Intelligent building decoration construction supervision method and system based on artificial intelligence

Abstract

The invention relates to the technical field of building decoration supervision, and discloses an intelligent building decoration construction supervision method and system based on artificial intelligence. The method comprises the following steps: acquiring real-time construction data through sensing equipment deployed at a construction site, and inputting the real-time construction data into a pre-trained building decoration quality analysis model; according to a quality evaluation result output by the model, generating a digital construction plane graph containing a construction abnormal region mark; automatically generating a recheck instruction set for the abnormal region based on the abnormal region coordinates; the recheck instruction set is distributed to the corresponding construction terminal equipment, and the terminal equipment is triggered to execute secondary data collection of the designated area; and dynamically adjusting a detection threshold parameter of the building decoration quality analysis model according to a deviation value between the data of the secondary data acquisition and a preset construction standard. The method can achieve the real-time and precise supervision of the building decoration construction process, optimizes the supervision process, improves the construction supervision efficiency and quality, and meets the demands of the building decoration industry for construction supervision.

Description

Technical Field

[0001] This invention relates to the field of building decoration supervision technology, specifically to an intelligent building decoration construction supervision method and system based on artificial intelligence. Background Technology

[0002] In the field of architectural decoration construction, the quality of supervision during the construction process directly affects the final decorative effect and project safety. Traditional construction supervision relies heavily on manual inspection, requiring supervisors to check each construction area on-site, judging whether the construction quality meets standards through visual observation and simple tool measurements. This method is limited by the experience and energy of supervisors, making it difficult to achieve comprehensive coverage and real-time monitoring of the construction process. Especially in large-scale architectural decoration projects, where the construction area is vast and the construction stages are complex, manual inspections often miss some construction areas and fail to detect construction anomalies in a timely manner. As a result, construction defects are not addressed immediately, and subsequent rectification requires more manpower, material resources, and time costs.

[0003] Under traditional regulatory models, the collection and analysis of construction data are fragmented. Various data generated during construction, such as construction site condition data, material usage data, and environmental data, are mostly stored in paper records or scattered electronic documents, lacking a unified collection and management mechanism. This makes it difficult to effectively integrate and utilize the data. When it is necessary to assess construction quality, supervisors must manually organize various data, which is time-consuming, and the assessment results are easily influenced by subjective factors, making it difficult to guarantee the objectivity and accuracy of the assessment.

[0004] Traditional regulatory models lack a dynamic optimization mechanism for construction quality assessment models. During construction, factors such as the construction environment and the characteristics of construction materials may change, making it difficult for existing quality assessment standards and methods to adapt to the changed construction scenarios. Furthermore, traditional models cannot adjust assessment parameters in a timely manner based on actual construction conditions, leading to a gradual decrease in the applicability of the assessment model and consequently affecting the effectiveness of construction quality supervision. With the continuous improvement of construction quality and efficiency requirements in the building decoration industry, traditional regulatory models are no longer sufficient to meet the industry's development needs, necessitating a new construction supervision method that can achieve real-time, comprehensive, and precise monitoring. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent building decoration construction supervision method and system based on artificial intelligence, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent building decoration construction supervision method based on artificial intelligence, the method comprising:

[0007] Real-time construction data is collected by sensing devices deployed at the construction site, and the real-time construction data is input into a pre-trained building decoration quality analysis model.

[0008] Based on the quality assessment results output by the building decoration quality analysis model, a digital construction plan containing markings of construction anomalies is generated.

[0009] Based on the coordinates of the abnormal areas marked on the digital construction plan, a set of re-inspection instructions for the abnormal areas is automatically generated.

[0010] The re-inspection instruction set is distributed to the corresponding construction terminal equipment, triggering the terminal equipment to perform secondary data collection in the specified area;

[0011] Based on the deviation between the secondary data collection data and the preset construction standards, the detection threshold parameters of the building decoration quality analysis model are dynamically adjusted.

[0012] Preferably, the real-time construction data includes three-dimensional point cloud data, material spectral feature data, and environmental temperature and humidity data. The three-dimensional point cloud data is acquired by a laser scanning device at a frequency of twice per second, and the material spectral feature data is collected by a multi-band spectrometer at a vertical distance of 30 centimeters from the construction surface.

[0013] Preferably, the training process of the building decoration quality analysis model specifically includes:

[0014] Construct a construction defect database containing 100,000 labeled samples. Each sample contains multimodal sensing data of the completed construction surface and corresponding process defect labels.

[0015] A hybrid architecture of deep convolutional neural network and long short-term memory network was used to train the construction defect database for twenty rounds of iteration.

[0016] The network weights are dynamically updated using the gradient backpropagation algorithm until the model achieves a classification accuracy of 95% on the validation set.

[0017] Preferably, the process of generating the digital construction plan includes:

[0018] The defect levels in the quality assessment results are mapped to polygon overlays of different colors;

[0019] The transparency of the overlay is gradually changed based on the construction progress timestamp.

[0020] A composite visualization layer is formed by overlaying a two-dimensional foundation layout diagram of the construction site.

[0021] Preferably, the logic for generating the re-inspection instruction set is as follows:

[0022] Extract the geometric center coordinates of the anomaly region markers as the reference positioning points;

[0023] A circular scanning area with a radius of 1.5 meters is generated with the reference positioning point as the center;

[0024] Match one of the six preset data acquisition modes based on the anomaly type.

[0025] Preferably, the execution process of the secondary data acquisition includes:

[0026] The drone is used to take a 10x magnified picture of the tile joints using a microscope camera.

[0027] The ground-penetrating radar of the ground robot was activated to perform millimeter-wave scanning of the concealed engineering layer;

[0028] Simultaneously record ambient light intensity and device posture data during data acquisition.

[0029] Preferably, the method for adjusting the detection threshold parameter includes:

[0030] The variance of three consecutive retest results was calculated as a stability index.

[0031] When the stability index exceeds the critical threshold, the material color difference tolerance will be reduced by 20%.

[0032] When the stability index is below 50 percent of the critical threshold, the historically optimal parameter combination is activated.

[0033] Preferably, the method for obtaining the historical optimal parameter combination is as follows:

[0034] Search for quality assessment records of similar construction projects within the past thirty days;

[0035] Filter out the model parameters corresponding to the top 10% of projects in the acceptance score;

[0036] Establish a database of matching relationships between parameters and construction environment characteristics.

[0037] Preferably, the method further includes:

[0038] Automatically generate a material loss comparison report after the end of each day's construction;

[0039] The procurement list update process is triggered based on the deviation data in the report;

[0040] The updated procurement list will be intelligently matched and verified with the construction plan for the next day.

[0041] Preferably, the present invention also includes an intelligent building decoration construction supervision system based on artificial intelligence, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the above-mentioned intelligent building decoration construction supervision method based on artificial intelligence.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] This AI-based intelligent construction and decoration supervision method collects real-time construction data through sensing devices deployed at the construction site and inputs it into a pre-trained construction and decoration quality analysis model. This changes the traditional model that relies on manual data collection and quality assessment, making the data collection and quality evaluation process more real-time and objective. The sensing devices can continuously and stably acquire various types of data from the construction site, ensuring the comprehensiveness of data collection and avoiding data omissions due to human error during manual collection. Simultaneously, the pre-trained analysis model processes the data according to fixed algorithms and logic, reducing subjective interference in the manual assessment process and making the quality assessment results more closely reflect the actual construction situation.

[0044] Based on the quality assessment results output by the building decoration quality analysis model, a digital construction plan is generated, including markers for construction anomalies. This transforms abstract quality assessment results into intuitive visual graphics, allowing supervisors to quickly locate anomalies and clearly understand their specific locations and extent. Unlike traditional supervisory methods, this eliminates the need for on-site inspections, significantly reducing the time spent searching for anomalies and improving the efficiency of construction supervision. Furthermore, the digital construction plan serves as a crucial record-keeping tool for construction supervision, facilitating subsequent tracing and analysis of construction anomalies and providing clear data support for post-construction review.

[0045] The system automatically generates re-inspection instruction sets based on the coordinates of abnormal areas in digital construction plans, achieving precise generation of re-inspection tasks. This process generates instructions based on clearly defined abnormal area coordinates, avoiding the inaccuracies in re-inspection areas and unclear content caused by human judgment errors in traditional re-inspection processes. This ensures that re-inspection tasks are precisely targeted at abnormal construction areas. The re-inspection instruction sets are distributed to corresponding construction terminal equipment and trigger secondary data collection, enabling rapid implementation of the re-inspection work. Construction personnel can directly proceed to the designated area for data collection based on the instructions received from the terminal equipment, reducing intermediate communication steps and improving the response speed and execution efficiency of the re-inspection work.

[0046] The detection threshold parameters of the building decoration quality analysis model are dynamically adjusted based on the deviation between the secondary data collection and the preset construction standards, enabling the model to adapt to actual changes during construction. During construction, factors such as the construction environment and material properties may change. The deviation between the secondary data collection and the preset standards reflects the impact of these changes on construction quality assessment. By adjusting the detection threshold parameters, the evaluation logic of the analysis model can be matched with the actual construction scenario, ensuring that the model maintains good evaluation performance under different construction stages and conditions. This continuously guarantees the effectiveness of construction quality supervision, promotes a virtuous cycle in the building decoration construction supervision process, and better meets the quality and efficiency requirements of building decoration construction. Attached Figure Description

[0047] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent building decoration construction supervision method based on artificial intelligence as described in this invention.

[0048] Figure 2 A flowchart illustrating real-time construction data collection;

[0049] Figure 3 A flowchart for training a building decoration quality analysis model. Detailed Implementation

[0050] 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.

[0051] Please see Figure 1 This invention provides an intelligent building decoration construction supervision method and system based on artificial intelligence. The method includes integrated sensing technology, artificial intelligence models and automated equipment to achieve real-time monitoring and dynamic optimization of construction quality.

[0052] Multiple sensing devices, including laser scanners and multi-band spectrometers, are deployed at the construction site to continuously collect real-time construction data. This data is transmitted to a pre-trained construction quality analysis model for processing. This model, built on a deep learning architecture, outputs construction quality assessment results. Based on the assessment results, the system automatically generates a digital construction plan, visually marking abnormal areas. Based on the coordinate information of these abnormal areas, a targeted re-inspection instruction set is generated, containing specific scanning areas and data acquisition modes. This re-inspection instruction set is distributed to corresponding construction terminal devices, such as drones or ground robots, triggering these devices to perform secondary data acquisition operations. Finally, the system dynamically adjusts the detection threshold parameters of the construction quality analysis model based on the deviation between the secondary data acquisition data and the preset construction standards, thereby optimizing the model's adaptability and accuracy. The entire process forms a closed-loop control system, achieving intelligent and adaptive construction supervision.

[0053] Example 1: See Figure 2 The real-time construction data acquisition process relies on the collaborative configuration and data fusion of multiple sensing devices. The acquisition of 3D point cloud data is accomplished through high-precision laser scanning equipment deployed at fixed points on the construction site. These devices are typically installed on height-adjustable supports at the four corners and center of the construction area. Their scanning heads perform a composite horizontal and vertical scan at a fixed frequency of twice per second. The laser point cloud data generated during the scanning process contains millions of spatial coordinate points, each with intensity information, accurately reflecting the microscopic flatness and macroscopic geometry of completed surfaces such as walls, floors, and ceilings. Before starting, the laser scanning equipment needs to be calibrated on-site by measuring the relative positions of known control points to calculate coordinate transformation parameters. The scanning operation covers the entire construction area according to a preset path plan, with adjacent scanning areas maintaining a 20% overlap to ensure data continuity. The acquired raw point cloud data is processed by a real-time filtering algorithm to remove noise caused by dust interference, and then transmitted via gigabit industrial Ethernet to a local edge computing node for preliminary registration.

[0054] The acquisition of material spectral characteristic data is performed by a handheld multi-band spectrometer. Operators periodically inspect key areas according to the construction progress. The spectrometer is stably hovered at a vertical distance of 30 centimeters from the construction surface, and its built-in distance sensor ensures the accuracy of the measurement height. The spectrometer's optical probe is aimed at the surface to be measured at a 45-degree angle to avoid specular reflection interference. The acquisition bands cover the visible and near-infrared range, and a continuous spectral reflectance curve is obtained for each measurement point, containing values ​​from 256 sampling points. For large-area construction surfaces such as wall paint, a serpentine scanning mode is used, with the spacing between adjacent measurement points set to 10 centimeters. For unit-based construction such as tile laying, sampling is performed at the center and seams of each unit. Spectral data is displayed on the instrument screen in real time, and repeated measurements are triggered immediately if abnormal fluctuations are detected. All spectral data files are accompanied by GPS coordinates and timestamps and are stored in association with the corresponding construction surface number.

[0055] Environmental temperature and humidity data are continuously monitored via a wireless sensor network, with sensor nodes deployed in a three-dimensional grid within the construction space. Each node includes a digital temperature and humidity sensor, a microprocessor, and a low-power wireless communication module. The node spacing is dynamically adjusted according to the construction area, typically maintaining a communication distance of less than ten meters. The sensors collect environmental data every five minutes, covering a temperature range of 0 to 50 degrees Celsius and a relative humidity range of 5% to 95%, with accuracies of ±0.5 degrees Celsius and ±3%, respectively. Data is transmitted to a gateway device via a wireless mesh network, where the gateway performs time synchronization and packet processing on the data from multiple nodes. Considering the environmental sensitivity of decorative materials, the system automatically increases the sampling frequency to once per minute during critical process stages such as paint spraying and cement curing. Long-term monitoring data generates temperature and humidity trend charts, which are then correlated with the material curing cycle for analysis. Data collected by various sensing devices employs a unified encryption protocol and data compression algorithm during transmission. Laser scan data uses a point cloud-specific compression format to reduce bandwidth consumption, while spectral data uses lossy compression to preserve characteristic peak values. After the data arrives at the central server, it first undergoes time alignment processing. Using the laser scanning time point as a baseline, interpolation algorithms are used to synchronize data from other modalities to a unified timeline. Multimodal data fusion is performed at the feature level, for example, by concatenating the 3D geometric features and spectral features of the same spatial coordinate point into a multidimensional vector. The fused data is stored in a spatiotemporal database, with each data record containing metadata such as spatial coordinates, timestamp, equipment number, and data quality identifier. The database establishes an index relationship with the BIM model, supporting multi-dimensional queries by construction stage, spatial location, material type, and other dimensions.

[0056] The operational status of sensing devices is monitored in real time. Laser scanning equipment detects vibration shifts using a built-in gyroscope; when the tilt angle exceeds a threshold, scanning automatically pauses and a calibration reminder is issued. The multi-band spectrometer is equipped with a self-cleaning device that periodically triggers airflow to clean the optical lenses. The wireless sensor network has a self-healing function; when a node fails, adjacent nodes automatically adjust their routing paths to maintain network connectivity. All devices have centralized operational logs, including startup time, runtime, and abnormal alarms, providing a basis for equipment maintenance. The data acquisition process is embedded in the construction management program; after a specific process is completed in a certain area, the system automatically generates a testing task sheet for that area and pushes it to the mobile terminals of relevant personnel. A data quality control mechanism is implemented throughout the entire acquisition process. Laser scanning equipment executes a standard plate calibration procedure after each startup, verifying measurement accuracy by scanning a calibration plate of known size. The multi-band spectrometer performs baseline correction daily using a standard white board to eliminate the influence of ambient light changes. The wireless temperature and humidity sensor is periodically compared on-site with a portable high-precision measuring instrument; when the deviation exceeds the limit, the sensor calibration process is triggered. If data loss or anomalies are detected during acquisition, the system automatically marks the data for that time period and generates a re-acquisition instruction. All quality control records are stored in association with the original data, forming a complete data traceability chain.

[0057] The deployment scheme for sensing equipment takes into account the dynamic changes in construction. Laser scanning equipment uses a track-mounted mobile support, allowing for position adjustments based on construction progress. Wireless sensor network nodes employ magnetic installation for rapid deployment on steel structure surfaces. Multi-band spectrometers are equipped with portable tripods and remote control capabilities, supporting unattended measurements. The equipment power supply system uses a hybrid mode of AC power and batteries, with uninterruptible power supplies (UPS) for critical nodes. Communication links between devices employ a redundant design, automatically switching to backup wireless channels when the main data transmission path is interrupted. The entire acquisition system is linked to the construction schedule, automatically increasing data acquisition density during construction breaks and employing a sampling detection mode to balance efficiency and accuracy during peak periods. Preprocessing of the acquired data is completed at edge computing nodes. Point cloud data is downsampled using voxel grids to reduce data volume while preserving key geometric features. Spectral data undergoes smoothing filtering and baseline correction to highlight material characteristic absorption peaks. Temperature and humidity data uses a moving average algorithm to eliminate instantaneous fluctuations. Preprocessed data is converted to standard JSON format and streamed to the cloud platform, with each data packet containing a checksum to prevent transmission errors. The platform receiver performs integrity checks on the data, discarding damaged data packets and requesting retransmission. All data operations are recorded in audit logs to meet the traceability requirements of the engineering quality management system.

[0058] Example 2: See Figure 3The model training is built upon a construction defect database containing 100,000 labeled samples. The construction of this database involved the collection of massive amounts of data from historical engineering projects. Each sample contains multimodal sensor data of the completed surface and corresponding process defect labels. The multimodal sensor data includes high-resolution image sequences, 3D laser point clouds, and material spectral data. The defect labels are manually annotated by senior supervising engineers according to industry acceptance standards, and the label system includes common process problems such as hollow areas, cracks, and unevenness. The database employs a stratified sampling strategy to organize samples, ensuring a balanced proportion of samples from different defect types, construction stages, and material categories. Each sample is accompanied by detailed metadata, including project number, construction date, and work team information. The database update mechanism is synchronized with model training; newly completed engineering project data is periodically added to the training set after annotation.

[0059] The model employs a hybrid architecture of deep convolutional neural networks (CNNs) and long short-term memory (LSTM) networks for training. The CNN branch handles spatial data such as images and point clouds, and its network structure includes multiple convolutional and pooling layers to extract feature maps at different scales. The LSTM branch processes time-series data from the construction process, such as environmental parameter variation curves. The training process involves twenty iterations on a construction defect database. In each iteration, the dataset is randomly shuffled and input into the network in batches, with the batch size optimized based on GPU memory capacity. Network weights are dynamically updated using gradient backpropagation, employing an adaptive moment estimation optimizer. The learning rate dynamically decays according to training progress. An early stopping mechanism is introduced during training: training terminates when the validation set loss function no longer decreases for several consecutive cycles to prevent overfitting. Once the model's classification accuracy on the validation set reaches a predetermined standard, the network structure and parameter matrix are saved for deployment.

[0060] After model training, the process moves to the generation stage of the digital construction plan. This process first maps the defect levels from the quality assessment results to polygon overlays of different colors. Defect levels are categorized based on severity, with each category corresponding to a color coding scheme: red for urgent defects, orange for important defects, and yellow for general defects. The polygon overlays are generated based on the boundary coordinates of the defect areas. A contour extraction algorithm connects consecutive defect points into closed polygons, and the polygon vertex coordinates are converted to the plan coordinate system. The overlays undergo a gradual transparency change based on construction progress timestamps; newly discovered defects are displayed opaquely, while defects with longer histories gradually increase in transparency. The rate of transparency change is related to the defect repair cycle. The processed overlays are then overlaid with a two-dimensional foundation layout map of the construction site. This map, derived from a vector graphics file exported from the building information model, contains basic information such as wall axes, door and window openings, and pipeline routes. The overlay process employs a coordinate system transformation algorithm to convert the defect polygon coordinates from the measurement coordinate system to the construction drawing coordinate system, ensuring accurate spatial correspondence.

[0061] The visualization layers are generated using a layered rendering technique. The bottom layer is a wireframe display of the 2D basic layout, the middle layer is filled with color blocks representing the construction progress status, and the top layer is a semi-transparent overlay of defect annotations. Layers support multi-level zooming, displaying an overview of defect distribution in the global view and detailed defect information in a zoomed-in view. Each defect polygon is associated with an attribute database; clicking on it displays detailed information such as defect type, discovery time, and responsible work team. The plan view update mechanism is synchronized with quality assessment; when new detection data is input into the model, the system automatically updates the defect overlay layer and adjusts the transparency gradient. Historical defect data is archived and managed, supporting queries for construction status within a specific time range. The model training phase also includes data augmentation strategies, performing random rotations, scaling, and color adjustments on training samples to increase data diversity. During training, changes in various indicators are monitored, including training loss, validation accuracy, and recall, and training curves are generated for analyzing model convergence. Before deployment, the model undergoes rigorous testing, using an independent test set to evaluate its generalization ability under different construction scenarios. The digital floor plan output format supports multiple standards, including commonly used engineering formats such as DWG and PDF, facilitating integration with other management systems. Both model parameters and floor plan data are version-identified, supporting historical version backtracking and comparison. The model inference process optimizes computational efficiency, preprocessing input data to reduce computational load and employing model quantization technology to minimize memory usage. Digital floor plan rendering utilizes hardware acceleration technology, supporting real-time visualization of large datasets. The system establishes an automatic backup mechanism, regularly backing up training data, model parameters, and floor plan data to a disaster recovery server. User access is managed hierarchically, allowing different roles to access different levels of data and operational functions.

[0062] Example 3: The generation of the re-inspection instruction set begins with the analysis of abnormal area markers on the digital construction plan. The system extracts the geometric center coordinates of each abnormal area marker as a reference positioning point. These coordinates are converted through the mapping relationship between the plan coordinate system and the actual construction site coordinate system. The calculation of the reference positioning point adopts a weighted centroid algorithm, comprehensively considering the defect confidence weight of each pixel within the abnormal area. A circular scanning area with a radius of 1.5 meters is generated with the reference positioning point as the center. This radius value is set based on the statistical data analysis of the impact range of common decoration construction defects, which can cover the core area and potential diffusion range of typical defects. The generated circular area is presented as a geometric shape with specific markings on the two-dimensional plan. The coordinates of the shape boundary are converted into the GPS coordinates of the construction site. According to the anomaly type, one of the six preset data acquisition modes is matched. The matching process adopts a decision tree algorithm based on rule reasoning, which performs multi-dimensional association mapping between defect features and acquisition modes.

[0063] The following formula is used to calculate the location of the center point of the abnormal region:

[0064]

[0065] in: The coordinate vector of the reference point in the abnormal region. Represents the first in the abnormal region The planar coordinates of each pixel The defect confidence weighting coefficient for that pixel should be... This represents the total number of pixels contained in the abnormal region. Weighting coefficient. The value of is derived from the probability value output by the quality assessment model, ranging from 0 to 1. Pixels with higher confidence levels have a greater impact on center point localization. This calculation process ensures that the reference point can accurately reflect the spatial distribution characteristics of the abnormal area.

[0066] The secondary data acquisition relies on the collaborative operation of intelligent terminal devices, controlling the microscopic camera mounted on the drone to magnify the tile joints tenfold. The drone autonomously navigates to the target point using received GPS coordinates, hovers at a height of three meters, and adjusts its attitude to ensure the camera lens is perpendicular to the ground. The microscopic camera activates its autofocus mechanism, setting the optical magnification to tenfold, and uses digital zoom to achieve micron-level resolution image acquisition. During the shooting process, the drone maintains stable attitude, using a stabilization platform composed of a four-axis gyroscope and accelerometer to counteract external disturbances. Three sets of image data are collected at each shooting point, targeting indicators such as joint width, grout fullness, and edge smoothness. The ground robot's ground-penetrating radar system performs millimeter-wave scanning of the concealed engineering layer. The robot moves at a constant speed of 0.5 meters per second along a predetermined path. The radar antenna transmits millimeter waves at a frequency of 60 GHz towards the ground at a 45-degree angle, and the receiver records the time-series data of the reflected signals. The scanning depth can be adjusted from five to twenty centimeters by adjusting the transmission power to meet the different levels of concealed engineering inspection needs. Radar data is transmitted to the processing unit in real time, where time-domain analysis algorithms are used to identify features such as hollow areas in the base layer and the depth of pipeline burial. Simultaneously, the robot records the coordinate data of its trajectory, establishing a correspondence between the scanned data and its spatial location.

[0067] Environmental parameters are recorded synchronously during the data acquisition process. Ambient light intensity is measured by a photometric sensor mounted on the UAV, with a sampling frequency of ten times per second, covering a measurement range from zero to 100,000 lux. Equipment attitude data is acquired via an inertial measurement unit, including three-dimensional acceleration, angular velocity, and magnetometer readings. All sensor data is timestamped with a precise time protocol to ensure strict synchronization of multi-source data. Data storage adopts a hierarchical structure, retaining complete information in the raw data, and generating a standard-format inspection report after preprocessing. Six data acquisition modes for anomaly area matching each have their own focus: high-resolution imaging mode is suitable for surface defect detection, thermal imaging mode is used to investigate hollow defects, acoustic detection mode is for substrate adhesion quality, multispectral scanning analyzes material consistency, three-dimensional morphology measurement assesses flatness, and humidity distribution detection detects potential leakage. For each mode, the system automatically selects the optimal acquisition scheme based on the anomaly characteristics, considering the corresponding equipment parameter combinations and scanning path planning. The acquisition task scheduling algorithm takes into account equipment resource allocation, and generates the optimal inspection path after clustering multiple anomaly areas by spatial location.

[0068] A quality control mechanism is established for the data acquisition process. Before each shot, the drone performs a self-check to calibrate the camera's white balance and exposure parameters. Upon startup, the ground robot performs radar system diagnostics to verify transmit power and receive sensitivity. Acquired data is transmitted in real-time to edge computing nodes for initial verification; if data quality is found to be substandard, a re-acquisition process is triggered. All operation records form a complete log, including equipment status, environmental parameters, and abnormal events. The acquisition task progress is updated in real-time to the central management system, supporting remote monitoring and intervention. Device collaboration is achieved through a wireless mesh network, with the drone and ground robot establishing point-to-point communication to share location information and task status. Network transmission employs an anti-interference protocol to maintain connection stability in complex construction environments. Data security mechanisms encrypt the transmission channel to prevent unauthorized access. The entire secondary acquisition system dynamically adapts to the construction progress, automatically adjusting the work plan during periods of high personnel activity to avoid interfering with normal construction. Complete metadata records of the acquired data, including equipment serial numbers, operator identifiers, and calibration certificate numbers, meet the traceability requirements of the quality management system.

[0069] Example 4: The dynamic adjustment mechanism of the detection threshold parameter optimizes the model's judgment criteria by analyzing the stability characteristics of continuously re-inspected data. Assume that during the construction of a wall tile laying project, the system continuously monitored the northwest facade area. Referring to Table 1, it shows the data records of five consecutive re-inspections of this area and their corresponding stability analysis:

[0070] Table 1: Record of Re-inspection Data for Northwest Facade Tile Paving Area

[0071] Re-inspection time point Testing items Retest results Permissible deviation range Result Status Calculate the variance value 2023-08-1009:00 flatness deviation 1.2mm ≤2.0mm qualified 0.15 2023-08-1011:30 flatness deviation 1.8mm ≤2.0mm qualified 0.28 2023-08-1014:00 flatness deviation 2.1mm ≤2.0mm Unqualified 0.42 2023-08-1109:00 Seam width 2.5mm 2.0±0.3mm Unqualified 0.31 2023-08-1114:30 Seam height difference 0.8mm ≤1.0mm qualified 0.19

[0072] The adjustment of the detection threshold parameters is based on data stability analysis within a sliding window. The system uses the results of the three most recent re-inspections as the calculation unit, and automatically updates the calculation window each time new re-inspection data is added. The stability index is calculated using a variance formula to measure the dispersion of three consecutive measurements; a larger variance value indicates more significant fluctuations in construction quality. When the system detects that the stability index of a certain inspection item exceeds the preset critical threshold, it automatically triggers the parameter adjustment program. The critical threshold is set to 0.35 based on historical data statistical analysis. Taking the flatness deviation data in Table 1 as an example, the variance value calculated after the third re-inspection is 0.42, exceeding the critical threshold. At this time, the system lowers the material color difference tolerance parameter by 20%. The parameter adjustment operation is implemented by modifying the model judgment logic, specifically by narrowing the acceptable judgment range and increasing the stringency of the inspection standards.

[0073] The process of obtaining the historically optimal parameter combination is closely related to the characteristics of the current construction environment. The system searches the database of similar projects that have been completed and accepted within the past 30 days. Search criteria include features such as construction process type (tile laying), material specifications (600mm×600mm ceramic tiles), and working environment (dry indoor area). The screening mechanism first excludes projects with acceptance scores below 90 points, then selects the top 10% of the remaining projects and extracts the model parameter configurations corresponding to these high-quality projects. The parameter combination includes more than ten adjustable parameters such as flatness tolerance threshold, joint width allowable deviation, and color difference detection sensitivity. These parameters are stored in the optimization parameter library in vector form. The matching relationship between parameters and construction environment characteristics is achieved through a feature weighting algorithm, with the system assigning different weight coefficients to each environmental feature. For example, the weight of ambient temperature is set to 0.15, humidity to 0.12, construction team experience value to 0.23, and material batch consistency to 0.18. The matching process calculates the cosine similarity between the current project environment feature vector and the historical project feature vectors, selects the five project parameters with the highest similarity as a candidate set, and then generates the final parameter combination through a weighted average. When the system detects that the stability index is below 50% of the critical threshold, it automatically calls the historically optimal parameter combination with the highest matching degree from the optimized parameter library. This adjustment method is suitable for situations where construction quality remains consistently stable and helps maintain the consistency of testing standards.

[0074] The data flow during parameter adjustment has a two-way feedback characteristic. The detection results after each parameter modification are recorded and the effect is evaluated. The system establishes a parameter adjustment log, recording in detail the time, reason, adjustment range, and quality data changes for three detection cycles after each adjustment. The evaluation mechanism judges the effectiveness of parameter adjustments by comparing indicators such as changes in defect detection rate and false alarm rate fluctuations before and after the adjustment. If two consecutive adjustments fail to improve stability indicators, the system will initiate expert intervention mode, reporting the parameter configuration problem to technical personnel for handling. All parameter adjustment operations are audited, supporting later traceability analysis. The historical parameter database is maintained using a dynamic update mechanism; parameters for newly completed projects with excellent acceptance scores are regularly added to the database. Data cleaning and standardization are performed before data is added to the database to eliminate system errors caused by differences in measurement equipment. The database supports parameter version management; each parameter combination is marked with applicable conditions, effective time, and verification results. The system automatically generates a parameter usage effect report weekly, analyzing the adaptability of different parameter combinations under different construction environments, providing data support for subsequent parameter optimization. The threshold parameter adjustment mechanism is linked to the construction schedule, pre-adjusting parameter settings before the start of key processes. For example, during the waterproofing construction phase, the system will proactively increase the sensitivity of humidity detection parameters; during the final stages of interior decoration, it will appropriately raise the inspection standards for surface smoothness. This predictive adjustment is based on matching with a process feature library, which includes common quality problems at each construction stage and their corresponding parameter optimization strategies. The system also considers seasonal environmental changes, such as automatically relaxing the detection threshold for wood moisture content during the rainy season to avoid an increase in false alarm rates due to generally high ambient humidity.

[0075] The boundary conditions for parameter adjustments are ensured through a safety constraint mechanism, with each adjustable parameter having a maximum allowable adjustment range. For example, the color difference tolerance parameter is adjusted no more than 30% of its original value each time, and the flatness threshold adjustment range is limited to ±0.5 mm. These constraints prevent excessive adjustments from causing the detection standards to deviate from a reasonable range, maintaining the stability of the quality assessment system. The system monitors the parameter adjustment frequency in real time. If a parameter is found to be adjusted frequently within a short period, a system self-check procedure will be triggered to investigate whether there are sensor malfunctions or data anomalies. Verification of the historically optimal parameter combination uses a cross-validation method. Newly added parameter combinations must be trial-run in three different construction areas of the current project. During the trial run, the system runs the old and new sets of parameters in parallel, comparing the consistency of defect detection and the difference in false alarms. Only when the new parameter combination shows better adaptability will it be officially and fully implemented. This gradual parameter update strategy reduces the risk of quality detection caused by parameter incompatibility, ensuring the robust operation of the monitoring system. The parameter management interface provides a visual operating environment, allowing technicians to observe the trend of quality data changes after parameter adjustments through graphs. The system supports manual fine-tuning of parameter combinations, allowing experienced engineers to temporarily override automatically adjusted results in special circumstances. All manual adjustments require double-checking and confirmation, with detailed records of the reasons for the adjustments and the expected results, ensuring that every parameter change is traceable. Parameter configurations can be ported between projects, and successfully applied parameter combinations can be exported as templates for reference in similar projects.

[0076] Example 5: Material Management Process After Construction Completion. A material loss comparison report is automatically generated daily after construction ends. Taking an office building decoration project as an example, the daily construction work involved the installation of gypsum board on the walls of floors five through eight. The system automatically starts the report generation program at 6:00 PM. The report data comes from the integration of multiple heterogeneous systems. The warehouse management system provides daily outbound records for various materials, including the quantity and specifications of major materials such as gypsum board, light steel keel, and self-tapping screws. The on-site intelligent shelving records the actual quantity of materials received by each construction team through gravity sensors and links it to the construction task sheet. The image recognition system installed in the construction area estimates the amount of various scrap materials generated by analyzing images of waste piles. The system extracts planned usage data from the BIM model and compares it item by item with actual usage data to calculate the loss rate of each material. After the report is generated, it is automatically pushed to the mobile terminals of the project manager, material handler, and relevant team leaders.

[0077] The system triggers a purchase order update process based on deviation data in the reports. The system is configured to activate the update mechanism when the actual wastage rate of a certain material exceeds 5% of the planned value. Taking gypsum board as an example, the planned daily usage was 320 standard sheets, the actual outbound record was 340 sheets, and the on-site image recognition system detected approximately 15 discarded sheets, resulting in a wastage rate of 9.4%, exceeding the set threshold. The system automatically generates a purchase requisition form, specifying the quantity, specifications, and required delivery time of the required gypsum board. The purchase order update process is integrated with the supplier management system, automatically selecting partner manufacturers based on preset supplier ratings and generating a draft purchase order with price information. After confirmation by the materials clerk, the draft order is directly sent to the supplier order processing center via the enterprise resource planning system interface.

[0078] The updated procurement list is intelligently matched and verified against the next day's construction plan. The system first analyzes the material requirements in the next day's construction plan. The plan for the next day is to construct the ceiling on floors nine through twelve, requiring materials such as gypsum board, mineral wool board, and hangers. The system matches the gypsum board delivery time in the procurement list with the next day's construction plan to detect any potential material supply disruptions. The verification algorithm analyzes the construction progress network diagram, identifies material demand time nodes on the critical path, and ensures that the delivery time of critical materials is earlier than the start of construction. For materials on non-critical paths, the system optimizes delivery times to reduce on-site storage pressure. The material delivery plan is dynamically linked to the construction area. Based on the next day's construction area distribution map, the system generates a detailed plan for material delivery to each floor. Taking the ninth-floor construction area as an example, the system calculates that 180 sheets of gypsum board are needed for this area. Based on the construction schedule, the optimal delivery time is determined to be between 8:30 and 9:00 the next day. The delivery plan considers the frequency of construction elevator use, avoiding peak commuting hours to improve vertical transportation efficiency. The system automatically generates delivery task orders, specifying the handling team and delivery route to avoid transportation conflicts between different types of work.

[0079] The inventory early warning mechanism works in tandem with the procurement process, with the system monitoring material inventory in the on-site warehouse in real time. When the on-site inventory of a certain material falls below the safety stock level, the replenishment quantity is automatically added to the purchase order. The safety stock level is dynamically adjusted based on historical consumption data, with the safety stock standard appropriately increased during peak construction periods. For special materials such as flammable and explosive materials, the system strictly adheres to the first-in, first-out (FIFO) principle, marking the material batch information in the purchase order to ensure timely turnover of inventory materials. The material acceptance process is linked to the quality supervision system; materials delivered the next day must undergo quality sampling inspection before being put into storage. The system determines the sampling rate based on the supplier's historical supply quality records, increasing the sampling frequency for suppliers with lower quality ratings. During acceptance, the material's QR code is scanned using a mobile device to verify the consistency of specifications and model with the purchase order. Acceptance data is transmitted back to the central database in real time, updating inventory records and generating acceptance reports. When substandard materials are found, the system automatically triggers a return process and generates a new emergency purchase order. The material allocation mechanism is linked to construction tasks, with the system allocating materials according to the construction workload of each work team. When work teams collect materials, they need to scan their personal work badges and the material's QR code. The system records the work team that received each batch of materials and its usage. This refined traceability mechanism can accurately calculate the material usage efficiency of each work team, providing data support for cost accounting. Material usage data is fed back to the reporting system, forming a closed-loop management system.

[0080] The system performs a final verification at 10 PM, reassessing the match between the procurement list and the construction plan. Verification includes aspects such as material sufficiency, reasonable delivery times, and feasibility of delivery routes. When conflicts are found, adjustment suggestions are automatically generated, such as postponing the construction time of non-critical processes or adjusting material delivery batches. The final verification results are compiled into a report, highlighting potential risks and corresponding countermeasures. A cross-project material allocation mechanism optimizes resource utilization. When a project experiences material shortages while other projects have surplus inventory, the system initiates allocation suggestions. These suggestions consider factors such as transportation costs and material storage conditions, prioritizing projects with closer proximity and higher material matching rates. The allocation process generates allocation slips, recording the material's origin, recipient, and allocation time, ensuring traceability of asset transfers. The system supports plan adjustments in unforeseen circumstances. If weather conditions cause changes to the construction plan, material requirements are automatically recalculated. When plans change, the system assesses the storage requirements of purchased materials and generates protective measures suggestions for materials requiring special storage.

[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart construction and decoration supervision method based on artificial intelligence, characterized in that, Includes the following steps: Real-time construction data is collected by sensing devices deployed at the construction site, and the real-time construction data is input into a pre-trained building decoration quality analysis model. Based on the quality assessment results output by the building decoration quality analysis model, a digital construction plan containing markings of construction anomalies is generated. Based on the coordinates of the abnormal areas marked on the digital construction plan, a set of re-inspection instructions for the abnormal areas is automatically generated. The re-inspection instruction set is distributed to the corresponding construction terminal equipment, triggering the terminal equipment to perform secondary data collection in the specified area; Based on the deviation between the secondary data collection data and the preset construction standards, the detection threshold parameters of the building decoration quality analysis model are dynamically adjusted.

2. The intelligent building decoration construction supervision method based on artificial intelligence according to claim 1, characterized in that, The real-time construction data includes three-dimensional point cloud data, material spectral feature data, and environmental temperature and humidity data. The three-dimensional point cloud data is acquired twice per second by a laser scanning device, and the material spectral feature data is collected by a multi-band spectrometer at a vertical distance of 30 centimeters from the construction surface.

3. The intelligent building decoration construction supervision method based on artificial intelligence according to claim 1, characterized in that, The training process of the building decoration quality analysis model specifically includes: Construct a construction defect database containing 100,000 labeled samples. Each sample contains multimodal sensing data of the completed construction surface and corresponding process defect labels. A hybrid architecture of deep convolutional neural network and long short-term memory network was used to train the construction defect database for twenty rounds of iteration. The network weights are dynamically updated using the gradient backpropagation algorithm until the model achieves a classification accuracy of 95% on the validation set.

4. The intelligent building decoration construction supervision method based on artificial intelligence according to claim 3, characterized in that, The process of generating the digital construction site plan includes: The defect levels in the quality assessment results are mapped to polygon overlays of different colors; The transparency of the overlay is gradually changed based on the construction progress timestamp. A composite visualization layer is formed by overlaying a two-dimensional foundation layout diagram of the construction site.

5. The intelligent building decoration construction supervision method based on artificial intelligence according to claim 1, characterized in that, The specific logic for generating the re-inspection instruction set is as follows: Extract the geometric center coordinates of the anomaly region markers as the reference positioning points; A circular scanning area with a radius of 1.5 meters is generated with the reference positioning point as the center; Match one of the six preset data acquisition modes based on the anomaly type.

6. The intelligent building decoration construction supervision method based on artificial intelligence according to claim 5, characterized in that, The execution process of the secondary data acquisition includes: The drone is used to take a 10x magnified picture of the tile joints using a microscope camera. The ground-penetrating radar of the ground robot was activated to perform millimeter-wave scanning of the concealed engineering layer; Simultaneously record ambient light intensity and device posture data during data acquisition.

7. The intelligent building decoration construction supervision method based on artificial intelligence according to claim 1, characterized in that, The method for adjusting the detection threshold parameter includes: The variance of three consecutive retest results was calculated as a stability index. When the stability index exceeds the critical threshold, the material color difference tolerance will be reduced by 20%. When the stability index is below 50 percent of the critical threshold, the historically optimal parameter combination is activated.

8. The intelligent building decoration construction supervision method based on artificial intelligence according to claim 7, characterized in that, The method for obtaining the historical optimal parameter combination is as follows: Search for quality assessment records of similar construction projects within the past thirty days; Filter out the model parameters corresponding to the top 10% of projects in the acceptance score; Establish a database of matching relationships between parameters and construction environment characteristics.

9. The intelligent building decoration construction supervision method based on artificial intelligence according to claim 1, characterized in that, Also includes: Automatically generate a material loss comparison report after the end of each day's construction; The procurement list update process is triggered based on the deviation data in the report; The updated procurement list will be intelligently matched and verified with the construction plan for the next day.

10. An intelligent building decoration construction supervision system based on artificial intelligence, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent building decoration construction supervision method based on artificial intelligence as described in any one of claims 1 to 9.

Images