Intelligent detection system for construction engineering quality
By using multi-source sensing modules, AI intelligent analysis, and digital twin modeling, the problems of low efficiency and difficulty in identifying hidden defects in traditional building engineering quality inspection have been solved, achieving efficient and accurate building engineering quality inspection and providing a full-cycle quality control solution.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 阳强
- Filing Date
- 2026-03-10
- Publication Date
- 2026-06-09
AI Technical Summary
Traditional construction quality inspection relies on manual inspection, which is inefficient and inconsistent, and cannot cover the entire project. The inspection results are greatly affected by human factors and cannot effectively identify damage and defects in hidden works. Existing intelligent inspection has failed to achieve deep integration of multi-dimensional data and dynamic modeling, making it difficult to meet the accuracy requirements of inspection under complex working conditions.
By employing multi-source sensing modules, AI intelligent analysis, and digital twin modeling, data is acquired through a multi-source sensor array, transmitted via 5G and industrial IoT, preprocessed using edge computing, and a digital twin is constructed. AI models are then applied for defect identification and risk assessment, and blockchain is used for data storage to ensure the immutability of the data.
Significantly improves detection efficiency and coverage, enhances the accuracy and reliability of surface and hidden defect identification, extends the early warning time of potential risks, and provides an efficient and accurate full-cycle quality control solution.
Smart Images
Figure CN122170956A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of building engineering quality testing technology, and in particular to an intelligent building engineering quality testing system. Background Technology
[0002] Traditional construction quality inspection relies on manual inspection and single-equipment sampling. Manual operation is time-consuming and labor-intensive, making it difficult to cover the entire project scope, resulting in low inspection efficiency. At the same time, the inspection results are greatly affected by the experience of the inspectors, resulting in poor consistency. Furthermore, the data obtained by different inspection methods are not in the same format and lack correlation analysis. Moreover, existing inspections can only detect exposed surface defects and cannot predict some hidden defects, resulting in a lag in defect identification.
[0003] Although there have been attempts to partially solve the problems of manual inspection through intelligent detection, they have failed to achieve deep integration and dynamic modeling of multi-dimensional data, making it difficult to meet the accuracy requirements of detection under complex working conditions. Therefore, this invention proposes an intelligent detection system for building engineering quality to solve the problems existing in the prior art. Summary of the Invention
[0004] To address the aforementioned problems, the present invention aims to propose an intelligent inspection system for building engineering quality. This intelligent inspection system for building engineering quality achieves automation of the inspection process and intelligent data processing through multi-source sensing fusion, AI intelligent analysis, and digital twin modeling. It significantly improves inspection efficiency and coverage, enhances the accuracy and reliability of surface and hidden defect identification, extends the early warning time of potential risks, and strengthens data traceability and immutability through blockchain evidence storage.
[0005] To achieve the objectives of this invention, the following technical solution is provided: an intelligent inspection system for building engineering quality, comprising a multi-source sensing module, a data transmission module, an intelligent inspection module, and an intelligent application module. The multi-source sensing module acquires multi-source sensor data based on a multi-source sensor array. The data transmission module transmits the multi-source sensor data based on a hybrid communication network of 5G and industrial IoT. The intelligent inspection module is used to fuse the multi-source sensor data and perform AI analysis. The intelligent application module provides real-time monitoring dashboards, defect classification and early warning, automatic generation of inspection reports, and blockchain evidence storage functions based on a terminal interactive interface.
[0006] Further improvements are made in that: the multi-source sensing module includes a non-contact sensing unit, a contact sensing unit, and a motion detection unit. The non-contact sensing unit includes a lidar, an infrared thermal imager, a millimeter-wave radar, and a high-definition camera, which are used to acquire point cloud data of building surfaces, monitor temperature distribution data, distance, deformation, and reflection data of concealed targets on structural surfaces, and visible / near-infrared dual-spectral images, respectively.
[0007] The contact sensing unit includes an embedded fiber optic grating sensor, a piezoelectric ceramic accelerometer, a strain gauge, a temperature sensor, and a humidity sensor, which are used to acquire internal strain data of concrete, structural vibration acceleration data, static / dynamic strain data of the structure, internal / ambient temperature data of building materials, and relative humidity data, respectively.
[0008] The mobile detection unit includes a drone equipped with lightweight sensors and a wall-climbing robot, which are used for high-altitude and large-span area inspection and vertical wall inspection, respectively.
[0009] Further improvements are made in that the data transmission module includes an edge computing node unit, which is used to preprocess and perform local real-time analysis on multi-source sensor data. Specifically, it performs noise reduction, normalization and outlier removal on the data, and performs local real-time analysis on key data to trigger a primary early warning.
[0010] Further improvements are made in that: the intelligent detection module includes a data fusion unit, a model library unit, and a knowledge graph unit. The data fusion unit is based on a multi-source data fusion engine and integrates the data obtained by the multi-source perception module through a spatiotemporal registration algorithm to construct a digital twin of the building structure. The model library unit is used to store pre-trained defect identification models and material performance evaluation models. The knowledge graph unit is used to store historical detection cases, building code standards, and material property databases.
[0011] Further improvements are made in the following ways: The construction of the digital twin is specifically carried out by first aligning the point cloud data acquired by the lidar at different time periods using the ICP point cloud registration algorithm, then integrating the infrared thermal imaging temperature data, fiber optic strain data and contact sensing unit data through timestamps, and finally correcting the geometric deviation and stress field distribution between the digital twin and the actual structure based on finite element simulation to complete the high-fidelity mapping of the internal structure.
[0012] Further improvements are made in that: the defect identification model includes a crack / hole visual identification model based on YOLOv8, a stress time series prediction model based on LSTM, and a microcrack identification model based on transfer learning; the material performance evaluation model includes a concrete strength inversion model and a steel corrosion rate prediction model.
[0013] Further improvements are made in that: the input of the crack / hole visual recognition model based on YOLOv8 is a visible light / near infrared dual-spectrum image captured by a high-definition camera, which is cropped into a ROI region containing the building surface by edge computing node units, and the output is the type, location information, size parameters and confidence level of the defect;
[0014] The input to the LSTM-based stress time series prediction model is the strain time series data collected by the contact sensing unit, which is a three-dimensional tensor in the format of [number of samples, time step, number of features]. The output is the stress prediction value sequence and stress change trend label for the next 24 hours.
[0015] The microcrack identification model based on transfer learning takes a high-resolution image after image enhancement as input and outputs the existence, location and size of microcracks.
[0016] The input to the concrete strength inversion model is sensor data from embedded fiber optic grating sensors and temperature sensors, as well as engineering parameters such as concrete mix proportion and pouring time. The output is the estimated value of the concrete cube compressive strength.
[0017] The steel corrosion rate prediction model takes environmental data, electrochemical data, and image data as input and outputs the annual corrosion rate and corrosion risk level of the steel.
[0018] Further improvements include: the intelligent detection module also includes a dynamic risk assessment unit, which constructs a risk propagation model based on a Bayesian network, inputs real-time monitoring data, outputs the structural security level, and pushes early warning information to the project leader via SMS / APP. The security levels include Level I Normal, Level II Attention, Level III Warning, and Level IV Danger.
[0019] Further improvements include: the detection report is automatically generated including the three-dimensional coordinates of the defect marked by the digital twin, the classification of the defect severity, and the recommended repair scheme based on the knowledge graph combined with the intelligent detection module. The generated report is automatically archived to the blockchain for evidence storage.
[0020] The beneficial effects of this invention are as follows: This invention achieves automation of the detection process and intelligent data processing through multi-source sensing fusion, AI intelligent analysis and digital twin modeling, significantly improving detection efficiency and coverage, enhancing the accuracy and reliability of surface and hidden defect identification, extending the early warning time of potential risks, and strengthening data traceability and immutability through blockchain evidence storage, providing an efficient, accurate and reliable full-cycle solution for the quality control of construction projects. Attached Figure Description
[0021] Figure 1 This is a system architecture diagram of the present invention.
[0022] Figure 2 This is a flowchart of the method of the system of the present invention. Detailed Implementation
[0023] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.
[0024] according to Figure 1 and Figure 2 As shown in the figure, this embodiment provides an intelligent inspection system for building engineering quality, including a multi-source sensing module, a data transmission module, an intelligent inspection module, and an intelligent application module.
[0025] The multi-source sensing module acquires multi-source sensing data based on a multi-source sensor array, including a non-contact sensing unit, a contact sensing unit, and a motion detection unit.
[0026] The non-contact sensing unit includes a lidar, an infrared thermal imager, a millimeter-wave radar, and a high-definition camera. The high-definition camera supports visible light / near-infrared dual spectrum, which is used to acquire point cloud data of building surface with an accuracy of ±2mm, monitor temperature distribution data, distance, deformation and reflection data of structural surface and concealed targets, and visible light / near-infrared dual spectrum images, respectively.
[0027] The contact sensing unit includes an embedded fiber optic grating sensor, a piezoelectric ceramic accelerometer, a strain gauge, a temperature sensor, and a humidity sensor, which are used to acquire internal strain data of concrete, structural vibration acceleration data, static / dynamic strain data of the structure, internal / ambient temperature data of building materials, and relative humidity data, respectively.
[0028] The mobile inspection unit includes drones equipped with lightweight sensors and wall-climbing robots, which are used for high-altitude and large-span area inspections and vertical wall inspections, respectively.
[0029] The sensor deployment scheme is based on the architectural design drawings (BIM model). The sensors are densely deployed at key locations such as beam-column joints, mid-span of floor slabs, steel structure welds, and concrete pouring surfaces, with a sensor spacing of ≤500mm.
[0030] The data transmission module transmits multi-source sensor data based on a hybrid communication network of 5G and industrial IoT, supporting low-latency, high-bandwidth data transmission. It includes edge computing node units for preprocessing and local real-time analysis of multi-source sensor data. Specifically, it performs noise reduction, normalization, and outlier removal on the data, and conducts local real-time analysis of key data such as stress mutations and temperature exceedances to issue preliminary early warnings.
[0031] The intelligent detection module is used to fuse multi-source sensor data and perform AI analysis, including a data fusion unit, a model library unit, and a knowledge graph unit;
[0032] The data fusion unit is based on a multi-source data fusion engine. It integrates data acquired by multi-source sensing modules through a spatiotemporal registration algorithm to construct a digital twin of the building structure. Specifically, the construction of the digital twin involves first aligning point cloud data acquired by lidar at different time periods using an ICP point cloud registration algorithm, then synchronously integrating infrared thermal imaging temperature data, fiber optic strain data, and contact sensing unit data through timestamps, and finally correcting the geometric deviation and stress field distribution between the digital twin and the actual structure based on finite element simulation to complete a high-fidelity mapping of the internal structure.
[0033] The model library unit is used to store pre-trained defect recognition models and material performance evaluation models. The defect recognition models include a crack / hole visual recognition model based on YOLOv8, a stress time series prediction model based on LSTM, and a microcrack recognition model based on transfer learning. The material performance evaluation models include a concrete strength inversion model based on strain-strength mapping relationship and a steel corrosion rate prediction model based on environmental temperature, humidity and electrochemical data.
[0034] The input to the YOLOv8-based crack / hole visual recognition model is a visible light / near infrared dual-spectrum image captured by a high-definition camera. After being cropped into ROI regions containing the building surface by edge computing node units, the output includes the type, location information, size parameters, and confidence level of the defect, expressed by the following formula.
[0035]
[0036] Where IoU is the intersection-over-union ratio, and the predicted bounding box B = (x, y, w, h) and the ground truth bounding box Bgt = (x, y, w, h) are compared. gt y gt w gt h gt The degree of overlap; ρ 2 (b, b) gt ) represents the center of the predicted bounding box (x, y) and the center of the ground truth bounding box (x, y). gt y gt The squared Euclidean distance of ), c is the length of the diagonal of the smallest bounding rectangle containing both frames; v represents the aspect ratio consistency measure. α represents the weighting coefficient.
[0037] The input to the LSTM-based stress time series prediction model is the strain time series data collected by the contact sensing unit, which is a three-dimensional tensor in the format of [number of samples, time step, number of features]. The output is the stress prediction value sequence and stress change trend label for the next 24 hours, expressed by the following formula.
[0038]
[0039] Where x t This indicates the input strain data at time t; ht W represents the hidden state of the LSTM at time t; lstm b lstm The weight matrix and bias of the LSTM; For the future T-hour stress prediction sequence; W o b o The weights and biases for the fully connected output layer.
[0040] The microcrack identification model based on transfer learning takes a high-resolution image after image enhancement as input and outputs the existence, location and size of microcracks, expressed by the following formula;
[0041] P(y|x) = Softmax(W fc ·F(x)+b fc )
[0042] Where x is the input high-resolution image; F(x) is the image features extracted by the ResNet-50 backbone network; W fc b fc The weights and biases of the fine-tuned 2-layer fully connected classifier are: P(y|x) represents the probability distribution of the output, where y=1 means "with micro-cracks" and y=0 means "without", and a confidence level ≥ 0.9 indicates validity.
[0043] The input to the concrete strength inversion model is sensor data from embedded fiber optic grating sensors and temperature sensors, as well as engineering parameters such as concrete mix proportion and pouring time. The output is the estimated value of the concrete cube compressive strength, expressed by the following formula.
[0044] f c =[(A / w)-B]+MLP(ε, T, t, w, θ)
[0045] Where f c ε represents the estimated compressive strength of the concrete cube; A and B represent the empirical coefficients of Abrams' water-cement ratio law; w is the water-cement ratio; ε is the fiber grating strain; T represents the temperature; t represents the age; MLP(·, θ) is the multilayer perceptron.
[0046] The input to the steel corrosion rate prediction model is environmental data, electrochemical data, and image data. The output is the annual corrosion rate and corrosion risk level of the steel, expressed by the following formula.
[0047]
[0048] Where v represents the annual corrosion rate; N=100 represents the number of trees in the random forest, and T i Let X represent the i-th decision tree; X represents the input feature vector. Represents the splitting rules and weight parameters of the i trees; g(E) hum, T) represents the electrochemical correction function; k represents the weighting coefficient of the mechanism term.
[0049] The knowledge graph unit is used to store historical inspection cases, building codes and standards, and material property databases, supporting semantic retrieval and reasoning.
[0050] The intelligent detection module also includes a dynamic risk assessment unit, which builds a risk propagation model based on a Bayesian network. It takes in real-time monitoring data such as concrete shrinkage rate and ambient temperature and humidity as input, outputs the structural safety level, and pushes early warning information to the project manager via SMS / APP. The safety levels include Level I Normal, Level II Attention, Level III Warning, and Level IV Danger.
[0051] The intelligent application module provides real-time monitoring dashboards, defect classification and early warning, automatic generation of inspection reports, and blockchain evidence storage functions based on the terminal interactive interface;
[0052] The inspection report is automatically generated, including the three-dimensional coordinates of the defect based on the digital twin annotation error ≤5cm, the severity level of the defect (minor / moderate / severe), and the recommended repair solution based on the knowledge graph of the intelligent inspection module, such as surface sealing for microcracks and pressure grouting for cavities. The generated report is automatically archived to the blockchain for evidence storage.
[0053] The testing methods of intelligent testing systems for building engineering quality include:
[0054] Step 1: Initialize the configuration. Based on the BIM model, plan the sensor deployment scheme, deploy sensors in key locations, and establish an initial digital twin.
[0055] Step 2: Data Acquisition and Transmission. The sensor collects data at a set frequency (e.g., once every 30 minutes), and the edge computing node preprocesses the data before uploading it to the platform layer.
[0056] Step 3: Multi-source fusion and digital twin update. The platform layer integrates data through a spatiotemporal registration algorithm and uses finite element simulation to correct the digital twin.
[0057] Multi-source fusion specifically includes aligning lidar point cloud data with ICP algorithm and synchronizing it with infrared thermal imaging temperature data and fiber optic strain data according to timestamps.
[0058] Data noise is eliminated by Kalman filtering, and a dynamic digital twin containing geometry, temperature field, and stress field is constructed.
[0059] Step 4: Defect identification and risk assessment. Use AI models to identify surface / hidden defects and predict potential risks by combining the stress concentration areas of twins.
[0060] Defect identification includes: identifying surface defects such as cracks and holes using high-definition camera images and a YOLOv8 model. Cracks wider than 0.3 mm are deemed to require treatment, and holes with an area greater than 100 mm² are also identified. 2 The defect is classified as serious. Internal voids are predicted by combining fiber optic strain abrupt change (such as local strain > 150% of the design value) with an LSTM model, or hidden defects such as non-dense areas of concrete are identified by infrared thermal imaging in low-temperature zones.
[0061] Step 5: Results Feedback and Optimization. Generate a visual report and optimize sensor deployment strategy based on historical data.
[0062] The feedback results include: real-time monitoring dashboards displaying defect distribution heatmaps and safety level change curves; early warning information being pushed according to level, with Level II and above pushed to project managers and Level III and above pushed to enterprise leaders; and reports being automatically archived to the blockchain for evidence storage to ensure that the data is tamper-proof.
[0063] Optimization strategies include: increasing the density of sensor deployment in areas with high-frequency triggering of early warnings, such as beam-column nodes that trigger warnings ≥2 times per month, reducing the spacing between sensors to 200mm, or changing the detection method, such as adding radar scanning.
[0064] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. An intelligent inspection system for building engineering quality, characterized in that: It includes a multi-source sensing module, a data transmission module, an intelligent detection module, and an intelligent application module. The multi-source sensing module acquires multi-source sensing data based on a multi-source sensor array. The data transmission module transmits the multi-source sensing data based on a hybrid communication network of 5G and industrial IoT. The intelligent detection module is used to fuse the multi-source sensing data and perform AI analysis. The intelligent application module provides real-time monitoring dashboards, defect classification and early warning, automatic generation of inspection reports, and blockchain evidence storage functions based on a terminal interactive interface.
2. The intelligent inspection system for building engineering quality according to claim 1, characterized in that: The multi-source sensing module includes a non-contact sensing unit, a contact sensing unit, and a motion detection unit. The non-contact sensing unit includes a lidar, an infrared thermal imager, a millimeter-wave radar, and a high-definition camera, which are used to acquire point cloud data of building surfaces, monitor temperature distribution data, distance, deformation, and reflection data of concealed targets on structural surfaces, and visible / near-infrared dual-spectrum images, respectively. The contact sensing unit includes an embedded fiber optic grating sensor, a piezoelectric ceramic accelerometer, a strain gauge, a temperature sensor, and a humidity sensor, which are used to acquire internal strain data of concrete, structural vibration acceleration data, static / dynamic strain data of the structure, internal / ambient temperature data of building materials, and relative humidity data, respectively. The mobile detection unit includes a drone equipped with lightweight sensors and a wall-climbing robot, which are used for high-altitude and large-span area inspection and vertical wall inspection, respectively.
3. The intelligent inspection system for building engineering quality according to claim 1, characterized in that: The data transmission module includes an edge computing node unit, which is used to preprocess and perform local real-time analysis on multi-source sensor data. Specifically, it performs noise reduction, normalization, and outlier removal on the data, and performs local real-time analysis on key data to trigger preliminary early warnings.
4. The intelligent inspection system for building engineering quality according to claim 1, characterized in that: The intelligent detection module includes a data fusion unit, a model library unit, and a knowledge graph unit. The data fusion unit is based on a multi-source data fusion engine and integrates the data obtained by the multi-source sensing module through a spatiotemporal registration algorithm to construct a digital twin of the building structure. The model library unit is used to store pre-trained defect identification models and material performance evaluation models. The knowledge graph unit is used to store historical detection cases, building code standards, and material property databases.
5. The intelligent inspection system for building engineering quality according to claim 4, characterized in that: The construction of the digital twin involves first aligning the point cloud data acquired by the lidar at different time periods using the ICP point cloud registration algorithm, then synchronously integrating infrared thermal imaging temperature data, fiber optic strain data, and contact sensing unit data through timestamps, and finally correcting the geometric deviation and stress field distribution between the digital twin and the actual structure based on finite element simulation to complete a high-fidelity mapping of the internal structure.
6. The intelligent inspection system for building engineering quality according to claim 4, characterized in that: The defect identification model includes a crack / hole visual identification model based on YOLOv8, a stress time series prediction model based on LSTM, and a microcrack identification model based on transfer learning; the material performance evaluation model includes a concrete strength inversion model and a steel corrosion rate prediction model.
7. The intelligent inspection system for building engineering quality according to claim 6, characterized in that: The input to the YOLOv8-based crack / hole visual recognition model is a visible light / near infrared dual-spectrum image captured by a high-definition camera. After being cropped into a ROI region containing the building surface by edge computing node units, the output is the type, location information, size parameters and confidence level of the defect. The input to the LSTM-based stress time series prediction model is the strain time series data collected by the contact sensing unit, which is a three-dimensional tensor in the format of [number of samples, time step, number of features]. The output is the stress prediction value sequence and stress change trend label for the next 24 hours. The microcrack identification model based on transfer learning takes a high-resolution image after image enhancement as input and outputs the existence, location and size of microcracks. The input to the concrete strength inversion model is sensor data from embedded fiber optic grating sensors and temperature sensors, as well as engineering parameters such as concrete mix proportion and pouring time. The output is the estimated value of the concrete cube compressive strength. The steel corrosion rate prediction model takes environmental data, electrochemical data, and image data as input and outputs the annual corrosion rate and corrosion risk level of the steel.
8. The intelligent inspection system for building engineering quality according to claim 1, characterized in that: The intelligent detection module also includes a dynamic risk assessment unit, which constructs a risk propagation model based on a Bayesian network, inputs real-time monitoring data, outputs the structural security level, and pushes early warning information to the project leader via SMS / APP. The security levels include Level I Normal, Level II Attention, Level III Early Warning, and Level IV Danger.
9. The intelligent inspection system for building engineering quality according to claim 1, characterized in that: The detection report is automatically generated, including the three-dimensional coordinates of the defect marked by the digital twin, the classification of the defect severity, and the recommended repair scheme based on the knowledge graph combined with the intelligent detection module. The generated report is automatically archived to the blockchain for evidence storage.