strategy including path planning, hydraulic parameters, and obstacle avoidance instructions;

[0060] A human-machine collaboration terminal 104, for parsing the collaborative control strategy into boom movement trajectories and valve control parameters, and sending them to the concrete placing unit to achieve multi-position synchronization and stress balance control during the pouring process.

[0061] Based on the embodiments provided in this application, multi-dimensional construction parameters such as slump and pumping pressure are collected in real time by an embedded controller, and combined with the standardized feature stream processing of an edge gateway to solve the problem of data islands in traditional systems and improve the integrity and timeliness of parameter collection; the cloud federated learning platform constructs a multi-objective optimization model based on a digital twin engine, dynamically generates a collaborative control strategy, and realizes the global optimal matching of the pouring path and hydraulic parameters, overcoming the lag of manual experience decision-making, and can help effectively cope with concrete slump fluctuations or pumping pressure imbalances; through the AR visualization interface and the multi-modal interaction module, complex cloud strategies are converted into executable boom movement instructions, reducing the operation threshold and improving construction accuracy.

[0062] Further, the dual-rail configuration:

[0063] An RFID positioning tag array with a spacing less than or equal to 1 meter is used for calibrating the absolute position of the individual concrete placing boom.

[0064] An anti-collision warning module calculates the safe distance between adjacent individual concrete placing booms in real time through UWB ranging.

[0065] A dynamic leveling hydraulic outrigger is used to automatically adjust the inclination angle of the support plate according to the fiber optic strain data to ensure balanced reaction forces.

[0066] Based on the embodiments provided in this application, the absolute position calibration of the individual concrete placing boom is realized through the RFID tag array, and the hydraulic outriggers are dynamically leveled by combining the fiber optic strain data to ensure balanced reaction forces during the multi-machine collaborative slip and prevent the risk of overturning caused by foundation settlement.

[0067] Further, the three-section hydraulic folding boom includes:

[0068] The first section of the boom is hinged to the chassis through a pin shaft and is internally provided with a double-redundancy balance valve and a pressure compensator.

[0069] The second section of the boom is equipped with distributed strain sensors for real-time monitoring of the dynamic load distribution on the guide rail support surface.

[0070] The end of the third section of the boom is integrated with a microwave slump detector and a MEMS inertial navigation unit; among them, the microwave slump detector is used to measure the rheological properties of concrete based on the microwave resonance principle, and the MEMS inertial navigation unit is used to real-time calculate the pose of the boom end.

[0071] In this embodiment, the first section of the boom, the second section of the boom, and the third section of the boom are 8 meters, 7 meters, and 6 meters respectively.

[0072] Based on the embodiments provided in this application, the three-section hydraulic folding boom integrates strain sensors and an inertial navigation unit, real-time monitors the flexural deformation and the end pose of the boom, and dynamically adjusts the concrete rheological parameters in combination with the microwave slump detector to improve the consistency of pouring quality under complex working conditions.

[0073] Further, as Figure 2 shown, the edge gateway executes:

[0074] S201, filtering out hydraulic shock noise based on wavelet packet decomposition and extracting the spectral characteristics of the pumping pressure;

[0075] S202, when it is detected that the slump fluctuation exceeds the set threshold, triggering the local PID controller to adjust the hydraulic flow rate;

[0076] Among them, the slump is an important indicator to measure the fluidity of concrete, and its fluctuation range should be determined according to the construction requirements and the performance of the concrete. Generally speaking, there is a design range for the slump of concrete. For example, for pumped concrete, the slump is usually between 100 and 200 mm. When the slump fluctuation exceeds this range, it may affect the construction performance and structural quality of the concrete.

[0077] For example, assuming that the designed slump is 150 mm and the allowable fluctuation range is ±20 mm, then the set threshold can be taken as 130 mm to 170 mm. When it is detected that the slump is lower than 130 mm or higher than 170 mm, the local PID controller is triggered to adjust the hydraulic flow rate.

[0078] S203, adopting an event-driven transmission mode, and uploading data to the cloud federated learning platform when the change rate of the eigenvalue is greater than the dynamically adjusted threshold interval; among them, the dynamically adjusted threshold interval is optimized in real time by combining historical construction data and environmental parameters.

[0079] For example, the dynamically adjusted threshold interval can be 10% or 20%.

[0080] In the embodiment of the present application, the dynamic event threshold model is:

[0081]

[0082] Among them, is the dynamically adjusted threshold interval; is the current timestamp (unit: second); is the sensor type index (e.g., k = 1 is the slump sensor, k = 2 is the pressure sensor, k = 3 is the inclination sensor, k = 4 is the wind speed sensor); is the total number of sensors (for example, in this system K = 4, corresponding to slump, pressure, inclination, and wind speed); is the data change amount of the k-th type of sensor within the time window ; is the sensor weight coefficient; is the index of the environmental parameter; is the number of environmental parameters; is the environmental disturbance sensitivity coefficient (e.g., wind speed , temperature ); is the normalized value of environmental parameters (e.g., when the wind speed is level 6 ); is the hydraulic pressure gain factor (obtained through training historical data, typical value μ = 1.2); is the real-time hydraulic system pressure value; is the critical pressure threshold (set according to the rated parameters of the hydraulic valve); is the hyperbolic tangent function.

[0083] In addition, the hydraulic circuit numbers in the system are used to identify different hydraulic circuits (e.g., if there are two hydraulic circuits in the system, the numbers can be 1 or 2, representing the main circuit and the auxiliary circuit respectively); the pressure changes in these hydraulic circuits indirectly affect the calculation of the dynamic event threshold through sensor data.

[0084] Specifically, the slump is an important indicator to measure the fluidity of concrete and directly affects the pouring quality. When dynamically adjusting the threshold, the slump fluctuation has the greatest impact on the adjustment of construction parameters; the pumping pressure is a key parameter affecting the stability of concrete transportation, and pressure fluctuations may lead to poor pumping or equipment damage; the boom angle has a certain impact on the stability of the placing boom, but its weight is relatively low because the angle change usually does not directly affect the rheological properties of concrete; the wind speed has a certain impact on the operation stability of the placing boom, but the weight is low because the wind speed change usually does not directly affect the construction parameters of concrete.

[0085] Based on this, for the sensor weight coefficient , it can be set as: slump sensor , pressure sensor , inclination sensor , wind speed sensor .

[0086] If it is found during the construction process that the pressure fluctuation has a greater impact on the system, the weight of the pressure sensor can be appropriately increased ( ). If the wind speed has a greater impact on the construction stability, the weight of the wind speed sensor can be appropriately increased (e.g., ).

[0087] Based on the above formula, by integrating the sensor change rate and environmental disturbance, the data upload trigger condition is automatically adjusted to avoid high-frequency ineffective communication; the pressure mutation is smoothed through the hyperbolic tangent function tanh to prevent false triggering caused by hydraulic shock; the online optimization of the weight coefficients and enables the system to adapt to the requirements of different construction stages.

[0088] Based on the embodiments provided in this application, wavelet packet decomposition is used to filter out hydraulic noise, and the event-driven transmission mode is combined to dynamically adjust the data upload threshold, reducing the cloud communication load while retaining key construction features, and realizing the efficient cooperation between local PID control and cloud decision-making.

[0089] Further, the event-driven transmission mode includes:

[0090] Screen key features based on spatio-temporal attention mechanism, and assign different transmission weights to high-frequency changing boom vibration data and low-frequency pumping pressure data;

[0091] Adopt a lightweight federated filtering algorithm to pre-aggregate the local data of multiple concrete placer monomers at the edge gateway, generate a compressed gradient tensor and then upload it to the cloud federated learning platform.

[0092] Based on the embodiments provided in this application, a heterogeneous federated learning framework is used to differentially encrypt hydraulic parameters and path planning gradients, and the Monte Carlo tree search is combined to optimize the path weight matrix, realizing the dynamic reconstruction of multi-machine obstacle avoidance paths while protecting data privacy.

[0093] Further, as Figure 3 shown, the construction of the digital twin engine includes:

[0094] S301, discretize the imported BIM model into voxel grids and map it to the working space of the concrete placer unit;

[0095] S302, fuse lidar point cloud data to reconstruct the three-dimensional construction scene and generate a dynamic obstacle topology map;

[0096] S303, adopt a heterogeneous federated learning framework to aggregate the local gradient updates of multiple concrete placer monomers. Among them, differential privacy encryption is used for the hydraulic parameter gradient, and adaptive Laplace noise is injected; homomorphic encryption transmission is used for the path planning gradient, and after decryption by the cloud, the path weight matrix is optimized through Monte Carlo tree search.

[0097] In the embodiments of this application, heterogeneous federated gradient aggregation can be realized based on the following formula:

[0098]

[0099] where, is the federated gradient after global aggregation; is the concrete placer monomer number (such as j = 1, 2,..., J); is the hydraulic parameter gradient of the j-th monomer (including pumping pressure, flow regulation amount); is the path planning gradient of the j-th monomer (boom trajectory curvature, obstacle avoidance priority); It is a differential privacy encryption function (injecting Laplace noise L(0,β), β = 0.1); It is a homomorphic encryption function (supporting ciphertext addition operation); It is the total number of cloth laying machines in the system (for example, J = 5); It is the Hadamard product (element-wise multiplication, used for mask filtering); It is the encrypted weight of the hydraulic gradient (set according to the monomer reliability, such as ); It is the encrypted weight of the path gradient (such as ); It is the abnormal node mask matrix (the corresponding position is 0 when the communication delay > 200ms);

[0100] Based on the above formula, a differential encryption strategy is adopted for the hydraulic parameters and the path planning gradient to prevent the leakage of sensitive data; the abnormal nodes are filtered through the mask matrix to improve the robustness of federated learning; homomorphic encryption supports ciphertext operations, reduces the decryption overhead of the cloud, and accelerates the update of the global model.

[0101] Furthermore, the cloud federated learning platform receives the feature streams of each edge gateway, constructs a multi-objective optimization model, and generates a collaborative control strategy including path planning, hydraulic parameters, and obstacle avoidance instructions, including:

[0102] Define a multi-objective loss function according to the pouring efficiency, energy consumption cost, and structural stress;

[0103] Iteratively update the model parameters of the multi-objective optimization model through the adaptive momentum estimation optimizer (AdamW);

[0104] Based on the optimized multi-objective optimization model, generate a collaborative control strategy including path planning, hydraulic parameter adjustment, and obstacle avoidance instructions;

[0105] Introduce a hybrid optimization strategy of reinforcement learning and federated learning, and use the output of the multi-objective optimization model as a reward signal to feedback to the edge gateway, dynamically adjusting the local model training priority to further optimize the collaborative control strategy.

[0106] Furthermore, the hybrid optimization strategy of reinforcement learning and federated learning specifically includes:

[0107] Construct a virtual construction sandbox to simulate the multi-machine collaborative pouring process and generate adversarial training samples;

[0108] Optimize the path planning action space through the deep deterministic policy gradient algorithm;

[0109] Fuse the virtual training results with the real construction data to generate an enhanced dataset for federated learning.

[0110] In the embodiment of this application, the multi-objective hybrid reward function is:

[0111]

[0112] wherein, is the mixed reward value; is the decision time step (different from the time stamp in Formula 1, here it is a discretized step); is the federated learning reward (used to measure the pouring uniformity and energy consumption efficiency); is the federated reward weight (at the initial stage of pouring , and at the later stage ); is the reinforcement learning reward (used to measure the obstacle avoidance success rate and structural stress balance); is the weight ( is 0.3 at the initial stage of pouring and is adjusted to 0.6 at the later stage); is the decision delay time (unit: second); is the total number of decision steps (for example, for a single pouring task T = 100 steps); is the system state (including boom pose, concrete state, environmental parameters); is the action (such as the hydraulic valve opening degree, boom rotation angle); is the time decay coefficient (typical value ).

[0113] Based on the above formula, the reward mechanism combining federated learning and reinforcement learning balances construction efficiency and safety; adjusts the value according to the construction stage (such as emphasizing efficiency at the initial stage of pouring and stress balance at the later stage); the exponential decay term reduces the weight of high-delay decisions and improves the real-time performance of the system.

[0114] Based on the embodiments provided in this application, adversarial training samples are generated through a virtual construction sandbox, and the action space is optimized by combining the deep deterministic policy gradient algorithm to improve the decision-making ability of the model in complex scenarios.

[0115] Furthermore, the human-machine collaboration terminal includes:

[0116] The AR virtual-real registration error compensation algorithm is used to control the virtual path projection deviation by combining MEMS pose data;

[0117] The multi-modal sensor fusion positioning module is used to fuse UWB positioning data, visual SLAM point cloud and inertial navigation information in real time;

[0118] The dynamic priority arbitration module is used to automatically select the execution instruction based on the safety level weight when the gesture instruction conflicts with the collaborative control strategy.

[0119] Furthermore, the AR virtual-real registration error compensation algorithm includes:

[0120] A construction scene understanding module based on semantic segmentation to identify the semantic labels of the pouring area boundary and obstacles;

[0121] Adopt the neural radiance field strategy to render the lighting consistency between the virtual path and the real environment in real time;

[0122] Embed a feedback correction loop to inversely adjust the AR projection parameters through the pose error at the end of the boom.

[0123] In the embodiment of the present application, the AR virtual-real projection error compensation is calculated based on the following formula:

[0124]

[0125] Wherein, is the comprehensive projection error (to be minimized); is the virtual projection pose of the th modality (visual SLAM, UWB, inertial navigation); is the measured value of the real pose corresponding to the modality; is the modality fusion weight (e.g., visual weight , UWB weight ); is the KL divergence (measuring the difference in positioning confidence between SLAM and UWB); is the lighting parameter of the neural radiance field rendering; is the real environment lighting intensity; is the sensor modality index (e.g., c = 1 for visual SLAM, c = 2 for UWB); is the total number of modalities (in this system C = 3, including SLAM, UWB, inertial navigation); is the positioning confidence distribution of visual SLAM; is the confidence distribution of UWB positioning; represents the L2 norm; represents the L1 norm.

[0126] Based on the embodiments provided in the present application, by fusing multi-modal sensor positioning data and neural radiance field rendering technology, the AR projection deviation is controlled within millimeters. Combining semantic segmentation and a feedback correction loop ensures the lighting and geometric consistency between the virtual path and the real environment, and improves the efficiency of human-machine collaboration.

[0127] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. An orbital fabric distribution system based on cloud factory collaboration, characterized in that, Including: A fabricating unit, including multiple fabricating machine monomers that slide along double guide rails. Each monomer integrates a three-section hydraulic folding arm, a rotary drive assembly, and an embedded controller. The embedded controller is used to collect data on slump, pumping pressure, boom inclination, and ambient wind speed in real time; An edge gateway, deployed inside each monomer, configured with a sliding time window filtering algorithm and a feature extraction module, for compressing sensor data into a standardized construction feature stream containing timestamps; A cloud federated learning platform, integrating a digital twin engine, for receiving the feature streams from each edge gateway, constructing a multi-objective optimization model, and generating a collaborative control strategy containing path planning, hydraulic parameters, and obstacle avoidance instructions; A human-machine collaborative terminal, for parsing the collaborative control strategy into boom movement trajectories and valve control parameters, and sending them to the fabricating unit to achieve multi-position synchronization and stress balance control during the pouring process; The edge gateway performs: Filtering hydraulic shock noise based on wavelet packet decomposition and extracting the pumping pressure spectrum features; When it detects that the slump fluctuation exceeds the set threshold, triggering a local PID controller to adjust the hydraulic flow rate; Adopting an event-driven transmission mode, and uploading data to the cloud federated learning platform when the feature value change rate is greater than a dynamically adjusted threshold interval; wherein, the dynamically adjusted threshold interval is optimized in real time by combining historical construction data and environmental parameters.

2. The orbital cloth spreading system based on cloud factory collaboration according to claim 1, wherein The double guide rails are configured with: An RFID positioning tag array with a spacing less than or equal to 1 meter, for calibrating the positions of the fabricating machine monomers; An anti-collision warning module, for calculating the safety distance between adjacent fabricating machine monomers in real time through UWB ranging; Dynamically leveling hydraulic outriggers, for automatically adjusting the inclination angle of the support plate according to fiber optic strain data to ensure balanced reaction forces.

3. The orbital fabric distribution system based on cloud factory collaboration according to claim 1, characterized in that, The three-section hydraulic folding arm includes: The first section of the arm is hinged to the chassis through a pin shaft, and is internally equipped with a double-redundancy balance valve and a pressure compensator; The second section of the arm is equipped with distributed strain sensors, for real-time monitoring of the dynamic load distribution on the guide rail support surface; The end of the third section of the arm integrates a microwave slump detector and a MEMS inertial navigation unit; wherein, the microwave slump detector is used to measure the rheological properties of concrete based on the microwave resonance principle, and the MEMS inertial navigation unit is used to solve the pose of the end of the boom in real time.

4. The orbital cloth spreading system based on cloud factory collaboration according to claim 1, characterized in that, The event-driven transmission mode includes: Screening key features based on a spatio-temporal attention mechanism, and assigning different transmission weights to high-frequency changing boom vibration data and low-frequency pumping pressure data; Adopting a lightweight federated filtering algorithm, pre-aggregating the local data of multiple fabricating machine monomers at the edge gateway end, generating a compressed gradient tensor, and then uploading it to the cloud federated learning platform.

5. The orbital fabric distribution system based on cloud factory collaboration according to claim 1, characterized in that The construction of the digital twin engine includes: Discretizing the imported BIM model into a voxel grid and mapping it to the working space of the fabricating unit; Fusing lidar point cloud data to reconstruct a three-dimensional construction scene and generating a dynamic obstacle topology map; Adopting a heterogeneous federated learning framework to aggregate the local gradient updates of multiple fabricating machine monomers. Among them, differential privacy encryption is used for the hydraulic parameter gradient, and adaptive Laplace noise is injected; homomorphic encryption transmission is used for the path planning gradient, and after decryption in the cloud, the path weight matrix is optimized through Monte Carlo tree search.

6. The orbital cloth spreading system based on cloud factory collaboration according to claim 5, characterized in that, The cloud-based federated learning platform receives the feature streams of each edge gateway, constructs a multi-objective optimization model, and generates a collaborative control strategy including path planning, hydraulic parameters, and obstacle avoidance instructions, including: Define a multi-objective loss function based on pouring efficiency, energy consumption cost, and structural stress; Iteratively update the model parameters of the multi-objective optimization model through an adaptive momentum estimation optimizer; Based on the optimized multi-objective optimization model, generate a collaborative control strategy including path planning, hydraulic parameter adjustment, and obstacle avoidance instructions; Introduce a hybrid optimization strategy of reinforcement learning and federated learning, feedback the output of the multi-objective optimization model as a reward signal to the edge gateway, and dynamically adjust the local model training priority to further optimize the collaborative control strategy.

7. The orbital cloth spreading system based on cloud factory collaboration according to claim 6, wherein, The hybrid optimization strategy of reinforcement learning and federated learning specifically includes: Construct a virtual construction sandbox to simulate the multi-machine collaborative pouring process and generate adversarial training samples; Optimize the path planning action space through the deep deterministic policy gradient algorithm; Fuse the virtual training results with real construction data to generate an enhanced dataset for federated learning.

8. The orbital cloth spreading system based on cloud factory collaboration according to claim 1, characterized in that, The human-machine collaborative terminal includes: AR virtual-real registration error compensation algorithm for controlling the virtual path projection deviation by combining MEMS pose data; Multi-modal sensor fusion positioning module for real-time fusion of UWB positioning data, visual SLAM point cloud, and inertial navigation information; Dynamic priority arbitration module for automatically selecting an execution instruction based on the safety level weight when a gesture instruction conflicts with the collaborative control strategy.

9. The orbital cloth spreading system based on cloud factory collaboration according to claim 8, characterized in that The AR virtual-real registration error compensation algorithm includes: A construction scene understanding module based on semantic segmentation to identify the pouring area boundary and obstacle semantic labels; Adopt a neural radiance field strategy to render the lighting consistency between the virtual path and the real environment in real time; Embed a feedback correction loop to reverse-adjust the AR projection parameters through the end-effector pose error.

Images