AI-based building engineering construction progress optimization system

Abstract

The invention relates to the field of building construction supervision, in particular to an AI-based building engineering construction progress optimization system, which comprises an environment sensing module, an edge calculation module, a progress inversion module, a material prediction module and a construction progress prediction module. According to the invention, the construction progress of each layer is monitored in real time, and the construction condition is tracked in all directions through the environment sensing module (such as dust concentration, personnel distribution, voiceprint characteristics and the like), so that the accurate mastering of the construction progress is ensured. The progress inversion module effectively reversely deduces the construction progress through an improved Transformer time sequence model based on data such as the dust change rate and the personnel density index, the error is controlled within 3%, the accuracy and operability of progress control are further improved, the material prediction module predicts the material requirement of the next stage in advance by analyzing the construction progress, and the construction progress control efficiency is improved. And a material delivery instruction is sent to the automatic equipment in real time through the intelligent regulation and control interface.

Description

Technical Field

[0001] The present invention belongs to the field of construction supervision, and specifically relates to an AI-based construction project progress optimization system. Background Art

[0002] With the rapid development of science and technology, especially the continuous advancement of artificial intelligence (AI) technology, the construction industry is undergoing profound changes. As the scale and complexity of construction projects continue to expand, higher requirements are placed on the accuracy and efficiency of construction progress management.

[0003] Traditional construction project construction progress management often relies on manual monitoring and experience-based judgment. This method is not only inefficient, but also difficult to achieve real-time monitoring and accurate prediction. When the construction progress optimization system in the existing technology is used, it is usually difficult to monitor the construction progress of each floor of the building in real time, and it is impossible to monitor the lack of consumables for the construction of each floor. It is impossible to quickly adjust the overall construction progress according to the progress of each floor, and further improvement is needed.

[0004] Therefore, an AI-based construction project construction progress optimization system is proposed, which can monitor the construction progress of each floor of the building in real time, monitor the lack of consumables for the construction of each floor, and quickly adjust the overall situation according to the construction progress of each floor. Summary of the invention

[0005] In order to overcome the problems that the construction progress optimization system in the prior art is usually difficult to monitor the construction progress of each floor of the building in real time, cannot monitor the lack of consumables for the construction of each floor, and cannot quickly adjust the overall construction progress according to the construction progress of each floor, therefore, an AI-based construction project construction progress optimization system is proposed.

[0006] The technical solution of the present invention is: an AI-based construction project construction progress optimization system, including an environment perception module, an edge computing module, a progress inversion module, a material prediction module and a construction progress prediction module;

[0007] The environmental perception module is used to detect the dust concentration in each construction area, the distribution density and activity intensity of construction personnel, and the voiceprint characteristics generated at the construction site;

[0008] The edge computing module includes a process mapping model and a thermal imaging analysis model. The personnel density index of each area is calculated through the process mapping model and the thermal imaging analysis model, and the construction surface expansion rate is inferred by combining historical data.

[0009] The progress inversion module combines the data calculated by the edge computing module with the time series change rate of dust concentration, the thermal imaging personnel distribution heat map and the voiceprint feature classification results to output the quantitative value of the construction progress of each floor;

[0010] The material prediction module predicts the classification results of the voiceprint features of the edge computing module and triggers the material pre-delivery instruction for the next process;

[0011] The construction progress prediction module predicts the construction progress based on the data of the edge computing module and the data of the material prediction module;

[0012] The data of each module is synchronized using the NTP protocol, and the time deviation ≤ 100 ms.

[0013] Preferably, the environmental perception module: consists of a distributed sensor network deployed in each layer of the construction area, including: a. a laser dust sensor array (PM2.5 / PM10 dual-mode), arranged at a grid density of 5m × 5m, real-time monitoring the dust concentration distribution of each operation surface, integrating a barometric compensation algorithm to eliminate wind speed interference; b. an infrared thermal imaging camera group, identifying the distribution density and activity intensity of construction personnel through the human body's thermal radiation characteristics, densely deployed at a grid of 2.5m × 2.5m at the wall corners, and compensating for data differences through transfer learning; c. a sound spectrum analyzer, collecting the voiceprint features of construction machinery, classifying and identifying the current construction process and then outputting, outputting standardized JSON format data (including voiceprint tags, signal-to-noise ratio, ±1ms accuracy timestamp), with a built-in noise filtering algorithm.

[0014] Preferably, the edge computing module includes an embedded AI chip integrated in each layer's distribution box: the AI chip embeds a local path planning algorithm, and the robot only receives a sequence of coordinate points instead of real-time control signals. The AI performs: a. constructing a dust concentration - process mapping model, and when a specific dust peak is detected (such as PM10 > 200 μg / m 3 lasting for 5 minutes), it is determined as a concrete grinding operation; b. through clustering analysis of thermal imaging data, calculating the personnel density index of each area, and combining historical data to reverse-infer the development rate of the construction surface.

[0015] Preferably, the progress inversion module includes an improved Transformer time series model; the inputs include: a. the time series change rate of dust concentration (ΔPM / Δt); b. the thermal imaging personnel distribution heat map: including the classification results of voiceprint features as discrete process labels (such as "reinforcement cutting_01", "concrete vibration sound_01") and outputting the quantified construction progress value (0 - 100%) of each layer, with an error rate ≤ 3%; when the environmental wind speed > 5m / s or ≥ 3 types of voiceprint features are detected simultaneously, the allowable error rate is temporarily relaxed to 5%. The progress inversion module needs to cross-verify the process time nodes in the BIM model, and if the deviation > 5%, it will initiate an artificial review process.

[0016] Preferably, based on the progress inversion results, the material prediction module establishes a relationship map of material consumption for each process. When the progress of a certain layer reaches the critical threshold (such as 60% for steel structure installation), it automatically triggers the material pre-delivery instruction for the next process. Adaptive control interface: Sends adjustment instructions to the automatic plastering robot through LoRa wireless network or 5GNR-Light. Low-frequency instructions (such as path planning) use LoRa, and high-frequency control (such as emergency braking) switches to 5GNR-Light (air interface delay < 10ms). After the delivery instruction is triggered, it is necessary to reconfirm whether the personnel have reached the target area through thermal imaging data and dynamically reconstruct the construction path. The calculation formula for the threshold is: Threshold = α × baseline progress + β × average historical deviation, where α = 0.8 - 1.2 (process type coefficient) and β = 0.05 - 0.15 (fault tolerance coefficient).

[0017] Preferably, the construction progress prediction module combines the data of the edge computing module and the data of the material prediction module, and uses machine learning algorithms to predict the overall construction progress and generate progress optimization suggestions. The progress optimization suggestions include resource allocation optimization suggestions and abnormal condition detection suggestions.

[0018] Preferably, the formula for the change rate of dust concentration in the Transformer time series model of the progress inversion module is:

[0019] Where PM t and PM t+1 represent the dust concentrations at time t and time t + 1 respectively, and Δt is the time difference;

[0020] The formula for the personnel activity density index is: Where, Pi represents the number of personnel in the i-th area, ai is the posture coefficient of area i, and ti is the working duration of the personnel in this area. The posture coefficient ai of each area is given according to the work type, 1.0 for standing work, 0.8 for bending installation, and 0.6 for handling and moving;

[0021] The progress inversion calculation formula is: S layer = f(ΔPM, D, V sound , Δt)

[0022] Where, S layer represents the construction progress of a certain layer (0 - 100%), f is the prediction model function based on environmental data and sound characteristics, ΔPM is the change rate of dust concentration, D is the personnel activity density index, and V sound is the voiceprint feature data.

[0023] Preferably, the calculation formula for the material consumption rate is: M used = f(S layer × t), where, M usedRepresents the consumption of materials at time point t, S layer is the current construction progress, and f is the consumption function related to the process.

[0024] The calculation formula for triggering material pre-delivery is: M threshold = M used × thresholdfactor, where M threshold is the threshold of the material. When the material consumption in a certain progress stage approaches or exceeds this threshold, the system will automatically trigger the material delivery in the next stage.

[0025] Preferably, the formula for predicting the construction progress is: where S forecast is the prediction of the construction progress within the next 72 hours, W i is the weighting coefficient, and X i is the input data for multiple task predictions (such as personnel density, dust concentration, voiceprint characteristics, etc.);

[0026] The formula for optimizing resource allocation is: R adjusted = R current × (1 - delayfactor), where R adjusted is the adjusted resource allocation amount, R current is the current resource allocation amount, and delayfactor is the adjustment factor generated based on the lag of the construction progress;

[0027] The formula for detecting abnormal working conditions is: E anomaly = [PM mean - PM threshold ∪ [D discrepancy > ], where E anomaly represents the determination of the occurrence of abnormal situations, PM mean is the monitored average dust concentration, PM threshold is the dust concentration threshold, and D discrepancy is the dispersion degree of the personnel density. When E anomaly is greater than the set reference value, the staff needs to go to the site immediately to check the relevant floors or activate the drone to fly to the relevant floor height to take pictures of the interior of the floor.

[0028] The present invention also provides a method for using an AI-based construction project construction progress optimization system, and the steps are as follows:

[0029] S1: Sensor deployment: Deploy laser dust sensors in a 5m × 5m grid in each floor construction area, and install infrared thermal imaging cameras and sound spectrum analyzers;

[0030] S2: Data initialization: Import the logical relationship of construction processes and material consumption reference parameters through the BIM model;

[0031] S3: Environmental data collection: Real-time collection of dust concentration, thermal imaging personnel distribution data, and construction machinery voiceprint characteristics for each floor;

[0032] S4: Edge computing processing: Executed by an embedded AI chip: Dust concentration - process mapping analysis (detecting PM10 peak value to determine grinding operation) and clustering of thermal imaging data to generate personnel density index;

[0033] S5: Progress inversion calculation: Input the dust change rate, personnel density index D, and voiceprint classification result into the Transformer model, and output the progress value for each floor;

[0034] S6: Material demand prediction: Trigger the material pre-delivery instruction according to the progress value. When the steel structure installation progress of a certain floor reaches 60%, automatically generate the delivery requirements for bolts and connectors;

[0035] S7: Construction path optimization: Send adjustment instructions to the automatic plastering robot through LoRa wireless network or 5GNR-Light. Low-frequency instructions (such as path planning) use LoRa, and high-frequency control (such as emergency braking) switches to 5GNR-Light (air interface delay < 10ms) to send path update instructions to the automatic plastering robot to avoid high personnel density areas;

[0036] S8: Progress prediction and early warning: Use the random forest algorithm to predict the progress in the next 72 hours. When the deviation exceeds 8%, trigger a level-three early warning;

[0037] Level-three early warning classification standard:

[0038] Level-one early warning (deviation 8% - 12%): Automatically adjust the delayfactor in the resource allocation formula;

[0039] Level-two early warning (deviation 12% - 15%): Freeze material delivery and initiate manual review;

[0040] Level-three early warning (deviation > 15%): Forcefully switch to the conservative prediction mode + drone inspection;

[0041] S9: Dynamic resource allocation: Adjust the personnel and equipment configuration for each floor according to the formula (1 - delayfactor);

[0042] S10: Abnormal condition handling: When it is detected that >> the threshold value, it is determined as abnormal construction congestion. The staff needs to immediately go to the site to check the relevant floors. If the threshold value is continuously detected 3 times in the same area and there is no manual response, enable the drone to fly to the relevant floor height to take videos of the floor;

[0043] S11: Data closed-loop update: Feed the actual progress data back to the BIM model and dynamically feedback to the dynamic weight adjustment of the edge computing module to form a data closed-loop, and optimize the prediction parameters of subsequent processes.

[0044] Advantages of the present invention:

[0045] 1. Real-time monitoring and precise control: The system can monitor the construction progress of each layer in real time, and comprehensively track the construction situation through the environmental perception module (such as dust concentration, personnel distribution, voiceprint characteristics, etc.) to ensure accurate grasp of the construction progress. The progress inversion module effectively inverses the construction progress based on data such as dust change rate and personnel density index through an improved Transformer time series model, and the error is controlled within 3%, further improving the accuracy and operability of progress control;

[0046] 2. Optimization of material requirements and distribution: The material prediction module analyzes the construction progress, predicts the material requirements for the next stage in advance, and sends material distribution instructions to automated equipment in real time through an intelligent control interface. This intelligent material distribution system can significantly reduce construction delays caused by material shortages or overages, ensure that construction materials can arrive accurately when needed, and improve construction efficiency and resource utilization rate;

[0047] 3. Optimization of construction path and personnel allocation: By combining the edge computing module and thermal imaging data, the system can analyze the personnel distribution and activity intensity in the construction area in real time, automatically optimize the construction path, and adjust the path or operation content of automated equipment through LoRa and 5GNR-Light when necessary to avoid operations in crowded areas and reduce construction safety hazards. In addition, by dynamically allocating resources and making timely adjustments according to the lag of the construction progress, the continuous progress of the project is guaranteed;

[0048] 4. Abnormality detection and early warning mechanism: The system can quickly identify possible problems such as congestion and delays during the construction process through intelligent abnormality detection algorithms, and make timely responses through a three-level early warning mechanism. For example, when the construction progress deviation exceeds the set threshold, the system will automatically adjust the resource configuration, start the manual review process or conduct drone inspections to ensure that abnormal situations at the construction site can be quickly handled and avoid major delays caused by construction obstacles;

[0049] 5. AI-based intelligent prediction and optimization: The construction progress prediction module combines multi-dimensional data (such as personnel density, dust concentration, voiceprint characteristics, etc.) and machine learning algorithms to accurately predict the construction progress within the next 72 hours. Through the prediction results, the system can identify potential resource bottlenecks or progress deviations in advance, optimize the construction process, and make comprehensive resource configurations and emergency plans in the early stage of construction to improve the overall efficiency and response ability of construction; 6. Data Closed - Loop and Continuous Optimization: The system forms a data closed - loop through the feedback mechanism of the BIM model and the edge - computing module, enabling the construction progress prediction to be continuously updated and optimized. This further improves the intelligence and adaptability of the system. By continuously adjusting various construction parameters and prediction models through feedback, the system can effectively cope with complex and dynamic construction environments, improving the refinement and intelligence levels of construction management. Brief Description of the Drawings

[0050] Figure 1 The flowchart of the present invention is shown. Detailed Embodiment

[0051] The present invention will be further described below in conjunction with the drawings and embodiments.

[0052] Please refer to Figure 1 , the present invention provides an embodiment: an AI - based construction progress optimization system for building engineering, including an environment perception module, an edge - computing module, a progress inversion module, a material prediction module, and a construction progress prediction module;

[0053] The environment perception module is used to detect the dust concentration, the distribution density and activity intensity of construction workers, and the voiceprint characteristics generated in the construction site in each construction area.

[0054] The edge - computing module includes a process mapping model and a thermal imaging analysis model. The personnel density index of each area is calculated through the process mapping model and the thermal imaging analysis model, and the construction surface expansion rate is deduced by combining historical data.

[0055] The progress inversion module combines the data calculated by the edge - computing module with the time - series change rate of dust concentration, the thermal imaging personnel distribution heat map, and the voiceprint feature classification result, and outputs the quantified value of the construction progress of each layer.

[0056] The material prediction module predicts the voiceprint feature classification result of the edge - computing module and triggers the pre - distribution instruction of materials for the next process.

[0057] The construction progress prediction module predicts the construction progress based on the data of the edge - computing module and the data of the material prediction module.

[0058] The data of each module is synchronized using the NTP protocol, and the time deviation ≤ 100 ms.

[0059] Preferably, the environmental perception module is composed of a distributed sensor network deployed in each layer of the construction area, including: a. a laser dust sensor array (PM2.5 / PM10 dual-mode), arranged at a grid density of 5m×5m, to monitor the dust concentration distribution in each operation area in real time, and integrated with a barometric compensation algorithm to eliminate wind speed interference; b. an infrared thermal imaging camera group, to identify the distribution density and activity intensity of construction workers through the characteristics of human thermal radiation, densely deployed at a grid of 2.5m×2.5m at the wall corners, and compensating for data differences through transfer learning; c. a sound spectrum analyzer, to collect the acoustic fingerprint characteristics of construction machinery, classify and identify the current construction process and then output, outputting standardized JSON format data (including acoustic fingerprint tags, signal-to-noise ratio, ±1ms precision timestamp), with a built-in noise filtering algorithm.

[0060] Preferably, the edge computing module includes an embedded AI chip integrated in the distribution box of each layer: the AI chip embeds a local path planning algorithm, and the robot only receives a sequence of coordinate points instead of real-time control signals. The AI performs: a. constructing a dust concentration - process mapping model, and determining it as a concrete grinding operation when a specific dust peak is detected (such as PM10>200μg / m 3 lasting for 5 minutes); b. through clustering analysis of thermal imaging data, calculating the personnel density index of each area, and inversely inferring the development rate of the construction surface in combination with historical data.

[0061] Preferably, the progress inversion module includes an improved Transformer time series model; the inputs include: a. the time series change rate of dust concentration (ΔPM / Δt); b. the thermal imaging personnel distribution heat map: including the classification result of acoustic fingerprint characteristics as discrete process tags (such as "steel bar cutting_01", "concrete vibration sound_01") and outputting the quantitative value of the construction progress of each layer (0 - 100%), with an error rate ≤3%; when the environmental wind speed > 5m / s or ≥3 types of acoustic fingerprint characteristics are detected simultaneously, the allowable error rate is temporarily relaxed to 5%. The progress inversion module needs to cross-verify the process time nodes in the BIM model. If the deviation > 5%, the manual review process is initiated.

[0062] Preferably, based on the progress inversion result, the material prediction module establishes a relationship map of material consumption in the process. When the progress of a certain layer reaches the critical threshold (such as 60% of steel structure installation), it automatically triggers the material pre-delivery instruction for the next process. Adaptive control interface: sends adjustment instructions to the automatic plastering robot through LoRa wireless network or 5GNR-Light. Low-frequency instructions (such as path planning) use LoRa, and high-frequency control (such as emergency braking) switches to 5GNR-Light (air interface delay < 10ms). After the delivery instruction is triggered, it is necessary to reconfirm whether the personnel have reached the target area through thermal imaging data and dynamically reconstruct the construction path; the calculation formula for the threshold is: threshold = α × reference progress + β × average historical deviation, where α = 0.8 - 1.2 (process type coefficient) and β = 0.05 - 0.15 (fault tolerance coefficient).

[0063] Preferably, the construction progress prediction module combines the data of the edge computing module and the data of the material prediction module, and uses machine learning algorithms to predict the overall construction progress and generate progress optimization suggestions. The progress optimization suggestions include resource allocation optimization suggestions and abnormal working condition detection suggestions.

[0064] Preferably, the formula for the change rate of dust concentration in the Transformer time series model of the progress inversion module is:

[0065] where PM t and PM t+1 represent the dust concentrations at time t and time t + 1 respectively, and Δt is the time difference;

[0066] The formula for the personnel activity density index is: where Pi represents the number of personnel in the i-th area, ai is the posture coefficient of area i, and ti is the working duration of the personnel in this area. The posture coefficient ai of each area is given according to the work type, 1.0 for standing operation, 0.8 for bending installation, and 0.6 for handling and moving;

[0067] The progress inversion calculation formula is: S layer = f(ΔPM, D, V sound , Δt)

[0068] where S layer represents the construction progress (0 - 100%) of a certain layer, f is the prediction model function based on environmental data and sound characteristics, ΔPM is the change rate of dust concentration, D is the personnel activity density index, and V sound is the voiceprint feature data.

[0069] Preferably, the calculation formula for the material consumption rate is: M used = f(S layer × t), where M used

[0070] Indicates the consumption of materials at time point t, S layer is the current construction progress, and f is the consumption function related to the process.

[0071] The calculation formula for triggering material pre-delivery is: M threshold = M used × thresholdfactor, where M threshold is the threshold of the material. When the material consumption in a certain progress stage approaches or exceeds this threshold, the system will automatically trigger the material delivery in the next stage.

[0072] Preferably, the formula for predicting the construction progress is: where S forecast is the predicted construction progress within the next 72 hours, W i is the weighting coefficient, and X i is the input data for multiple task predictions (such as personnel density, dust concentration, voiceprint characteristics, etc.);

[0073] The formula for optimizing resource allocation is: R adjusted = R current × (1 - delayfactor), where R adjusted is the adjusted resource allocation amount, R current is the current resource allocation amount, and delayfactor is the adjustment factor generated based on the lag of the construction progress;

[0074] The formula for detecting abnormal working conditions is: E anomaly = [PM mean - PM threshold ∪ [D discrepancy > ], where E anomaly indicates the determination of the occurrence of abnormal situations, PM mean is the measured average dust concentration, PM threshold is the dust concentration threshold, and D discrepancy is the dispersion degree of the personnel density. When E anomaly is greater than the set reference value, the staff needs to go to the site immediately to check the relevant floors or enable the drone to fly to the relevant floor height to take pictures of the interior of the floor.

[0075] The present invention also provides a usage method of an AI-based construction progress optimization system for construction projects, and the steps are as follows:

[0076] S1: Sensor deployment: Deploy laser dust sensors in a 5m × 5m grid in each construction area, and install infrared thermal imaging cameras and sound spectrum analyzers;

[0077] S2: Data initialization: Import the logical relationships of construction processes and benchmark parameters of material consumption through the BIM model;

[0078] S3: Environmental data collection: Real-time collect the dust concentration of each floor, the thermal imaging personnel distribution data, and the acoustic fingerprint characteristics of construction machinery;

[0079] S4: Edge computing processing: Execute through an embedded AI chip: dust concentration - process mapping analysis (detect the PM10 peak value to determine the grinding operation) and thermal imaging data clustering to generate a personnel density index;

[0080] S5: Progress inversion calculation: Input the dust change rate, the personnel density index D, and the acoustic fingerprint classification result into the Transformer model, and output the progress value of each floor;

[0081] S6: Material demand prediction: Trigger the material pre-delivery instruction according to the progress value. When the steel structure installation progress of a certain floor reaches 60%, automatically generate the delivery requirements for bolts and connectors;

[0082] S7: Construction path optimization: Send adjustment instructions to the automatic plastering robot through the LoRa wireless network or 5GNR-Light. For low-frequency instructions (such as path planning), use LoRa, and for high-frequency control (such as emergency braking), switch to 5GNR-Light (air interface delay < 10ms), and send path update instructions to the automatic plastering robot to avoid high personnel density areas;

[0083] S8: Progress prediction and early warning: Use the random forest algorithm to predict the progress in the next 72 hours, and trigger a level 3 early warning when the deviation exceeds 8%;

[0084] Level 3 early warning classification standard:

[0085] Level 1 early warning (deviation 8% - 12%): Automatically adjust the delayfactor in the resource allocation formula;

[0086] Level 2 early warning (deviation 12% - 15%): Freeze material delivery and initiate manual review;

[0087] Level 3 early warning (deviation > 15%): Forcefully switch to the conservative prediction mode + drone inspection.

[0088] S9: Dynamic resource allocation: Adjust the personnel and equipment configuration of each floor according to the formula (1 - delayfactor);

[0089] S10: Abnormal working condition handling: When detecting >> the threshold value, it is determined as abnormal construction congestion. The staff needs to immediately go to the site to check the relevant floors. If the threshold value is continuously detected 3 times in the same area and there is no manual response, enable the drone to fly to the relevant floor height to take pictures of the floor;

[0090] S11: Data closed-loop update: Feed the actual progress data back to the BIM model and dynamically feedback to the edge computing module for dynamic weight adjustment to form a data closed-loop, and optimize the prediction parameters of subsequent processes.

[0091] Through the above steps, the system can monitor the construction progress of each floor in real time, comprehensively track the construction situation through the environmental perception module (such as dust concentration, personnel distribution, voiceprint characteristics, etc.), ensure the accurate grasp of the construction progress. The progress inversion module, based on data such as the dust change rate and personnel density index, effectively inversely deduces the construction progress through an improved Transformer time series model, with the error controlled within 3%, further improving the accuracy and operability of progress control;

[0092] The material prediction module analyzes the construction progress, predicts the material requirements for the next stage in advance, and sends material distribution instructions to the automated equipment in real time through the intelligent control interface. This intelligent material distribution system can significantly reduce construction delays caused by material shortages or overages, ensure that construction materials can be accurately in place when needed, and improve construction efficiency and resource utilization rate;

[0093] By combining the edge computing module and thermal imaging data, the system can analyze the personnel distribution and activity intensity in the construction area in real time, automatically optimize the construction path, and adjust the path or operation content of the automated equipment through LoRa and 5GNR-Light when necessary, avoid operations in crowded areas, and reduce construction safety hazards. In addition, through dynamic resource allocation and timely adjustment according to the lag of the construction progress, the continuous progress of the project is guaranteed.

[0094] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the purpose of the present invention.

Claims

1. An AI-based construction project construction progress optimization system, including an environment perception module, an edge computing module, a progress inversion module, a material prediction module and a construction progress prediction module; characterized by: The environmental perception module is used to detect the dust concentration in each construction area, the distribution density and activity intensity of construction personnel, and the voiceprint characteristics generated at the construction site; The edge computing module includes a process mapping model and a thermal imaging analysis model. Through the process mapping model and the thermal imaging analysis model, the personnel density index of each area is calculated, and the construction surface expansion rate is inferred by combining historical data. The progress inversion module combines the data calculated by the edge computing module with the dust concentration time series change rate, the thermal imaging personnel distribution heat map and the voiceprint feature classification results to output the quantitative value of the construction progress of each layer. The material prediction module predicts the voiceprint feature classification results of the edge computing module and triggers the material pre-delivery instruction for the next process; The construction progress prediction module predicts the construction progress based on the data from the edge computing module and the data from the material prediction module; The data of each module is synchronized using the NTP protocol, and the time deviation is ≤100ms.

2. The AI-based construction project construction progress optimization system according to claim 1 is characterized by: Environmental perception module: It is composed of a distributed sensor network deployed in each construction area, including: a. Laser dust sensor array (PM2.5 / PM10 dual-mode), arranged in a 5m×5m grid density, monitors the dust concentration distribution of each work surface in real time, and integrates an air pressure compensation algorithm to eliminate wind speed interference; b. Infrared thermal imaging camera group, identifies the distribution density and activity intensity of construction personnel through human thermal radiation characteristics, and is densely deployed at the corners of the wall to a 2.5m×2.5m grid, and compensates for data differences through transfer learning; c. Sound spectrum analyzer, collects the voiceprint features of construction machinery and classifies and identifies the current construction process before outputting, outputs standardized JSON format data (including voiceprint labels, signal-to-noise ratio, ±1ms precision timestamp), and has a built-in noise filtering algorithm.

3. The AI-based construction project construction progress optimization system according to claim 1 is characterized by: The edge computing module includes an embedded AI chip integrated in the distribution box on each floor: the AI ​​chip is embedded in the local path planning algorithm, the robot only receives the coordinate point sequence instead of the real-time control signal, and the AI ​​performs: a. Build a dust concentration-process mapping model, and when a specific dust peak is detected (such as PM10>200μg / m 3 If the grinding process lasts for 5 minutes, it is considered as concrete grinding operation; b. Calculate the personnel density index of each area through cluster analysis of thermal imaging data, and infer the construction surface expansion rate based on historical data.

4. The AI-based construction project construction progress optimization system according to claim 1 is characterized by: The progress inversion module includes the use of an improved Transformer time series model; the input includes: a. the time series change rate of dust concentration (ΔPM / Δt); b. the thermal imaging personnel distribution heat map: including the classification results of voiceprint features as discrete process labels (such as "rebar cutting_01", "concrete vibration sound_01") and the output of the quantitative value of the construction progress of each layer (0100%), with an error rate of ≤3%; when the ambient wind speed is greater than 5m / s or ≥3 types of voiceprint features are detected at the same time, the error rate is temporarily relaxed to 5%. The progress inversion module needs to cross-verify the process time nodes in the BIM model. If the deviation is greater than 5%, the manual review process is started.

5. The AI-based construction project construction progress optimization system according to claim 1 is characterized by: The material prediction module establishes a process material consumption relationship map based on the progress inversion results. When the progress of a certain layer reaches the critical threshold (such as 60% of the steel structure installation), it automatically triggers the material pre-delivery instruction for the next process. The adaptive control interface sends adjustment instructions to the automatic plastering robot through the LoRa wireless network or 5GNR-Light. Low-frequency instructions (such as path planning) use LoRa, and high-frequency control (such as emergency braking) switches to 5GNR-Light (air interface delay <10ms). After the delivery instruction is triggered, it is necessary to confirm whether the personnel have reached the target area through thermal imaging data for the second time, and dynamically reconstruct the construction path; the threshold calculation formula is: threshold = α × benchmark progress + β × historical deviation mean, where α = 0.8-1.2 (process type coefficient), β = 0.05-0.15 (fault tolerance coefficient).

6. The AI-based construction project construction progress optimization system according to claim 1 is characterized by: The construction progress prediction module combines the data from the edge computing module and the data from the material prediction module, uses a machine learning algorithm to predict the overall construction progress, and generates progress optimization suggestions, which include resource allocation optimization suggestions and abnormal working condition detection suggestions.

7. The AI-based construction project construction progress optimization system according to claim 4 is characterized by: The dust concentration change rate formula of the Transformer time series model of the progress inversion module is: PM t and PM t+1 They represent the dust concentration at time t and time t+1 respectively, and Δt is the time difference; The formula for the human activity density index is: Among them, Pi represents the number of people in the i-th area, ai is the posture coefficient of area i, and ti is the working time of the person in the area. The posture coefficient ai of each area is given according to the type of work, with standing work being 1.0, bending installation being 0.8, and carrying and moving being 0.6; The calculation formula for progress inversion is: S layer =f(ΔPM,D,V sound ,Δt) Among them, S layer represents the construction progress of a certain layer (0-100%), f is the prediction model function based on environmental data and sound characteristics, ΔPM is the dust concentration change rate, D is the personnel activity density index, V sound It is the voiceprint feature data.

8. The AI-based construction project construction progress optimization system according to claim 5 is characterized by: The material consumption rate is calculated as: M used =f(S layer ×t), where M used represents the material consumption at time t, S layer is the current construction progress, and f is the consumption function related to the process. The calculation formula for material pre-delivery trigger is: threshold =M used ×thresholdfactor, where M threshold It is the material threshold. When the material consumption in a certain progress stage approaches or exceeds the threshold, the system will automatically trigger the material distribution for the next stage.

9. The AI-based construction project construction progress optimization system according to claim 6 is characterized by: The formula for construction progress prediction is: Among them, S forecast For the construction progress forecast within the next 72 hours, W i is the weighting coefficient, X i Input data predicted for multiple tasks (such as personnel density, dust concentration, voiceprint characteristics, etc.); the formula for resource allocation optimization is: R adjusted =R current ×(1-delayfactor), where R adjusted is the adjusted resource allocation, R current is the current resource allocation, delayfactor is the adjustment factor based on the construction progress delay; The formula for abnormal condition detection is: E anomaly =[PM mean -PM threshold ]∪[D discrepancy >], where E anomaly Indicates the judgment of abnormal situation, PM mean is the average dust concentration monitored, PM threshold is the dust concentration threshold, D discrepancy is the dispersion of the population density, when E anomaly When it is greater than the set benchmark value, the staff needs to go to the site immediately to check the relevant floors or activate the drone to fly to the height of the relevant floors to take pictures inside the floors.

10. The AI-based construction project construction progress optimization system according to claim 9 is characterized in that: The steps for use are as follows: S1: Sensor deployment: Deploy laser dust sensors in a 5m×5m grid in each construction area, install infrared thermal imaging cameras and sound spectrum analyzers; S2: Data initialization: Import the logical relationship of construction process and material consumption benchmark parameters through the BIM model; S3: Environmental data collection: real-time collection of dust concentration on each floor, thermal imaging personnel distribution data, and construction machinery voiceprint characteristics; S4: Edge computing processing: Executed through embedded AI chip: dust concentration-process mapping analysis (detecting PM10 peak to determine polishing operation) and thermal imaging data clustering to generate personnel density index; S5: Progress inversion calculation: input the dust change rate ΔPM, the personnel density index D and the voiceprint classification results into the Transformer model, and output the progress value of each layer; S6: Material demand forecast: trigger material pre-delivery instructions based on the progress value. When the installation progress of a certain layer of steel structure reaches 60%, the bolt and connector delivery demand is automatically generated; S7: Construction path optimization: Adjustment instructions are sent to the automatic plastering robot via the LoRa wireless network or 5GNR-Light. Low-frequency instructions (such as path planning) use LoRa, and high-frequency control (such as emergency braking) is switched to 5GNR-Light (air interface delay <10ms). Path update instructions are sent to the automatic plastering robot to avoid areas with high personnel density. S8: Progress prediction and warning: Use the random forest algorithm to predict the progress in the next 72 hours, and trigger a level 3 warning when the deviation exceeds 8%; Three-level warning classification standards: Level 1 warning (deviation 8%-12%): automatically adjust the delay factor in the resource allocation formula; Level 2 warning (deviation 12%-15%): freeze material delivery and initiate manual review; Level 3 warning (deviation > 15%): forced switch to conservative prediction mode + drone inspection; S9: Dynamic allocation of resources: According to formula R adjusted =R current ×(1-delayfactor) adjust the personnel and equipment configuration at each level; S10: Abnormal working condition processing: When it is detected that D>ΔPM>threshold, it is determined as abnormal construction congestion, and the staff needs to go to the site to immediately check the relevant floors. If the same area is detected to be >threshold for three consecutive times and there is no manual response, the drone is activated to fly to the height of the relevant floor to take pictures of the floor; S11: Data closed-loop update: Feedback the actual progress data to the BIM model, and dynamically feed back the dynamic weight adjustment to the edge computing module to form a data closed loop and optimize the prediction parameters of subsequent processes.

Images