Intelligent construction equipment control method based on multi-modal large model

Abstract

The invention discloses an intelligent construction equipment control method based on a multi-modal large model, and relates to the technical field of intelligent construction and construction equipment control, and the method comprises the steps: collecting concrete pumping flow, pipeline pressure, construction area point cloud and multi-angle images, obtaining a voice instruction of an operator at the same time, and carrying out the noise suppression and voiceprint recognition; inputting the trained multi-modal large model to carry out cloth coverage precision judgment, calculating a regional granularity factor, comparing the regional granularity factor with a coverage threshold value, and automatically generating a task label or triggering a voice correction strategy; a dynamic feeding stability coefficient is constructed by calculating the standard deviation of flow and pressure data, and intelligent monitoring and regulation of the running state and the construction environment of the material distributing machine are achieved in combination with a load balancing coefficient and a construction risk probability; the system also establishes a full-link tracing mechanism, realizes traceability of operation instructions, personnel identities and equipment states, ensures controllable construction quality and clear responsibility, and realizes task process optimization and continuous improvement.

Description

Technical Field

[0001] This invention relates to the field of intelligent construction and construction equipment control technology, specifically to an intelligent construction equipment control method based on a multimodal large model. Background Technology

[0002] Intelligent construction equipment, as a crucial support system for modern engineering construction, is widely used in the concrete pouring and forming processes of high-rise buildings, bridges, and large-scale infrastructure, placing extremely high demands on the precision control, stability assurance, and safety management of the construction process. With the development of artificial intelligence and big data technologies, multimodal perception and intelligent decision-making are gradually becoming the core directions driving the intelligent upgrading of construction equipment. Among these, multimodal large-scale models, by integrating heterogeneous data from multiple sources such as vision, voice, sensor signals, and BIM design models, provide a new technical path for the precise control and dynamic scheduling of construction equipment.

[0003] In traditional construction equipment control methods, data acquisition and control decisions often rely on a single signal source, such as flow or pressure sensor feedback, making it difficult to achieve global perception and coordinated scheduling of the construction site. Due to the lack of systematic multimodal fusion and unified data standards, the following problems often exist: On the one hand, real-time determination of the coverage area of ​​the concrete placing boom depends on manual experience and limited monitoring methods, which can easily lead to a mismatch between the actual pouring area and the design area, resulting in missed pouring or over-pouring; on the other hand, in the coordinated construction of multiple concrete placing booms, the pressure and flow of each pipeline fluctuate frequently, and the material supply stability is insufficient, resulting in intermittent or locally uneven pouring.

[0004] Furthermore, the complex working environment at construction sites results in low recognition rates for operator voice commands against noisy backgrounds, and the lack of ability to convert natural speech into structured control commands hinders the real-time and accurate human-machine interaction. Simultaneously, existing equipment relies primarily on limited monitoring videos and manual inspections for risk prediction and control, making it difficult to promptly identify potential risks between equipment and personnel, posing safety hazards. Particularly in cases of quality defects or operational anomalies, the lack of a full-chain traceability mechanism makes it impossible to quickly pinpoint the responsible party and their instructions, leading to unclear responsibility allocation and a lack of a closed-loop mechanism for using feedback information for subsequent optimization. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a control method for intelligent construction equipment based on a multimodal large model, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a control method for intelligent construction equipment based on a multimodal large model, comprising the following steps:

[0007] Step 1: Real-time monitoring of the multi-concrete-pumping collaborative pouring scenario at the construction site; real-time acquisition of concrete pump flow and pressure data, construction area point cloud data, construction surface images and multi-angle views, as well as operator voice commands processed by noise suppression and voiceprint recognition; then, the point cloud is integrated with the BIM design model to generate a three-dimensional model of the construction area, and after data preprocessing, a standardized input dataset is established.

[0008] Step 2: Input the standardized input dataset into the trained multimodal large model, determine the accuracy of the construction area, and perform semantic segmentation based on the 3D model and normalized video frames to obtain the actual pouring coverage area Sa and the BIM design area Sb. Calculate the regional granularity factor Garea and compare it with the coverage threshold Gth to determine whether the material coverage accuracy is qualified. If qualified, generate a local task label Ltask; if unqualified, provide an appropriate strategy.

[0009] Step 3: Extract the flow rate data F(t) and pressure data P(t) of the corresponding feed pipe based on the local task label Ltask, calculate their standard deviation after time series normalization, and obtain the standard deviation of flow rate and pressure; further calculate the dynamic feeding stability coefficient Sdyn and compare it with the feeding threshold Sth to determine whether the feeding process is stable. If it is unstable, give an appropriate strategy.

[0010] Step 4: Collect instantaneous flow and pressure signals of each concrete placing machine through sensors. After wireless network transmission and Kalman filtering normalization, a standardized flow and pressure sequence is obtained. The average and maximum values ​​are obtained by using exponential weighted moving average and extreme value search algorithms. The equipment load balancing coefficient Leq is calculated and compared with the load balancing threshold Lth to determine whether the concrete placing machine is operating in a balanced manner. If it is unbalanced, an appropriate strategy is given.

[0011] Step 5: By collecting the real-time minimum distance Δd between the equipment and personnel, construct the construction risk probability Prisk, and compare it with the construction risk probability threshold Pth to determine whether the construction environment is safe. If it is not safe, give appropriate strategies.

[0012] Step Six: The system establishes a full-link traceability mechanism for voice commands, executed actions, and equipment status. It confirms the operator's identity through voiceprint recognition and records the command and time. When quality defects or operational abnormalities occur, the system can quickly trace back to the relevant personnel and commands, clearly defining responsibilities. The traceability results are fed back to the construction knowledge graph for task process optimization and strategy improvement, forming a closed-loop management system with controllable quality, clear responsibilities, and continuous optimization.

[0013] Preferably, step one includes:

[0014] S11. Real-time monitoring of the multi-concrete placing machine collaborative pouring scenario at the construction site; flow sensors and pressure sensors are installed at the inlet and end of the placing pipe of each placing machine at the construction site to collect the concrete pumping volume flow rate F(t) and the instantaneous pressure value P(t) of the pipeline per unit time in real time.

[0015] S12 collects point cloud data of the construction surface and structural components by deploying lidar scanning equipment in the construction area, and matches and fuses the point cloud results with the BIM design model to generate a three-dimensional model M3D of the construction area.

[0016] S13. Deploy panoramic cameras at high positions in the construction area and install wide-angle monitoring cameras in key passages to obtain two-dimensional image data and multi-angle coverage views of the construction surface in real time.

[0017] S14. By deploying voice receiving terminals in the operator's work area at the construction site and configuring directional microphone arrays and noise suppression modules, the voice command signals of the operators are collected in real time. The collected voice data is first processed by adaptive filtering through a dynamic noise database, then the operator's identity code is extracted by a voiceprint recognition algorithm, and then the natural speech is converted into structured control commands, and the timestamp of the command issuance is recorded.

[0018] S15. The collected data undergoes filtering and outlier removal, point cloud registration and noise suppression, image distortion correction and brightness equalization, and speech signal filtering and voiceprint recognition. It is then uniformly converted into a structured data format and timestamped, and a standardized input dataset is established.

[0019] Preferably, step two includes:

[0020] S21. By inputting the standardized input dataset into the trained multimodal large model, the accuracy of the construction area within the coverage of the material placement pipe is determined.

[0021] S22. Based on the 3D model of the construction area M3D, combined with the preprocessed normalized video frame data, perform image semantic segmentation on the construction surface; obtain the actual segmented pouring coverage area Sa and the design area Sb of the BIM model through the regional projection overlay method.

[0022] Preferably, step two further includes:

[0023] S23. Match the actual pouring area Sa with the BIM model design area Sb, and after dimensionless processing, calculate and obtain the area granularity factor Garea.

[0024] S24. By setting a preset coverage threshold Gth and comparing the region granularity factor Garea with the coverage threshold Gth, the first evaluation results are obtained, including:

[0025] When the coverage threshold Gth ≤ the area particle size factor Garea ≤ the coverage threshold Gth * 120%, it indicates that the material coverage accuracy is qualified, the system recognizes that the pouring is consistent with the design, and generates a local task label Ltask in combination with the construction knowledge graph;

[0026] When the area granularity factor Garea < the coverage threshold Gth, it indicates that the coverage accuracy of the material is unqualified, the coverage is insufficient, and there is a risk of missed pouring, triggering the first voice interaction correction process: the operator inputs the replenishment command through the voice receiving terminal, and the command is sent to the trained multimodal large model in the industrial control computer; the multimodal large model calls the dynamic noise database to perform adaptive noise suppression on the on-site voice and improve the clarity of the command; the multimodal large model converts the voice into structured commands through end-to-end semantic parsing, including: reducing the moving speed of the material pipe by 10% and increasing the current area's material placement time by 8%, while highlighting the insufficient area on the monitoring interface;

[0027] When the area granularity factor Garea > coverage threshold Gth * 120%, it indicates that the fabric coverage accuracy is unqualified, the coverage is excessive, and there is a risk of over-pouring. This triggers the second voice interaction correction process: the operator inputs the "reduce" command, which is converted into a structured command after model noise suppression and semantic parsing. This command includes: adjusting the swing amplitude of the fabric tube and limiting the local flow rate, performing dynamic correction of the flow rate and movement amplitude, and gradually reducing the risk of over-coverage until the coverage threshold Gth ≤ area granularity factor Garea ≤ coverage threshold Gth * 120%.

[0028] Preferably, step three includes:

[0029] S31. Based on the local task label Ltask, by calling the fabric pipe operation data corresponding to the local task label Ltask in the standardized input dataset, extract the real-time flow data F(t) and real-time pressure data P(t) of the fabric pipe, perform time series normalization processing to obtain the flow fluctuation sequence and pressure fluctuation sequence, calculate the standard deviation of the flow fluctuation sequence and the pressure fluctuation sequence respectively, and obtain the flow standard deviation σF and the pressure standard deviation σP.

[0030] Preferably, step three further includes:

[0031] S32. After dimensionless processing of the obtained flow rate standard deviation σF and pressure standard deviation σP, the dynamic feeding stability coefficient Sdyn is calculated.

[0032] S33. By setting a preset material supply stability threshold Sth, and comparing and analyzing the dynamic material supply stability coefficient Sdyn with the material supply stability threshold Sth, the second evaluation results are obtained, including:

[0033] When the dynamic feeding stability coefficient Sdyn ≤ feeding stability threshold Sth, it indicates that the feeding process is stable and there is no risk of fluctuation in the fabric operation status, proceed to step four;

[0034] When the dynamic feeding stability coefficient Sdyn > the feeding stability threshold Sth, it indicates that the feeding process is unstable and there is a risk of fluctuation in the fabric operation status, triggering the third voice interaction correction process: the correction command is input through the voice receiving terminal and sent to the trained multimodal large model in the industrial control computer; after the multimodal large model performs adaptive noise suppression in combination with the dynamic noise database, it performs semantic parsing and converts natural language into structured commands, including a 5%–10% reduction in the fabric tube speed and a local path correction offset of 5% of the design path; the system issues correction commands and updates the flow and pressure acquisition sequences in real time, recalculates, until the dynamic feeding stability coefficient Sdyn ≤ the feeding stability threshold Sth.

[0035] Preferably, step four includes:

[0036] S41. By installing flow sensors and pressure sensors at the inlet and end of the placing pipe of each placing boom at the construction site, the instantaneous flow rate F of each placing boom is collected in real time. i With pipeline pressure P j The collected flow and pressure signals are transmitted to the industrial control computer via a wireless sensor network. Kalman filtering and normalization algorithms are then used to denoise and standardize the data, resulting in a standardized flow sequence {F}. ′ 1, F ′ 2, ..., F ′ n} and standardized pressure sequence {P1 ′ P2 ′ ..., P ′ m};

[0037] S42. Based on the acquired standardized flow rate sequence and standardized pressure sequence, the average flow rate mean(F) of each placing boom is obtained using an exponentially weighted moving average algorithm. i ′ ) and mean pressure (P) j ′ The maximum flow rate (max(F)) is obtained through an extreme value search algorithm with threshold constraints. i ′ ) and maximum pressure max(P j ′ ).

[0038] Preferably, step four further includes:

[0039] S43, the average flow rate mean(F) of each fabric placing machine is obtained.i ′ Mean pressure (P) j ′ ), maximum flow rate max(F) i ′ ) and maximum pressure max(P j ′ After dimensionless processing, the equipment load balancing coefficient Leq is calculated and obtained.

[0040] S44. By comparing and analyzing the device load balancing coefficient Leq with the load balancing threshold Lth using a preset load balancing threshold, the third evaluation result is obtained, including:

[0041] When the equipment load balancing coefficient Leq ≤ load balancing threshold Lth, it indicates that the concrete placing machine is operating in a balanced state, and monitoring should continue.

[0042] When the equipment load balancing coefficient Leq > the load balancing threshold Lth, it indicates that the concrete placing machine is out of balance and the load distribution is uneven, triggering the fourth voice interaction correction process: the system calls the path planning algorithm to redistribute the concrete placing sequence and generate a flexible work order; the system dynamically adjusts the concrete placing speed and operation sequence of the overloaded equipment in the scheduling results to keep the overall construction progress stable; the correction results are pushed in real time through the monitoring interface and the operator is prompted to confirm on the voice interaction terminal.

[0043] Preferably, step five includes:

[0044] S51. By deploying lidar and ultrasonic sensors within the operating radius of the concrete placing boom at the construction site, the spatial position data between the equipment and personnel is collected in real time. Combined with a standardized input dataset, the collected spatial distance information is filtered and normalized to obtain the real-time minimum distance Δd between the equipment and personnel.

[0045] S52. By obtaining the real-time minimum distance Δd, the risk field model in the construction knowledge graph is called, and the exponential decay risk calculation method is used. After dimensionless processing, the construction risk probability Prisk is constructed.

[0046] S53. By setting a preset construction risk probability threshold Pth, and comparing and analyzing the construction risk probability Prisk with the construction risk probability threshold Pth, the fourth assessment results are obtained, including:

[0047] When the construction risk probability Prisk < the construction risk probability threshold Pth, it indicates that the construction environment is safe, the risk is controllable, and continuous monitoring is required.

[0048] When the construction risk probability Prisk < the construction risk probability threshold Pth, it indicates that the construction environment is unsafe and the risk is uncontrollable. A risk warning message is immediately generated, and an automatic shutdown command is issued through the control bus. At the same time, a pop-up window is displayed on the monitoring interface, and a warning signal is issued in conjunction with the voice interaction terminal. In the shutdown state, the system enters the risk elimination detection process, tracking the change of Δd in real time. Based on the change value of Δd, the construction risk probability Prisk is calculated in real time until the construction risk probability Prisk < the construction risk probability threshold Pth. At this point, the shutdown command is automatically lifted, and the concrete placing boom operation is resumed.

[0049] Preferably, step six includes:

[0050] S61. Throughout the construction process, the system establishes a full-link traceability mechanism for voice commands, executed actions, and equipment status; it uses voiceprint recognition technology to confirm the identity of operators and records and archives the control commands they issue along with the corresponding execution times, forming a traceable operation link.

[0051] S62. When quality defects or operational abnormalities occur during subsequent construction, the system can quickly trace back to the relevant operators and corresponding control commands, thus clarifying the division of responsibilities. At the same time, the traceability results will be fed back to the construction knowledge graph for the optimization of subsequent task processes and strategy improvement, thereby forming a closed loop of intelligent construction with controllable quality, clear responsibilities, and continuous optimization.

[0052] This invention provides a control method for intelligent construction equipment based on a multimodal large model. It has the following beneficial effects:

[0053] (1) This intelligent construction equipment control method based on multimodal large model deploys various types of sensing devices such as flow rate, pressure, lidar, panoramic / wide-angle cameras and voice receiving terminals, and combines point cloud and BIM model fusion, image correction and voice noise suppression to realize real-time acquisition and standardized processing of multi-source heterogeneous data at the construction site, ensuring the integrity and accuracy of the input dataset, and providing a reliable foundation for subsequent intelligent analysis of multimodal large model.

[0054] (2) The intelligent construction equipment control method based on multimodal large model introduces multimodal large model, performs image semantic segmentation and BIM matching on the construction area, calculates the regional granularity factor and dynamically evaluates the pouring accuracy in combination with the coverage threshold, can detect the risk of insufficient coverage or over-pouring in a timely manner, and generates correction strategies through voice interaction, which significantly improves the accuracy and consistency of material placement operation.

[0055] (3) The intelligent construction equipment control method based on multimodal large model calculates the dynamic material supply stability coefficient Sdyn by the standard deviation of flow rate and pressure, and sets a threshold for stability assessment; at the same time, combined with the load balance coefficient Leq, the speed and operation sequence of multiple material placing machines are dynamically adjusted to achieve the stability of the material supply process and the load balance of equipment operation, so as to avoid the impact of fluctuations and imbalances on construction progress and quality.

[0056] (4) This intelligent construction equipment control method based on a multimodal large model constructs a construction risk probability model based on the real-time minimum distance between equipment and personnel, which can trigger shutdown and early warning measures in a timely manner to ensure on-site operation safety. At the same time, it establishes a full-link traceability mechanism of voice command - execution action - equipment status, which can quickly locate the responsible personnel and operation links, and feed the results back to the construction knowledge graph for continuous optimization of the construction process, so as to realize a closed loop of intelligent construction with controllable quality, clear responsibility and continuous improvement. Attached Figure Description

[0057] Figure 1 This is a schematic diagram illustrating the steps of a control method for intelligent construction equipment based on a multimodal large model according to the present invention. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Example 1

[0060] Please see Figure 1 This invention provides a control method for intelligent construction equipment based on a multimodal large model, comprising the following steps:

[0061] Step 1: Real-time monitoring of the multi-concrete-pumping collaborative pouring scenario at the construction site; real-time acquisition of concrete pump flow and pressure data, construction area point cloud data, construction surface images and multi-angle views, as well as operator voice commands processed by noise suppression and voiceprint recognition; then, the point cloud is integrated with the BIM design model to generate a three-dimensional model of the construction area, and after data preprocessing, a standardized input dataset is established.

[0062] Step 2: Input the standardized input dataset into the trained multimodal large model, determine the accuracy of the construction area, and perform semantic segmentation based on the 3D model and normalized video frames to obtain the actual pouring coverage area Sa and the BIM design area Sb. Calculate the regional granularity factor Garea and compare it with the coverage threshold Gth to determine whether the material coverage accuracy is qualified. If qualified, generate a local task label Ltask; if unqualified, provide an appropriate strategy.

[0063] Step 3: Extract the flow rate data F(t) and pressure data P(t) of the corresponding feed pipe based on the local task label Ltask, calculate their standard deviation after time series normalization, and obtain the standard deviation of flow rate and pressure; further calculate the dynamic feeding stability coefficient Sdyn and compare it with the feeding threshold Sth to determine whether the feeding process is stable. If it is unstable, give an appropriate strategy.

[0064] Step 4: Collect instantaneous flow and pressure signals of each concrete placing machine through sensors. After wireless network transmission and Kalman filtering normalization, a standardized flow and pressure sequence is obtained. The average and maximum values ​​are obtained by using exponential weighted moving average and extreme value search algorithms. The equipment load balancing coefficient Leq is calculated and compared with the load balancing threshold Lth to determine whether the concrete placing machine is operating in a balanced manner. If it is unbalanced, an appropriate strategy is given.

[0065] Step 5: By collecting the real-time minimum distance Δd between the equipment and personnel, construct the construction risk probability Prisk, and compare it with the construction risk probability threshold Pth to determine whether the construction environment is safe. If it is not safe, give appropriate strategies.

[0066] Step Six: The system establishes a full-link traceability mechanism for voice commands, executed actions, and equipment status. It confirms the operator's identity through voiceprint recognition and records the command and time. When quality defects or operational abnormalities occur, the system can quickly trace back to the relevant personnel and commands, clearly defining responsibilities. The traceability results are fed back to the construction knowledge graph for task process optimization and strategy improvement, forming a closed-loop management system with controllable quality, clear responsibilities, and continuous optimization.

[0067] In this embodiment, by real-time acquisition and standardized processing of multi-source data during the collaborative pouring process of multiple concrete placing machines, combined with the accuracy determination of a multimodal large model, dynamic material supply stability analysis, equipment load balancing monitoring, construction risk assessment, and full-link voice command traceability, intelligent construction closed-loop management is achieved, which ensures controllable pouring accuracy, stable material supply process, balanced equipment operation, controllable construction safety, and traceable operational responsibility. This significantly improves construction quality, consistency, and safety, and promotes continuous optimization of the construction process.

[0068] Example 2

[0069] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, step one includes:

[0070] S11. Real-time monitoring of the multi-concrete placing machine collaborative pouring scenario at the construction site; flow sensors and pressure sensors are installed at the inlet and end of the placing pipe of each placing machine at the construction site to collect the concrete pumping volume flow rate F(t) and the instantaneous pressure value P(t) of the pipeline per unit time in real time.

[0071] S12 collects point cloud data of the construction surface and structural components by deploying lidar scanning equipment in the construction area, and matches and fuses the point cloud results with the BIM design model to generate a three-dimensional model M3D of the construction area.

[0072] S13. Deploy panoramic cameras at high positions in the construction area and install wide-angle monitoring cameras in key passages to obtain two-dimensional image data and multi-angle coverage views of the construction surface in real time.

[0073] S14. By deploying voice receiving terminals in the operator's work area at the construction site and configuring directional microphone arrays and noise suppression modules, the voice command signals of the operators are collected in real time. The collected voice data is first processed by adaptive filtering through a dynamic noise database, then the operator's identity code is extracted by a voiceprint recognition algorithm, and then the natural speech is converted into structured control commands, and the timestamp of the command issuance is recorded.

[0074] S15. The collected data undergoes filtering and outlier removal, point cloud registration and noise suppression, image distortion correction and brightness equalization, and speech signal filtering and voiceprint recognition. It is then uniformly converted into a structured data format and timestamped, and a standardized input dataset is established.

[0075] In this embodiment, by collecting flow rate, pressure, point cloud, images, and voice commands in real time during the collaborative pouring process of multiple concrete placing machines at the construction site, and performing filtering, registration, correction, and structuring processing on the multi-source data, a standardized input dataset is established. This enables comprehensive perception and high-precision digital representation of the construction site status, providing a reliable data foundation for subsequent intelligent decision-making and equipment control, and improving the accuracy and response efficiency of construction monitoring.

[0076] Example 3

[0077] This embodiment is an explanation based on Embodiment 2. Please refer to it. Figure 1 Specifically, step two includes:

[0078] S21. By inputting the standardized input dataset into the trained multimodal large model, the accuracy of the construction area within the coverage of the material placement pipe is determined.

[0079] S22. Based on the 3D model of the construction area M3D, combined with the preprocessed normalized video frame data, perform image semantic segmentation on the construction surface; obtain the actual segmented pouring coverage area Sa and the design area Sb of the BIM model through the regional projection overlay method.

[0080] In this embodiment, by inputting a standardized input dataset into a trained multimodal large model and combining it with a three-dimensional construction model and normalized video frames for image semantic segmentation, the accuracy of the concrete placement pipe coverage area can be determined. This allows for accurate acquisition of the difference between the actual pouring area and the design area, providing a quantitative basis for construction coverage quality assessment and improving the reliability and operability of pouring accuracy monitoring.

[0081] Example 4

[0082] This embodiment is an explanation based on Embodiment 3. Please refer to it. Figure 1 Specifically, step two also includes:

[0083] S23. Match the actual poured area Sa with the BIM model designed area Sb, and after dimensionless processing, calculate the area granularity factor Garea, as follows:

[0084]

[0085] S24. By setting a preset coverage threshold Gth and comparing the region granularity factor Garea with the coverage threshold Gth, the first evaluation results are obtained, including:

[0086] When the coverage threshold Gth ≤ the area particle size factor Garea ≤ the coverage threshold Gth * 120%, it indicates that the material coverage accuracy is qualified, the system recognizes that the pouring is consistent with the design, and generates a local task label Ltask in combination with the construction knowledge graph;

[0087] When the area granularity factor Garea < the coverage threshold Gth, it indicates that the coverage accuracy of the material is unqualified, the coverage is insufficient, and there is a risk of missed pouring, triggering the first voice interaction correction process: the operator inputs the replenishment command through the voice receiving terminal, and the command is sent to the trained multimodal large model in the industrial control computer; the multimodal large model calls the dynamic noise database to perform adaptive noise suppression on the on-site voice and improve the clarity of the command; the multimodal large model converts the voice into structured commands through end-to-end semantic parsing, including: reducing the moving speed of the material pipe by 10% and increasing the current area's material placement time by 8%, while highlighting the insufficient area on the monitoring interface;

[0088] When the area granularity factor Garea > coverage threshold Gth * 120%, it indicates that the fabric coverage accuracy is unqualified, the coverage is excessive, and there is a risk of over-pouring. This triggers the second voice interaction correction process: the operator inputs the "reduce" command, which is converted into a structured command after model noise suppression and semantic parsing. This command includes: adjusting the swing amplitude of the fabric tube and limiting the local flow rate, performing dynamic correction of the flow rate and movement amplitude, and gradually reducing the risk of over-coverage until the coverage threshold Gth ≤ area granularity factor Garea ≤ coverage threshold Gth * 120%.

[0089] The method for setting the coverage threshold Gth is as follows: Through statistical analysis of a large amount of concrete placing boom pouring coverage data from construction sites, the matching ratio range between the actual coverage area and the design area under ideal and insufficient / excessive coverage conditions is extracted. Combined with the judgment of experienced technical personnel, a reasonable coverage judgment value is determined. Construction specifications, BIM design model standards, and concrete construction process specifications are referenced, as these specifications usually provide a reasonable range for coverage accuracy. This threshold is used to effectively distinguish between qualified and unqualified states of concrete placing boom coverage accuracy, ensuring the uniformity and accuracy of construction quality.

[0090] In this embodiment, the system calculates the regional granularity factor Garea and compares it with the preset coverage threshold Gth to achieve real-time evaluation of the coverage accuracy of the material. When the coverage is insufficient or excessive, the system can trigger a voice interaction correction process to intelligently convert operator instructions into a structured control strategy and dynamically adjust the operating parameters of the material distribution pipe, thereby effectively reducing the risk of under-pouring or over-pouring and improving the accuracy and automation level of construction quality.

[0091] Example 5

[0092] This embodiment is an explanation based on Embodiment 4. Please refer to it. Figure 1 Specifically, step three includes:

[0093] S31. Based on the local task label Ltask, by calling the fabric pipe operation data corresponding to the local task label Ltask in the standardized input dataset, extract the real-time flow data F(t) and real-time pressure data P(t) of the fabric pipe, perform time series normalization processing to obtain the flow fluctuation sequence and pressure fluctuation sequence, calculate the standard deviation of the flow fluctuation sequence and the pressure fluctuation sequence respectively, and obtain the flow standard deviation σF and the pressure standard deviation σP.

[0094] In this embodiment, real-time flow rate F(t) and pressure P(t) data of the fabric delivery pipe are extracted based on the local task tag Ltask, and their fluctuation standard deviation is calculated to achieve precise monitoring of the material supply process. This helps to identify abnormal flow rate or pressure in a timely manner and improve the stability of fabric delivery and the controllability of the construction process.

[0095] Example 6

[0096] This embodiment is an explanation based on Embodiment 5. Please refer to it. Figure 1 Specifically, step three also includes:

[0097] S32. After dimensionless processing of the obtained flow rate standard deviation σF and pressure standard deviation σP, the dynamic feeding stability coefficient Sdyn is calculated using the following formula:

[0098] Sdyn=w1*σF+w2*σP;

[0099] In the formula, w1 and w2 represent weighting coefficients;

[0100] The acquisition methods for w1 and w2 are as follows: Statistical analysis is performed on time-series data of flow rate F(t) and pressure P(t) of placing booms in numerous construction scenarios. Combined with simulation results, the sensitivity and impact of the standard deviation of flow rate σF and the standard deviation of pressure σP on the stability of dynamic material supply and the uniformity of the material placement process are evaluated. A reasonable weighting ratio is comprehensively determined based on the needs of the multi-pump collaborative pumping model and dynamic material supply control process at the construction site. Referring to concrete pumping operation specifications, placing boom stability evaluation standards, and relevant academic research, and combining the experience judgment of on-site construction technicians, the following empirical weighting coefficients are set to ensure the accuracy and operability of the dynamic material supply stability coefficient Sdyn, supporting subsequent material supply correction and construction quality control.

[0101] w1 = 0.6: Characterizes the influence of the standard deviation of flow fluctuation σF on the stability of material supply, and has the dominant weight, directly reflecting the constraining effect of the uniformity of material flow on the stability of the material supply process;

[0102] w2 = 0.4: Characterizes the impact of pressure fluctuation standard deviation σP on the stability of material supply, accounting for auxiliary weight, and reflects the contribution of pipeline pressure changes to the continuity and safety of material supply.

[0103] S33. By setting a preset material supply stability threshold Sth, and comparing and analyzing the dynamic material supply stability coefficient Sdyn with the material supply stability threshold Sth, the second evaluation results are obtained, including:

[0104] When the dynamic feeding stability coefficient Sdyn ≤ feeding stability threshold Sth, it indicates that the feeding process is stable and there is no risk of fluctuation in the fabric operation status, proceed to step four;

[0105] When the dynamic feeding stability coefficient Sdyn > the feeding stability threshold Sth, it indicates that the feeding process is unstable and there is a risk of fluctuation in the fabric operation status, triggering the third voice interaction correction process: the correction command is input through the voice receiving terminal and sent to the trained multimodal large model in the industrial control computer; after the multimodal large model performs adaptive noise suppression in combination with the dynamic noise database, it performs semantic parsing and converts natural language into structured commands, including a 5%–10% reduction in the fabric tube speed and a local path correction offset of 5% of the design path; the system issues correction commands and updates the flow and pressure acquisition sequences in real time, recalculates, until the dynamic feeding stability coefficient Sdyn ≤ the feeding stability threshold Sth.

[0106] The method for setting the material supply stability threshold Sth is as follows: Through statistical analysis of a large amount of time-series data on the flow and pressure of the concrete placing boom, the distribution range of the dynamic material supply fluctuation coefficient under stable and unstable supply conditions is extracted. Combined with the experience and judgment of construction technicians, a reasonable material supply stability judgment value is determined. Reference is made to concrete pumping operation specifications, concrete placing boom equipment parameters, and construction process requirements, which typically provide an acceptable fluctuation range for material supply stability. This threshold is used to distinguish whether the material distribution process is stable, effectively reducing the risk of material placement fluctuations during construction.

[0107] In this embodiment, by calculating the dynamic material supply stability coefficient Sdyn and comparing it with the threshold Sth, the fluctuation of the material conveying process can be evaluated in real time. When the material supply is unstable, the voice interaction correction process can be triggered to automatically adjust the speed and path deviation of the material supply tube, so as to ensure the stability of the material supply process and improve the construction quality and operation safety.

[0108] Example 7

[0109] This embodiment is an explanation based on Embodiment 6. Please refer to it. Figure 1 Specifically, step four includes:

[0110] S41. By installing flow sensors and pressure sensors at the inlet and end of the placing pipe of each placing boom at the construction site, the instantaneous flow rate F of each placing boom is collected in real time. i With pipeline pressure P j The collected flow and pressure signals are transmitted to the industrial control computer via a wireless sensor network. Kalman filtering and normalization algorithms are then used to denoise and standardize the data, resulting in a standardized flow sequence {F}. ′ 1, F ′ 2, ..., F ′ n} and standardized pressure sequence {P1 ′ P2 ′ ..., P ′ m};

[0111] S42. Based on the acquired standardized flow rate sequence and standardized pressure sequence, the average flow rate mean(F) of each placing boom is obtained using an exponentially weighted moving average algorithm. i ′ ) and mean pressure (P) j ′ The maximum flow rate (max(F)) is obtained through an extreme value search algorithm with threshold constraints. i ′ ) and maximum pressure max(P j ′ ).

[0112] In this embodiment, by real-time acquisition, noise reduction and normalization processing, and calculation of average and extreme values ​​of the flow and pressure signals of each concrete placing boom, the operating status of the concrete placing boom is accurately monitored, providing reliable data support for subsequent load balancing assessment and dynamic control, and improving the stability and controllability of the construction process.

[0113] Example 8

[0114] This embodiment is an explanation based on Embodiment 7. Please refer to it. Figure 1 Specifically, step four also includes:

[0115] S43, the average flow rate mean(F) of each fabric placing machine is obtained. i ′ Mean pressure (P) j ′ ), maximum flow rate max(F) i ′ ) and maximum pressure max(P j ′ After dimensionless processing, the equipment load balancing coefficient Leq is calculated using the following formula:

[0116]

[0117] In the formula, a1 and a2 represent weighting coefficients;

[0118] The setting method for a1 and a2 is as follows: Real-time flow rate F of each concrete placing boom is collected under different construction scenarios. i ′ With pipeline pressure P j ′Data, combined with on-site concrete placing boom scheduling and pumping models, is statistically analyzed to assess the impact of flow rate fluctuations and pressure changes on the load distribution balance of equipment. Considering the load fluctuation range under multiple operating conditions and construction schedule requirements, reasonable weighting ratios are determined. Referring to concrete pumping operation specifications, construction equipment load balance assessment standards, and relevant academic research, and incorporating expert judgment from the construction site, the following empirical weighting coefficients are set:

[0119] a1 = 0.55: Characterizes the impact of fabric feeder flow rate on load balancing, which has a dominant weight and directly reflects the contribution of fabric feeder speed and distribution to overall load balancing;

[0120] a2 = 0.45: Characterizes the influence of the fabric distribution pipe pressure on load balancing, accounting for auxiliary weight, and reflects the constraining effect of pressure fluctuations on the stability of equipment load distribution.

[0121] S44. By comparing and analyzing the device load balancing coefficient Leq with the load balancing threshold Lth using a preset load balancing threshold, the third evaluation result is obtained, including:

[0122] When the equipment load balancing coefficient Leq ≤ load balancing threshold Lth, it indicates that the concrete placing machine is operating in a balanced state, and monitoring should continue.

[0123] When the equipment load balancing coefficient Leq > the load balancing threshold Lth, it indicates that the concrete placing machine is out of balance and the load distribution is uneven, triggering the fourth voice interaction correction process: the system calls the path planning algorithm to redistribute the concrete placing sequence and generate a flexible work order; the system dynamically adjusts the concrete placing speed and operation sequence of the overloaded equipment in the scheduling results to keep the overall construction progress stable; the correction results are pushed in real time through the monitoring interface and the operator is prompted to confirm on the voice interaction terminal.

[0124] The load balancing threshold Lth is set as follows: Statistical analysis of the flow and pressure distribution data of multiple concrete placing booms operating simultaneously extracts the range of load distribution coefficient changes under load balancing and load imbalance conditions. This, combined with the experience of technical personnel, determines a reasonable load balancing judgment value. Reference is made to construction scheduling specifications, performance parameters provided by equipment manufacturers, and construction efficiency standards; these specifications typically provide a reasonable range for load balancing. This threshold is used to effectively determine whether the concrete placing boom's operating status is balanced, ensuring stable construction progress and equipment lifespan.

[0125] In this embodiment, by calculating the equipment load balancing coefficient and comparing it with a threshold, the system can achieve real-time balanced assessment of the operating status of multiple concrete placing machines. When the load is unbalanced, the system can automatically adjust the placing sequence and speed to optimize the operation timing, thereby ensuring the stability of the overall construction progress and improving construction efficiency and resource utilization.

[0126] Example 9

[0127] This embodiment is an explanation based on Embodiment 8. Please refer to it. Figure 1 Specifically, step five includes:

[0128] S51. By deploying lidar and ultrasonic sensors within the operating radius of the concrete placing boom at the construction site, the spatial position data between the equipment and personnel is collected in real time. Combined with a standardized input dataset, the collected spatial distance information is filtered and normalized to obtain the real-time minimum distance Δd between the equipment and personnel.

[0129] S52. Using the obtained real-time minimum distance Δd, the risk field model in the construction knowledge graph is invoked. The exponential decay risk calculation method is adopted, and after dimensionless processing, the construction risk probability Prisk is constructed. The formula is as follows:

[0130] Prisk = 1 - exp(-λ*Δd);

[0131] In the formula, λ represents the risk sensitivity coefficient;

[0132] The method for obtaining λ: By collecting minimum distance Δd data between equipment and personnel under different construction scenarios, and combining historical construction risk events and simulation results, the sensitivity of distance changes to the probability of safety accidents is statistically analyzed. The magnitude of risk change under different minimum distance thresholds is evaluated. Taking into account the personnel density, equipment operating radius and complexity of the working environment at the construction site, a reasonable risk sensitivity coefficient λ is determined. With reference to construction safety specifications, construction machinery operation safety standards and industry experience, and combined with the judgment of on-site safety management experts, an empirical coefficient is set to ensure that the calculation of the construction risk probability Prisk can effectively identify potential hazards without excessively triggering shutdown or warning strategies.

[0133] S53. By setting a preset construction risk probability threshold Pth, and comparing and analyzing the construction risk probability Prisk with the construction risk probability threshold Pth, the fourth assessment results are obtained, including:

[0134] When the construction risk probability Prisk < the construction risk probability threshold Pth, it indicates that the construction environment is safe, the risk is controllable, and continuous monitoring is required.

[0135] When the construction risk probability Prisk < the construction risk probability threshold Pth, it indicates that the construction environment is unsafe and the risk is uncontrollable. A risk warning message is immediately generated, and an automatic shutdown command is issued through the control bus. At the same time, a pop-up window is displayed on the monitoring interface, and a warning signal is issued in conjunction with the voice interaction terminal. In the shutdown state, the system enters the risk elimination detection process, tracking the change of Δd in real time. Based on the change value of Δd, the construction risk probability Prisk is calculated in real time until the construction risk probability Prisk < the construction risk probability threshold Pth. At this point, the shutdown command is automatically lifted, and the concrete placing boom operation is resumed.

[0136] The method for setting the construction risk probability threshold Pth is as follows: Statistical analysis of equipment and personnel location data at the construction site is used to extract the risk probability distribution range under safe and high-risk conditions. This, combined with the experience and judgment of construction safety management personnel, determines a reasonable construction risk assessment value. Reference is made to construction safety specifications, hazard assessment standards, and industry safety operating procedures, which typically provide acceptable thresholds for construction risk probabilities. This threshold is used to effectively distinguish whether the construction environment is safe, ensuring the safety of construction personnel and equipment.

[0137] In this embodiment, by monitoring the minimum distance between equipment and personnel in real time and calculating the probability of construction risks, a dynamic assessment of the safety of the construction environment is achieved. When the risk exceeds the threshold, the system can automatically issue an early warning, shut down the machine, and provide voice prompts to ensure the safety of construction personnel and reduce the accident rate. At the same time, the system can automatically resume work after the risk is eliminated, thereby improving the intelligence and reliability of construction safety management.

[0138] Example 10

[0139] This embodiment is an explanation based on Embodiment 9. Please refer to it. Figure 1 Specifically, step six includes:

[0140] S61. Throughout the construction process, the system establishes a full-link traceability mechanism for voice commands, executed actions, and equipment status; it uses voiceprint recognition technology to confirm the identity of operators and records and archives the control commands they issue along with the corresponding execution times, forming a traceable operation link.

[0141] S62. When quality defects or operational abnormalities occur during subsequent construction, the system can quickly trace back to the relevant operators and corresponding control commands, thus clarifying the division of responsibilities. At the same time, the traceability results will be fed back to the construction knowledge graph for the optimization of subsequent task processes and strategy improvement, thereby forming a closed loop of intelligent construction with controllable quality, clear responsibilities, and continuous optimization.

[0142] In this embodiment, a full-link traceability mechanism is established for operation instructions, execution actions, and equipment status to achieve traceable management of the construction process. When quality defects or abnormalities occur, the responsible personnel and corresponding operation instructions can be quickly identified, and the traceability information can be fed back to the construction knowledge graph to optimize subsequent task processes. This achieves intelligent construction closed-loop management with controllable construction quality, clear responsibilities, and continuous optimization.

[0143] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value, it is acceptable.

[0144] The above formulas are all derived from software simulation using a large amount of data, and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art based on the actual situation. The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A control method for intelligent construction equipment based on a multimodal large model, characterized in that, Includes the following steps: Step 1: Real-time monitoring of the multi-concrete placing boom collaborative pouring scenario at the construction site; Real-time acquisition of concrete pump flow and pressure data, construction area point cloud data, construction surface images and multi-angle views, as well as operator voice commands processed by noise suppression and voiceprint recognition; then the point cloud is fused with the BIM design model to generate a three-dimensional model of the construction area, and after data preprocessing, a standardized input dataset is established. Step 2: Input the standardized input dataset into the trained multimodal large model, determine the accuracy of the construction area, and perform semantic segmentation based on the 3D model and normalized video frames to obtain the actual pouring coverage area Sa and the BIM design area Sb. Calculate the regional granularity factor Garea and compare it with the coverage threshold Gth to determine whether the material coverage accuracy is qualified. If qualified, generate a local task label Ltask; if unqualified, provide an appropriate strategy. Step 3: Extract the flow rate data F(t) and pressure data P(t) of the corresponding feed pipe based on the local task label Ltask, calculate their standard deviation after time series normalization, and obtain the standard deviation of flow rate and pressure; further calculate the dynamic feeding stability coefficient Sdyn and compare it with the feeding threshold Sth to determine whether the feeding process is stable. If it is unstable, give an appropriate strategy. Step 4: Collect instantaneous flow and pressure signals of each concrete placing machine through sensors. After wireless network transmission and Kalman filtering normalization, a standardized flow and pressure sequence is obtained. The average and maximum values ​​are obtained by using exponential weighted moving average and extreme value search algorithms. The equipment load balancing coefficient Leq is calculated and compared with the load balancing threshold Lth to determine whether the concrete placing machine is operating in a balanced manner. If it is unbalanced, an appropriate strategy is given. Step 5: By collecting the real-time minimum distance Δd between the equipment and personnel, construct the construction risk probability Prisk, and compare it with the construction risk probability threshold Pth to determine whether the construction environment is safe. If it is not safe, give appropriate strategies. Step Six: The system establishes a full-link traceability mechanism for voice commands, executed actions, and equipment status. It confirms the operator's identity through voiceprint recognition and records the command and time. When quality defects or operational abnormalities occur, the system can quickly trace back to the relevant personnel and commands, clearly defining responsibilities. The traceability results are fed back to the construction knowledge graph for task process optimization and strategy improvement, forming a closed-loop management system with controllable quality, clear responsibilities, and continuous optimization.

2. The intelligent construction equipment control method based on a multimodal large model according to claim 1, characterized in that, Step one includes: S11. Real-time monitoring of the multi-concrete placing machine collaborative pouring scenario at the construction site; flow sensors and pressure sensors are installed at the inlet and end of the placing pipe of each placing machine at the construction site to collect the concrete pumping volume flow rate F(t) and the instantaneous pressure value P(t) of the pipeline per unit time in real time. S12 collects point cloud data of the construction surface and structural components by deploying lidar scanning equipment in the construction area, and matches and fuses the point cloud results with the BIM design model to generate a three-dimensional model M3D of the construction area. S13. Deploy panoramic cameras at high positions in the construction area and install wide-angle monitoring cameras in key passages to obtain two-dimensional image data and multi-angle coverage views of the construction surface in real time. S14. By deploying voice receiving terminals in the operator's work area at the construction site and configuring directional microphone arrays and noise suppression modules, the voice command signals of the operators are collected in real time. The collected voice data is first processed by adaptive filtering through a dynamic noise database, then the operator's identity code is extracted by a voiceprint recognition algorithm, and then the natural speech is converted into structured control commands, and the timestamp of the command issuance is recorded. S15. The collected data undergoes filtering and outlier removal, point cloud registration and noise suppression, image distortion correction and brightness equalization, and speech signal filtering and voiceprint recognition. It is then uniformly converted into a structured data format and timestamped, and a standardized input dataset is established.

3. The intelligent construction equipment control method based on a multimodal large model according to claim 2, characterized in that, Step two includes: S21. By inputting the standardized input dataset into the trained multimodal large model, the accuracy of the construction area within the coverage of the material placement pipe is determined. S22. Based on the 3D model of the construction area M3D, combined with the preprocessed normalized video frame data, perform image semantic segmentation on the construction surface; obtain the actual segmented pouring coverage area Sa and the design area Sb of the BIM model through the regional projection overlay method.

4. The intelligent construction equipment control method based on a multimodal large model according to claim 3, characterized in that, Step two also includes: S23. Match the actual pouring area Sa with the BIM model design area Sb, and after dimensionless processing, calculate and obtain the area granularity factor Garea. S24. By setting a preset coverage threshold Gth and comparing the region granularity factor Garea with the coverage threshold Gth, the first evaluation results are obtained, including: When the coverage threshold Gth ≤ the area particle size factor Garea ≤ the coverage threshold Gth * 120%, it indicates that the material coverage accuracy is qualified, the system recognizes that the pouring is consistent with the design, and generates a local task label Ltask in combination with the construction knowledge graph; When the area granularity factor Garea < the coverage threshold Gth, it indicates that the coverage accuracy of the material is unqualified, the coverage is insufficient, and there is a risk of missed pouring, triggering the first voice interaction correction process: the operator inputs the replenishment command through the voice receiving terminal, and the command is sent to the trained multimodal large model in the industrial control computer; the multimodal large model calls the dynamic noise database to perform adaptive noise suppression on the on-site voice and improve the clarity of the command; the multimodal large model converts the voice into structured commands through end-to-end semantic parsing, including: reducing the moving speed of the material pipe by 10% and increasing the current area's material placement time by 8%, while highlighting the insufficient area on the monitoring interface; When the area granularity factor Garea > coverage threshold Gth * 120%, it indicates that the fabric coverage accuracy is unqualified, the coverage is excessive, and there is a risk of over-pouring. This triggers the second voice interaction correction process: the operator inputs the "reduce" command, which is converted into a structured command after model noise suppression and semantic parsing. This command includes: adjusting the swing amplitude of the fabric tube and limiting the local flow rate, performing dynamic correction of the flow rate and movement amplitude, and gradually reducing the risk of over-coverage until the coverage threshold Gth ≤ area granularity factor Garea ≤ coverage threshold Gth * 120%.

5. The intelligent construction equipment control method based on a multimodal large model according to claim 4, characterized in that, Step three includes: S31. Based on the local task label Ltask, by calling the fabric pipe operation data corresponding to the local task label Ltask in the standardized input dataset, extract the real-time flow data F(t) and real-time pressure data P(t) of the fabric pipe, perform time series normalization processing to obtain the flow fluctuation sequence and pressure fluctuation sequence, calculate the standard deviation of the flow fluctuation sequence and the pressure fluctuation sequence respectively, and obtain the flow standard deviation σF and the pressure standard deviation σP.

6. The intelligent construction equipment control method based on a multimodal large model according to claim 5, characterized in that, Step three also includes: S32. After dimensionless processing of the obtained flow rate standard deviation σF and pressure standard deviation σP, the dynamic feeding stability coefficient Sdyn is calculated. S33. By setting a preset material supply stability threshold Sth, and comparing and analyzing the dynamic material supply stability coefficient Sdyn with the material supply stability threshold Sth, the second evaluation results are obtained, including: When the dynamic feeding stability coefficient Sdyn ≤ feeding stability threshold Sth, it indicates that the feeding process is stable and there is no risk of fluctuation in the fabric operation status, proceed to step four; When the dynamic feeding stability coefficient Sdyn > the feeding stability threshold Sth, it indicates that the feeding process is unstable and there is a risk of fluctuation in the fabric operation status, triggering the third voice interaction correction process: the correction command is input through the voice receiving terminal and sent to the trained multimodal large model in the industrial control computer; after the multimodal large model performs adaptive noise suppression in combination with the dynamic noise database, it performs semantic parsing and converts natural language into structured commands, including a 5%–10% reduction in the fabric tube speed and a local path correction offset of 5% of the design path; the system issues correction commands and updates the flow and pressure acquisition sequences in real time, recalculates, until the dynamic feeding stability coefficient Sdyn ≤ the feeding stability threshold Sth.

7. The intelligent construction equipment control method based on a multimodal large model according to claim 6, characterized in that, Step four includes: S41. By installing flow sensors and pressure sensors at the inlet and end of the placing pipe of each placing boom at the construction site, the instantaneous flow rate F of each placing boom is collected in real time. i With pipeline pressure P j The collected flow and pressure signals are transmitted to the industrial control computer via a wireless sensor network. Kalman filtering and normalization algorithms are then used to denoise and standardize the data, resulting in a standardized flow sequence {F1, F...}. ′ 2, ..., F ′ n } and standardized pressure sequence {P1 ′ P2 ′ ..., P ′ m }; S42. Based on the acquired standardized flow rate sequence and standardized pressure sequence, the average flow rate mean(F) of each placing boom is obtained using an exponentially weighted moving average algorithm. i ′ ) and mean pressure (P) j ′ The maximum flow rate (max(F)) is obtained through an extreme value search algorithm with threshold constraints. i ′ ) and maximum pressure max(P j ′ ).

8. The intelligent construction equipment control method based on a multimodal large model according to claim 7, characterized in that, Step four also includes: S43, the average flow rate mean(F) of each fabric placing machine is obtained. i ′ Mean pressure (P) j ′ ), maximum flow rate max(F) i ′ ) and maximum pressure max(P j ′ After dimensionless processing, the equipment load balancing coefficient Leq is calculated and obtained. S44. By comparing and analyzing the device load balancing coefficient Leq with the load balancing threshold Lth using a preset load balancing threshold, the third evaluation result is obtained, including: When the equipment load balancing coefficient Leq ≤ load balancing threshold Lth, it indicates that the concrete placing machine is operating in a balanced state, and monitoring should continue. When the equipment load balancing coefficient Leq > the load balancing threshold Lth, it indicates that the concrete placing machine is out of balance and the load distribution is uneven, triggering the fourth voice interaction correction process: the system calls the path planning algorithm to redistribute the concrete placing sequence and generate a flexible work order; the system dynamically adjusts the concrete placing speed and operation sequence of the overloaded equipment in the scheduling results to keep the overall construction progress stable; the correction results are pushed in real time through the monitoring interface and the operator is prompted to confirm on the voice interaction terminal.

9. The intelligent construction equipment control method based on a multimodal large model according to claim 8, characterized in that, Step five includes: S51. By deploying lidar and ultrasonic sensors within the operating radius of the concrete placing boom at the construction site, the spatial position data between the equipment and personnel is collected in real time. Combined with a standardized input dataset, the collected spatial distance information is filtered and normalized to obtain the real-time minimum distance Δd between the equipment and personnel. S52. By obtaining the real-time minimum distance Δd, the risk field model in the construction knowledge graph is called, and the exponential decay risk calculation method is used. After dimensionless processing, the construction risk probability Prisk is constructed. S53. By setting a preset construction risk probability threshold Pth, and comparing and analyzing the construction risk probability Prisk with the construction risk probability threshold Pth, the fourth assessment results are obtained, including: When the construction risk probability Prisk < the construction risk probability threshold Pth, it indicates that the construction environment is safe, the risk is controllable, and continuous monitoring is required. When the construction risk probability Prisk < the construction risk probability threshold Pth, it indicates that the construction environment is unsafe and the risk is uncontrollable. A risk warning message is immediately generated, and an automatic shutdown command is issued through the control bus. At the same time, a pop-up window is displayed on the monitoring interface, and a warning signal is issued in conjunction with the voice interaction terminal. In the shutdown state, the system enters the risk elimination detection process, tracking the change of Δd in real time. Based on the change value of Δd, the construction risk probability Prisk is calculated in real time until the construction risk probability Prisk < the construction risk probability threshold Pth. At this point, the shutdown command is automatically lifted, and the concrete placing boom operation is resumed.

10. The intelligent construction equipment control method based on a multimodal large model according to claim 9, characterized in that, Step six includes: S61. Throughout the construction process, the system establishes a full-link traceability mechanism for voice commands, executed actions, and equipment status; it uses voiceprint recognition technology to confirm the identity of operators and records and archives the control commands they issue along with the corresponding execution times, forming a traceable operation link. S62. When quality defects or operational abnormalities occur during subsequent construction, the system can quickly trace back to the relevant operators and corresponding control commands, thus clarifying the division of responsibilities. At the same time, the traceability results will be fed back to the construction knowledge graph for the optimization of subsequent task processes and strategy improvement, thereby forming a closed loop of intelligent construction with controllable quality, clear responsibilities, and continuous optimization.

Images