Heterogeneous construction equipment fleet control system and method based on vla model

Abstract

This invention relates to a heterogeneous construction equipment cluster control system and method based on a VLA model, belonging to the field of building construction technology. The system includes a cluster multimodal perception unit, an elevation benchmark fusion unit, an equipment status modeling unit, a multimodal decision control unit, an instruction allocation and safety verification unit, a low-level execution control unit, and a human-machine interaction and remote monitoring platform. By introducing innovative technical solutions such as visual language action models, multi-source data fusion, hybrid control, and safety verification, this invention achieves high-precision synchronous jacking and stable control of the core tube and mega-column formwork cluster, and high-precision collaborative control of the platform's flatness and vertical attitude. This improves the intelligence level and stability of the heterogeneous cluster synchronous control, reduces the relative deformation and stress concentration risks of the equipment cluster connection structure, provides a more efficient and safer collaborative control method for super high-rise construction, and ensures the smooth implementation of the project.

Description

Technical Field

[0001] This invention belongs to the field of building construction technology, specifically relating to a heterogeneous construction equipment cluster control system and method based on VLA model. Background Technology

[0002] In current super high-rise building construction, the synchronous jacking and climbing coordination of core tube and mega-column construction equipment mostly adopts a hierarchical control architecture of distributed PLC + host computer. Basic synchronization is achieved by combining fixed rules such as equal step distance, equal pressure difference, or speed-limiting interlocks. The sensing side mainly relies on single or limited modal data such as tilt angle, stroke, and pressure. Cross-platform exchanges only involve a small number of state and interlocking signals. There is a lack of unified dynamic modeling and characteristic representation of the differences between the core tube upper-mounted lifting system and the mega-column lower-mounted climbing system in terms of power source, force transmission path, stiffness, and response time. This fails to form a multi-source fusion state vector that can be used for global decision-making. Regarding control objectives, existing solutions only provide coarse-grained constraints on the flatness (vertical height difference) between platforms, lacking explicit suppression and coupled control of the relative horizontal displacement and torsion of the corridor bridge interface. This leads to instability symptoms such as accumulated height difference, lateral swaying, and torsional amplification during the jacking process. Regarding safety mechanisms, existing solutions largely rely on fixed thresholds and rigid interlocks, making it difficult to proactively assess and optimize future structural risks in action sequences. They lack a unified, real-time measurable comprehensive safety risk indicator and have insufficient instruction allocation strategies, hindering the implementation of detailed constraints and limit-breaking prevention for different control logics of heterogeneous execution units. In terms of intelligence and adaptability, parameter tuning in existing solutions primarily relies on manual experience, lacking scenario migration capabilities and struggling to adapt to structural phase changes, load disturbances, and equipment aging. Regarding real-time performance, centralized scheduling and low-frequency cycles in existing solutions easily cause response lag, failing to meet the requirements of small step sizes and rapid closed-loop operation in hydraulic systems.

[0003] Therefore, existing technologies struggle to balance planarity control and vertical attitude stability in the synchronous lifting scenario of heterogeneous, strongly coupled construction equipment clusters. A more intelligent and efficient control system, device, and method for heterogeneous construction equipment clusters are needed to improve the informatization and intelligence level of on-site safety management. Summary of the Invention

[0004] This invention provides a heterogeneous construction equipment cluster control system and method based on the VLA model to solve the problems existing in the synchronous lifting scenario of construction equipment clusters in the construction of super high-rise buildings.

[0005] To solve the above technical problems, the present invention includes the following technical solutions: A cluster control system for heterogeneous construction equipment based on a VLA model, used for the synchronous jacking and climbing coordination of construction equipment for building core tubes and mega-columns, includes: The cluster multimodal sensing unit is used to measure the physical state of the equipment cluster in real time. The cluster multimodal sensing unit includes: an attitude sensing network deployed on the core tube platform and each mega-column formwork measuring point, stroke and pressure sensors installed on all hydraulic actuators, laser rangefinders deployed at each connecting corridor bridge interface, and a distributed independent static leveling measurement network covering the equipment cluster. The elevation datum fusion unit is used to process data from independent hydrostatic leveling networks and achieve a unified elevation datum. The equipment status modeling unit receives multi-source heterogeneous data streams from the sensing unit and unified elevation data processed by the elevation benchmark fusion unit, and models the heterogeneous construction equipment cluster control system. The multimodal decision control unit implements model inference through the VLA inference module. The VLA inference module integrates a hierarchical intent decomposition mechanism and a structural risk indicator perception and attention mechanism to generate a global collaborative action strategy based on structural risk indicators. ; The instruction allocation and security verification unit is located between the multimodal decision control unit and the underlying execution control unit. It receives the global collaborative action strategy, executes the instruction allocation operation, decomposes the unified action instruction into sub-instruction sets that conform to the control logic, and performs real-time verification on these sub-instruction sets. The underlying execution control unit is used to convert the instructions in the sub-instruction set into underlying electronic control signals that drive the actions of different mechanisms in various equipment. The human-computer interaction and remote monitoring platform provides operators with a graphical interface for inputting task commands and displaying the status of equipment clusters.

[0006] A cluster control method for heterogeneous construction equipment based on the VLA model is used for the synchronous jacking and climbing coordination of construction equipment for building core tubes and mega-columns, comprising the following steps: S1. Real-time acquisition of multi-source data, including the attitude of equipment clusters, the stroke and pressure of hydraulic actuators, and the relative displacement vector of the corridor bridge; acquisition of level instrument readings from each platform, and conversion of the measurements from each independent system into a unified elevation vector. Based on the collected multi-source data and elevation vectors This forms the global state vector. ; S2. The system decomposes and parses the input high-level natural language task instructions, breaking down the macro-task into a series of internal sub-cues. It then integrates the preset safety constraints—such as bridge stress, platform height differences, and cylinder pressure—with the current stage's sub-cues to form task instruction prompts. ; The S3.VLA model is based on a global state vector. and task instruction prompts Generate a global collaborative action strategy based on structural risk indicators ; S4. Based on the global collaborative action strategy The system allocates instructions and performs safety checks on the decomposed sub-instructions. Instructions that pass the checks are then sent to the hydraulic actuators for execution. S5. The hydraulic actuator performs actions according to the received instructions.

[0007] Furthermore, in step S1, the global state vector The equipment state modeling unit is used to create and output the data; the equipment state modeling unit is based on the collected multi-source data and elevation vectors. Performs modeling / mapping of the motion and force characteristics of heterogeneous devices, and outputs a global state vector. Global state vector This can be formally represented as: ; in, This represents the total number of subsystems in the cluster, including one core and... A giant pillar It is the first Real-time sensor data vectors of each subsystem It is a representation of the first The static embedding vector of the dynamic characteristics of each subsystem It is the first Elevation values ​​of each subsystem based on a unified benchmark This indicates a vector concatenation operation.

[0008] Furthermore, in step S2, the elevation difference constraint between platforms based on a unified elevation datum can be expressed as: ; in, The maximum allowable height difference threshold, This represents the total number of subsystems in the cluster, including one core and... A giant pillar , They are the first , Elevation values ​​of each subsystem based on a unified benchmark.

[0009] Furthermore, in step S3, a structural risk index function is first defined. This is used to quantify the risk of cluster instability in real time. , in: Calculate the standard deviation of the elevations of all platforms based on a unified benchmark to characterize the overall flatness deviation of the cluster; It is the relative displacement vector of all the covered bridges. The L2 norm is used to characterize the degree of horizontal swaying or torsion of the cluster as a whole; in the third term... For the core tube elevation, For the first The elevation of the giant column platform is used to accumulate and quantify the stress concentration risk at the connection point. , and These are the weighting coefficients; The objective function of the VLA model for: , in, Loss due to motion imitation; objective function Drive the VLA model to generate the current action At that time, predictive reasoning is performed to select a value that will make the next time step... Structural risk indicators The action with the lowest expected value.

[0010] Furthermore, step S3 is executed by the multimodal decision control unit; The multimodal decision control unit integrates an online adaptive learning module, which collects execution data at the end of each control cycle. When the accumulated data reaches a preset threshold, the incremental learning process of the model is triggered, which calculates the actual structural risk indicators. The deviation from the expected value is used to update the model parameters, and the attention layer of the VLA model is fine-tuned so that the model can learn the dynamic characteristics of a specific building structure and equipment combination.

[0011] Furthermore, step S3 is executed by the multimodal decision control unit; Within the multimodal decision control unit, a VLA policy caching and fast matching mechanism are implemented. This is achieved by constructing a high-dimensional index database based on vector similarity, which stores historical decision pairs. Persistent storage is performed; when a new state vector... Upon input, first calculate its cosine similarity to existing states in the database. If the similarity exceeds a certain threshold and the task instructions... If a match is found, the corresponding historical action strategy will be invoked directly. It skips the model inference process, reduces system response time, and meets the real-time requirements of rapid response in hydraulic systems.

[0012] Furthermore, in step S1, the measured values ​​of each independent system are converted into a unified elevation vector. Performed by the elevation datum fusion unit; A multi-point weighted fusion mechanism is set up in the elevation datum fusion unit; for the core tube and the... The reference transfer between the giant columns is set at the connection of the corridor bridge. The reference transfer measurement point pair; let the first... The core tube side readings for each measuring point pair are: The reading on the side of the giant column is Then the first on the giant pillar platform The fused elevation of each measuring point is calculated as follows: , Weight Dynamically determined through stability analysis of historical data at measurement points: ; in, For the first Each measurement point in the past Difference in readings within each sampling period The variance; when the variance of a pair of measurement points exceeds the threshold. When this happens, its weight will be automatically reduced or it will be removed, triggering a maintenance alarm. As the global elevation benchmark, This is the reading of the reference point in the core tube system.

[0013] Furthermore, in step S4, the verification command is issued through the underlying execution control unit; A hybrid VLA and PID control architecture is built within the underlying execution control unit. An independent PID control loop is embedded in the underlying controller of each hydraulic actuator. The decomposed sub-commands are converted into dynamic setpoints for the PID controllers of each actuator. The PID controllers adjust the setpoints based on the deviations from the real-time sensor feedback. Through the proportional term Integral terms and differential terms The linear combination of these signals generates the actual control signal, achieving the fusion of VLA global planning and PID local control; among which, For the purpose of global coordinated action, For actual implementation feedback, , , These are the PID proportional control parameters, integral control parameters, and derivative control parameters, respectively.

[0014] Furthermore, between steps S3 and S4, a digital twin simulation and security verification step is added, which is executed by the digital twin simulation and security verification module. The digital twin simulation and security verification module maintains a digital twin model synchronized with the construction equipment cluster, when generating a global collaborative action strategy. Then, a virtual rehearsal of multiple future control cycles is first conducted in an accelerated mode in a digital twin environment. The stress distribution and deformation at the bridge connection are calculated through finite element analysis. If the predicted stress or relative deformation exceeds the allowable value, the dangerous state characteristics in the rehearsal results are fed back to the multimodal decision control unit, triggering the model to regenerate the global collaborative action strategy based on safety constraints, thus forming a safety assurance mechanism of rehearsal-evaluation-replanning.

[0015] Compared with existing technologies, this invention, by adopting the above technical solutions, has the following advantages and positive effects: This invention provides a heterogeneous construction equipment cluster control system and method based on a VLA model. It acquires platform attitude, stroke, pressure, relative displacement, and other state information through multi-source sensing and visual perception. Combined with a VLA multimodal model, it performs heterogeneous dynamics identification and global state fusion. It performs semantic parsing and sub-task decomposition of natural language task instructions, and optimizes and verifies action safety based on comprehensive structural risk indicators. This achieves high-precision synchronous lifting and stable control of the core tube and mega-column formwork cluster, and high-precision collaborative control of platform flatness and vertical attitude. This improves the intelligence level and stability of heterogeneous cluster synchronous control, reduces the relative deformation and stress concentration risks of the equipment cluster connection structure, and provides a more efficient and safer collaborative control method for super high-rise construction, ensuring the smooth implementation of the project. Attached Figure Description

[0016] Figure 1 This is an overall architecture diagram of a heterogeneous manufacturing equipment cluster control system based on a VLA model according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the distributed hydrostatic leveling network layout in one embodiment of the present invention; Figure 3 This is a flowchart of a heterogeneous construction equipment cluster control method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a hybrid VLA and PID control architecture and a safety verification mechanism in one embodiment of the present invention. Detailed Implementation

[0017] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, further illustrates the heterogeneous manufacturing equipment cluster control system and method based on the VLA model provided by the present invention. The advantages and features of the present invention will become clearer with the following description. It should be noted that the accompanying drawings are all in a very simplified form and use non-precise scales, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0018] Example 1 This embodiment provides a heterogeneous construction equipment cluster control system based on a VLA model for the synchronous lifting and climbing coordination of construction equipment for building core tubes and mega-columns. The application background of this embodiment is a super high-rise building construction scenario. During the construction of a super high-rise building, there is a core tube in the middle, surrounded by several mega-columns. As an example, the equipment cluster includes a core tube integral steel platform using an upper-mounted hydraulic steel column lifting method, and multiple mega-column formworks using a lower-mounted hydraulic climbing method. The core tube integral steel platform and the mega-column formworks are connected as a whole by a corridor bridge. Because the constituent units of this equipment cluster differ in their power sources, force transmission paths, and motion mechanisms, they exhibit heterogeneity and strong coupling, presenting technical challenges in attitude coordination control during synchronous lifting. Specifically, two issues need to be addressed simultaneously: first, maintaining the overall flatness of the cluster, i.e., controlling the vertical height difference between each platform unit within a preset tolerance range; second, ensuring the vertical attitude stability of the cluster, i.e., suppressing the relative horizontal displacement and torsion of each platform unit to prevent system motion interference caused by excessive relative deformation at the corridor bridge connection.

[0019] This embodiment uses the VLA model, which stands for Vision-Language-Action. The VLA model is based on an end-to-end decision-making framework of a multimodal large model, and realizes environmental interaction by integrating visual perception, language understanding and action control.

[0020] like Figure 1 As shown, the heterogeneous construction equipment cluster control system based on visual language action model provided in this embodiment includes a cluster multimodal perception unit, an elevation benchmark fusion unit, an equipment status modeling unit, a multimodal decision control unit, an instruction allocation and security verification unit, a low-level execution control unit, and a human-machine interaction and remote monitoring platform.

[0021] The cluster multimodal sensing unit is used to measure the physical state of the equipment cluster in real time. The cluster multimodal sensing unit includes: an attitude sensing network (such as multi-axis tilt sensors) deployed on the core tube platform and each mega-column formwork measuring point; stroke and pressure sensors installed on all hydraulic actuators (including the upper lifting mechanism of the core tube and the lower climbing cylinders of the mega-columns); and relative displacement measurement components (such as 3D laser measurement modules, visual target measurement systems, or orthogonally arranged laser rangefinder sensor groups) deployed at each connecting bridge interface. In terms of hardware implementation, this unit is implemented through a data acquisition interface module, configured with analog and digital input interfaces. The attitude of the equipment cluster is denoted as... ;all The stroke and pressure of each hydraulic actuator are denoted as follows: , , , ;pass The relative displacement vector measured by the sensors on the bridge is , .

[0022] In particular, the clustered multimodal sensing unit also includes a distributed, independent hydrostatic leveling network. For example... Figure 2 As shown, the core tube platform and each mega-column platform are equipped with independent static leveling instrument systems. Each system forms a closed loop through connecting pipes to prevent the entire measurement network from failing due to a single point of failure. To achieve a unified elevation benchmark among the heterogeneous platforms, benchmark transfer measurement point pairs are set up at the locations where the core tube and each mega-column are connected by corridors: a measurement point is led from the static leveling instrument system of the mega-column platform to the core tube side through the internal pipes of the corridor, forming a spatially overlapping measurement point pair with the measurement point of the static leveling instrument system of the core tube at that location. The elevation transfer relationship between the different independent systems is established through this physical overlap configuration.

[0023] The elevation datum fusion unit is used to process data from independent hydrostatic leveling networks and achieve a unified elevation datum. Specifically, it selects a measuring point from the core tube hydrostatic leveling system as the global elevation datum point. The elevations of other measuring points within the core tube system are obtained directly through the system's internal connectivity. For the first... The elevation of the internal measuring points of the mega-pillar platform is calculated as follows: First, the data of the measuring point pair at the connection between the mega-pillar and the core tube is read. Let the reading of the measuring point on the core tube side be... The current reading at the side measuring point of the giant column is Based on the incremental changes between the initial state and the current state, the following absolute elevation conversion relationship is established: ; in For the first The first giant pillar platform The absolute elevation of each measuring point This is the reading at this measuring point in the independent system of giant columns. This refers to the readings of the reference point within the core tube system. This method of integrating distributed independent measurement with reference transfer ensures both system redundancy and reliability, while also achieving globally unified elevation measurement.

[0024] The equipment state modeling unit is used to model heterogeneous systems. Its function is to receive multi-source heterogeneous data streams from the sensing unit and unified elevation data processed by the elevation benchmark fusion unit. The algorithm associates real-time data from different types of subsystems (such as the core tube top-mounted lifting system and the mega-column bottom-mounted climbing system) with prior model labels characterizing their unique dynamic properties (such as response time, stiffness characteristics, and load sensitivity). Through this process, a structured multimodal global state vector is generated, containing the physical properties of each subsystem and unified elevation information. This unit is implemented in the device through a processor module and a memory module, where the memory is used to cache real-time data and intermediate calculation results.

[0025] The multimodal decision control unit, serving as the system's decision-making center, is implemented within the device using a dedicated VLA inference module, employing hardware acceleration to optimize model inference performance. Its internally deployed large VLA model integrates two mechanisms: (1) Hierarchical intent decomposition mechanism: This mechanism can parse and decompose the high-level, goal-oriented natural language instructions input by the operator into a series of time-sequential sub-task sequences with clear physical meaning, and generate corresponding internal control prompts for each sub-task; (2) Structural risk indicator perception and attention mechanism: In the model processing of global state vector At that time, the mechanism is based on a structural risk index function calculated in real time. The output value dynamically adjusts the weight distribution of its internal attention network, enabling it to focus on key state features that characterize cluster stability during decision-making.

[0026] The instruction allocation and safety verification unit is located between the multimodal decision control unit and the underlying execution control unit. Its function is to receive the global coordinated action strategy, first performing an instruction allocation operation, decomposing the unified action instruction into sub-instruction sets, each conforming to its own control logic, specifically for the core tube upper-mounted lifting system and the giant column lower-mounted climbing system. Subsequently, these sub-instruction sets undergo real-time verification through a rule-based safety check to ensure that no instruction instantaneously exceeds the preset hard safety thresholds for speed, stroke, pressure, etc., of a single execution unit. In the device implementation, this unit includes an independent safety monitoring module that can directly cut off the control signal when an over-limit situation is detected.

[0027] The underlying execution control unit converts the decoupled instructions, verified by the safety verification unit, into underlying electrical control signals that drive the actions of different mechanisms in various equipment. This unit has a built-in protocol converter for communication with different types of controllers (such as the main controller of the core tube lifting system and the hydraulic PLCs of each mega-column formwork). At the device level, this is achieved through communication interface modules, including industrial fieldbus interfaces and Ethernet interfaces.

[0028] The human-machine interface and remote monitoring platform provides operators with a graphical interface for inputting advanced task commands and presents the cluster status processed by the equipment status modeling unit and the current intent of the multimodal decision control unit in real time using a 3D visualization. The platform enables remote access via the device's wireless communication interface.

[0029] Example 2 To address the problems existing in the synchronous jacking and climbing coordination of core tube construction equipment and mega-column construction equipment in the construction of super high-rise buildings, this embodiment provides a cluster control method for heterogeneous construction equipment based on a VLA model. The following section combines... Figures 1 to 4 The control method is further described below. The control method includes the following steps: S1. Real-time acquisition of multi-source data, including the attitude of equipment clusters, the stroke and pressure of hydraulic actuators, and the relative displacement vector of the corridor bridge; acquisition of level instrument readings from each platform, and conversion of the measurement values ​​from each independent system to a unified global benchmark to obtain a unified elevation vector. Based on the collected multi-source data and elevation vectors This forms the global state vector. .

[0030] The equipment cluster includes core tube construction equipment and mega-pillar construction equipment. Real-time acquisition of multi-source data is achieved through the cluster's multimodal sensing units. The attitude of the equipment cluster is denoted as... ;all The stroke and pressure of each hydraulic actuator are denoted as follows: , , , ;pass The relative displacement vector measured by the sensors on the bridge is , .

[0031] The acquisition of level instrument readings from each platform is performed through a distributed, independent hydrostatic leveling network. The elevation datum fusion unit, based on the coincidence relationship of measurement point pairs, unifies the measurement values ​​from each independent system to a global datum, resulting in a unified elevation vector. ,in This is the transpose symbol, indicating that a horizontal row vector written in writing should be treated as a vertical column vector in mathematical calculations to conform to matrix operation rules. The conversion relationship and the unified elevation vector represent a "process and result" relationship: the system uses the reading differences of overlapping measurement points in each independent network to calculate the conversion relationship (i.e., the mathematical model for aligning the benchmark), and then uses this relationship to convert and stitch together the scattered and independent local measurement values ​​from each platform in real time, ultimately outputting a unified elevation vector that can be compared under the same global benchmark. .

[0032] Global state vector The equipment status modeling unit is used to create and output data. This unit is based on collected multi-source data and elevation vectors. Performs modeling / mapping of the motion and force characteristics of heterogeneous devices, and outputs a global state vector. Global state vector This can be formally represented as: ; in, This represents the total number of subsystems in the cluster, including one core and... A giant pillar It is the first Real-time sensor data vectors of each subsystem It is a representation of the first The static embedding vector of the dynamic characteristics of each subsystem It is the first Elevation values ​​of each subsystem based on a unified benchmark This represents the vector concatenation operation. This step enables the VLA model to distinguish the intrinsic properties of different components at the perceptual level and obtain an accurate global elevation distribution.

[0033] S2. The system decomposes and parses the input high-level natural language task instructions, breaking down the macro-task into a series of internal sub-cues. It then integrates the preset safety constraints—such as bridge stress, platform height differences, and cylinder pressure—with the current stage's sub-cues to form task instruction prompts. .

[0034] The operator inputs advanced natural language task instructions through a human-computer interaction platform. Upon receiving the instructions, the multimodal decision control unit initiates the task decomposition and instruction parsing steps, breaking down the macro-task into a series of internal sub-cues. Simultaneously, the system integrates preset safety constraints such as bridge stress, platform height differences, and cylinder pressure with the current stage's sub-cues to form task instruction prompts input to the model's decision layer. In this process, the elevation difference constraint between platforms based on a unified elevation datum can be expressed as: ; in, The maximum allowable height difference threshold, This represents the total number of subsystems in the cluster, including one core and... A giant pillar , They are the first , Elevation values ​​of each subsystem based on a unified benchmark.

[0035] S3. Generate a global collaborative action strategy based on structural risk indicators. .

[0036] This step is executed by the multimodal decision control unit. Within the multimodal decision control unit, a global cooperative action strategy is implemented. The goal of this algorithm is to minimize a composite objective function that includes action imitation loss and an expected future risk indicator. To this end, a structural risk indicator function is defined. This is used to quantify the risk of cluster instability in real time. , in: Calculate the standard deviation of the elevations of all platforms based on a unified benchmark to characterize the overall flatness deviation of the cluster; It is the relative displacement vector of all the covered bridges. The L2 norm is used to characterize the degree of horizontal swaying or torsion of the cluster as a whole; in the third term... For the core tube elevation, For the first The elevation of the giant column platform is used to accumulate and quantify the stress concentration risk at the connection point. , and These are the weighting coefficients. The objective function of the VLA model. for: , in, Loss due to motion imitation.

[0037] This objective function drives the VLA model to generate the current action. At that time, predictive reasoning is performed to select a value that will make the next time step... Structural risk indicators The action with the lowest expected value. This mechanism enables the model to generate action sequences that suppress platform elevation differences and prevent lateral instability.

[0038] S4. Based on the global collaborative action strategy The system allocates instructions and performs safety checks on the decomposed sub-instructions. Instructions that pass the checks are then sent to the hydraulic actuators for execution.

[0039] This step is performed by the instruction allocation and security verification unit, which receives the action strategy output by the VLA core unit (i.e., the multimodal decision control unit). Then, the execution instruction allocation is performed, and the decomposed sub-instructions are subjected to security checks. Instructions that pass the checks are issued and executed by the underlying execution control unit.

[0040] S5. The hydraulic actuator performs actions according to the received instructions.

[0041] After the hydraulic actuator of the construction equipment completes a coordinated action, it updates the equipment status and the system enters the next control cycle. If the target height has not been reached, it returns to step S1 and repeats steps S1 to S5, forming a continuous "perception-decision-execution" control closed loop.

[0042] In one specific embodiment, step S1 involves setting a multi-point weighted fusion mechanism in the elevation benchmark fusion unit. For the core tube and the... The reference transfer between the giant columns is set at the connection of the corridor bridge. There are typically three (3) reference transfer measurement point pairs. Let the first... The core tube side readings for each measuring point pair are: The reading on the side of the giant column is Then the first on the giant pillar platform The fused elevation of each measuring point is calculated as follows: , Among them, weight Dynamically determined through stability analysis of historical data at measurement points: ; here For the first Each measurement point in the past Difference in readings within each sampling period The variance. When the variance of a pair of measurement points exceeds a threshold. When an error occurs, its weight is automatically reduced or it is removed, triggering a maintenance alarm. This adaptive weighted fusion method suppresses measurement noise and system errors at individual measurement points, ensuring the accuracy and robustness of the reference transfer.

[0043] In one specific embodiment, in step S1, based on the sensor network of the clustered multimodal sensing unit, a high-precision real-time dynamic (RTK) differential GPS receiver is configured for each platform. These receivers receive differential correction data through a wireless network to obtain the absolute three-dimensional coordinates. The data is transmitted in real time to the equipment status modeling unit, where it is fused with relative sensor data and static level data using Kalman filtering to eliminate the cumulative drift error of the tilt sensor, laser rangefinder, and static level. At the same time, by comparing the actual climbing trajectory with the original design trajectory, a trajectory deviation correction is generated and incorporated into the action decision-making process of the VLA model.

[0044] In one specific embodiment, in step S3, a VLA policy caching and fast matching mechanism is set up inside the multimodal decision control unit. This mechanism is located between the hierarchical intent decomposition module and the action policy generation module. It constructs a high-dimensional index database based on vector similarity to store historical decision pairs. Persistent storage is performed when a new state vector is generated. Upon input, first calculate its cosine similarity to existing states in the database. If the similarity exceeds a certain threshold and the task instructions... If a match is found, the corresponding historical action strategy will be invoked directly. It skips the model inference process, reduces system response time, and meets the real-time requirements of rapid response in hydraulic systems.

[0045] In one specific embodiment, in step S4, as follows: Figure 4 As shown, a hybrid VLA and PID control architecture is constructed within the underlying execution control unit. An independent PID control loop is embedded in the underlying controller of each hydraulic actuator. The global coordinated action target is output by the multimodal decision control unit. After instruction allocation, the values ​​are converted into dynamic setpoints for the PID controllers of each execution unit. The PID controllers then adjust the setpoints based on real-time sensor feedback and the deviations from the setpoints. Through the proportional term Integral terms and differential terms The linear combination of these signals generates the actual control signal, thus achieving the fusion of VLA global planning and PID local control.

[0046] In one specific embodiment, a digital twin simulation and security verification module is added between steps S3 and S4, such as... Figure 4 As shown, this module maintains a digital twin model synchronized with the construction equipment cluster, when VLA generates action strategies. Then, in the digital twin environment, a virtual rehearsal of multiple future control cycles is conducted in accelerated mode. The stress distribution and deformation at the bridge connection are calculated through finite element analysis. If the predicted stress or relative deformation exceeds the allowable value, the dangerous state characteristics in the rehearsal results are fed back to the multimodal decision control unit, triggering the model to regenerate the action strategy based on safety constraints, thus forming a "rehearsal-evaluation-replanning" safety assurance mechanism.

[0047] In one specific embodiment, in step S5, an online adaptive learning module is integrated into the multimodal decision control unit. This module collects execution data at the end of each control cycle. When the accumulated data reaches a preset threshold, the incremental learning process of the model is triggered, which calculates the actual structural risk indicators. The deviation from the expected value is used to update the model parameters, and the attention layer of the VLA model is fine-tuned so that the model can learn the dynamic characteristics of a specific building structure and equipment combination.

[0048] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0049] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A cluster control system for heterogeneous construction equipment based on a VLA model, used for the synchronous jacking and climbing coordination of construction equipment for building core tubes and mega-columns, characterized in that, include: Cluster multimodal sensing unit is used to measure the physical state of equipment clusters in real time; The cluster multimodal sensing unit includes: an attitude sensing network deployed on the core tube platform and each mega-column formwork measuring point, stroke and pressure sensors installed on all hydraulic actuators, laser ranging sensors deployed at each connecting corridor bridge interface, and a distributed independent static leveling measurement network covering the equipment cluster. The elevation datum fusion unit is used to process data from independent hydrostatic leveling networks and achieve a unified elevation datum. The equipment status modeling unit receives multi-source heterogeneous data streams from the sensing unit and unified elevation data processed by the elevation benchmark fusion unit, and models the heterogeneous construction equipment cluster control system. The multimodal decision control unit implements model inference through the VLA inference module. The VLA inference module integrates a hierarchical intent decomposition mechanism and a structural risk indicator perception and attention mechanism to generate a global collaborative action strategy based on structural risk indicators. ; The instruction allocation and security verification unit is located between the multimodal decision control unit and the underlying execution control unit. It receives the global collaborative action strategy, executes the instruction allocation operation, decomposes the unified action instruction into sub-instruction sets that conform to the control logic, and performs real-time verification on these sub-instruction sets. The underlying execution control unit is used to convert the instructions in the sub-instruction set into underlying electronic control signals that drive the actions of different mechanisms in various equipment. The human-computer interaction and remote monitoring platform provides operators with a graphical interface for inputting task commands and displaying the status of equipment clusters.

2. A cluster control method for heterogeneous construction equipment based on a VLA model, used for the synchronous jacking and climbing coordination of construction equipment for building core tubes and mega-columns, characterized in that... Includes the following steps: S1. Real-time acquisition of multi-source data, including the attitude of equipment clusters, the stroke and pressure of hydraulic actuators, and the relative displacement vector of the corridor bridge; acquisition of level instrument readings from each platform, and conversion of the measurements from each independent system into a unified elevation vector. Based on the collected multi-source data and elevation vectors This forms the global state vector. ; S2. The input high-level natural language task instructions are decomposed and parsed, and the macro task is decomposed into a series of internal sub-hints; The system integrates preset safety constraints such as bridge stress, platform height difference, and cylinder pressure with the current stage's sub-prompts to form task instruction prompts. ; The S3.VLA model is based on a global state vector. and task instruction prompts Generate a global collaborative action strategy based on structural risk indicators ; S4. Based on the global collaborative action strategy The system allocates instructions and performs safety checks on the decomposed sub-instructions. Instructions that pass the checks are then sent to the hydraulic actuators for execution. S5. The hydraulic actuator performs actions according to the received instructions.

3. The heterogeneous manufacturing equipment cluster control method based on the VLA model as described in claim 2, characterized in that, In step S1, the global state vector The equipment state modeling unit is used to create and output the data; the equipment state modeling unit is based on the collected multi-source data and elevation vectors. Performs modeling / mapping of the motion and force characteristics of heterogeneous devices, and outputs a global state vector. Global state vector This can be formally represented as: ； in, This represents the total number of subsystems in the cluster, including one core and... A giant pillar It is the first Real-time sensor data vectors of each subsystem It is a representation of the first The static embedding vector of the dynamic characteristics of each subsystem It is the first Elevation values ​​of each subsystem based on a unified benchmark This indicates a vector concatenation operation.

4. The heterogeneous manufacturing equipment cluster control method based on the VLA model as described in claim 3, characterized in that, In step S2, the elevation difference constraint between platforms based on a unified elevation datum can be expressed as: ； in, The maximum allowable height difference threshold, This represents the total number of subsystems in the cluster, including one core and... A giant pillar , They are the first , Elevation values ​​of each subsystem based on a unified benchmark.

5. The heterogeneous manufacturing equipment cluster control method based on the VLA model as described in claim 4, characterized in that, In step S3, a structural risk index function is first defined. This is used to quantify the risk of cluster instability in real time. ， in: Calculate the standard deviation of the elevations of all platforms based on a unified benchmark to characterize the overall flatness deviation of the cluster; It is the relative displacement vector of all the covered bridges. The L2 norm is used to characterize the degree of horizontal swaying or torsion of the cluster as a whole; in the third term... For the core tube elevation, For the first The elevation of the giant column platform is used to accumulate and quantify the stress concentration risk at the connection point. , and These are the weighting coefficients; The objective function of the VLA model for: ， in, Loss due to motion imitation; objective function Drive the VLA model to generate the current action At that time, predictive reasoning is performed to select a value that will make the next time step... Structural risk indicators The action with the lowest expected value.

6. The heterogeneous manufacturing equipment cluster control method based on the VLA model as described in claim 5, characterized in that, Step S3 is executed by the multimodal decision control unit; The multimodal decision control unit integrates an online adaptive learning module, which collects execution data at the end of each control cycle. When the accumulated data reaches a preset threshold, the incremental learning process of the model is triggered, which calculates the actual structural risk indicators. The deviation from the expected value is used to update the model parameters, and the attention layer of the VLA model is fine-tuned so that the model can learn the dynamic characteristics of a specific building structure and equipment combination.

7. The heterogeneous manufacturing equipment cluster control method based on the VLA model as described in claim 2, characterized in that, Step S3 is executed by the multimodal decision control unit; Within the multimodal decision control unit, a VLA policy caching and fast matching mechanism are implemented. This is achieved by constructing a high-dimensional index database based on vector similarity, which stores historical decision pairs. Persistent storage is performed; when a new state vector... Upon input, first calculate its cosine similarity to existing states in the database. If the similarity exceeds a certain threshold and the task instructions... If a match is found, the corresponding historical action strategy will be invoked directly. It skips the model inference process, reduces system response time, and meets the real-time requirements of rapid response in hydraulic systems.

8. The heterogeneous manufacturing equipment cluster control method based on the VLA model as described in claim 2, characterized in that, In step S1, the measured values ​​of each independent system are converted into a unified elevation vector. Performed by the elevation datum fusion unit; A multi-point weighted fusion mechanism is set up in the elevation datum fusion unit; for the core tube and the... The reference transfer between the giant columns is set at the connection of the corridor bridge. The reference transfer measurement point pair; let the first... The core tube side readings for each measuring point pair are: The reading on the side of the giant column is Then the first on the giant pillar platform The fused elevation of each measuring point is calculated as follows: ， Weight Dynamically determined through stability analysis of historical data at measurement points: ； in, For the first Each measurement point in the past Difference in readings within each sampling period The variance; when the variance of a pair of measurement points exceeds the threshold. When this happens, its weight will be automatically reduced or it will be removed, and a maintenance alarm will be triggered. As the global elevation benchmark, This is the reading of the reference point in the core tube system.

9. The heterogeneous manufacturing equipment cluster control method based on the VLA model as described in claim 2, characterized in that, In step S4, the verification command is issued through the underlying execution control unit; A hybrid VLA and PID control architecture is built within the underlying execution control unit. An independent PID control loop is embedded in the underlying controller of each hydraulic actuator. The decomposed sub-commands are converted into dynamic setpoints for the PID controllers of each actuator. The PID controllers adjust the setpoints based on the deviations from the real-time sensor feedback. Through the proportional term Integral terms and differential terms The linear combination of these signals generates the actual control signal, achieving the fusion of VLA global planning and PID local control; among which, For the purpose of global coordinated action, For actual implementation feedback, , , These are the PID proportional control parameters, integral control parameters, and derivative control parameters, respectively.

10. The heterogeneous manufacturing equipment cluster control method based on the VLA model as described in claim 2, characterized in that, Between steps S3 and S4, a digital twin simulation and security verification step is added, which is executed by the digital twin simulation and security verification module. The digital twin simulation and security verification module maintains a digital twin model synchronized with the construction equipment cluster, when generating a global collaborative action strategy. Then, a virtual rehearsal of multiple future control cycles is first conducted in an accelerated mode in a digital twin environment. The stress distribution and deformation at the bridge connection are calculated through finite element analysis. If the predicted stress or relative deformation exceeds the allowable value, the dangerous state characteristics in the rehearsal results are fed back to the multimodal decision control unit, triggering the model to regenerate the global collaborative action strategy based on safety constraints, thus forming a safety assurance mechanism of rehearsal-evaluation-replanning.

Images