Multi-modal sensor array-based intelligent detection system and method for leakage of large-thickness bottom plate of existing building

Abstract

The invention belongs to the technical field of leakage detection, and particularly relates to an existing building large-thickness bottom plate leakage intelligent detection system and method based on a multi-mode sensor array. According to the technical scheme, the existing building large-thickness bottom plate leakage intelligent detection system based on the multi-mode sensor array comprises a cloud analysis server and an on-site mobile detection platform; the cloud server performs deep learning reasoning, big data analysis and digital twin modeling, and realizes data interaction and cooperative work with the field mobile detection platform through 5G or WiFi wireless communication; the field mobile detection platform serves as an operation execution unit and integrates an intelligent mobile trolley, an active excitation subsystem, a multi-modal sensor array subsystem and an edge calculation unit. The invention provides an intelligent detection system and method for leakage of a large-thickness bottom plate of an existing building based on a multi-mode sensor array, and aims to realize full-thickness detection coverage from a surface layer to a deep layer, span from qualitative judgment to quantitative evaluation, and change from manual operation to intelligence and automation.

Description

Technical Field

[0001] This invention belongs to the field of leakage detection technology, and specifically relates to an intelligent detection system and method for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array. Background Technology

[0002] Detecting leakage in the thick foundation slabs of existing buildings faces a series of technical challenges, which severely restrict the accuracy and efficiency of leakage diagnosis.

[0003] Traditional infrared detection technology suffers from severely insufficient detection depth, typically only able to detect temperature anomalies within 50 millimeters of the surface. However, the foundation slabs of existing buildings such as basements, subway stations, and civil defense projects are generally 300 to 800 millimeters thick, and leaks often occur deep within or at the bottom of the slab, where surface infrared signals are weak or even undetectable. This limitation in detection depth leads to a large number of deep leaks going undetected, and by the time water seeps to the surface, severe damage has often already occurred, missing the optimal time for repairs.

[0004] Insufficient positioning accuracy is another prominent problem. While existing technologies can roughly determine the area of ​​leakage, they have significant shortcomings in accurately locating the source of the leak. Planar positioning errors are typically over 100 millimeters, and there is a lack of effective positioning methods in the depth direction, making it impossible to determine whether the leak is occurring in the upper, middle, or bottom of the base slab. This uncertainty in positioning greatly hinders maintenance work, forcing workers to rely on experience to blindly excavate or treat large areas, wasting materials and time, and potentially damaging the normal structure.

[0005] The lack of quantitative assessment capabilities severely impacts leak management decisions. Traditional detection methods mostly only provide qualitative descriptions such as "minor leak," "moderate leak," and "severe leak," failing to quantify the degree and volume of leakage. Property management and maintenance companies struggle to develop reasonable repair plans and budgets based on this information, and insurance claims often become embroiled in disputes due to the lack of quantitative data.

[0006] Poor environmental adaptability is a significant factor limiting the reliability of detection. Existing building foundations are often covered with decorative materials such as tiles and floor paint, which alter surface thermal and reflective properties, interfering with detection signals. Changes in ambient temperature and humidity also significantly affect detection results, especially since passive infrared detection is highly dependent on diurnal temperature variations and is almost impossible to operate in cloudy or rainy weather.

[0007] Low testing efficiency increases both testing costs and time costs. Traditional methods require multiple on-site tests, each followed by complex data interpretation and analysis by professionals, often taking several days or even weeks for the entire testing cycle. For continuously operating facilities such as subway stations, the testing window is extremely limited, and low efficiency means that large-scale testing cannot be completed.

[0008] The lack of leakage path tracing capabilities leads to a widespread problem of treating the symptoms but not the root cause. Current technology can only detect surface signs of leakage, but cannot trace where the water entered from or what path it took to reach the detection location. This results in repairs often only addressing the surface symptoms, while the true source and path of the leakage remain undiscovered and untreated. As a result, the leakage quickly reappears in other locations, creating a situation where "repairing is never enough."

[0009] Insufficient automation leads to a high reliance on professional personnel. Existing testing methods require experienced experts at every stage, from data collection and analysis to diagnosis and treatment plan development. Human interpretation is highly subjective, resulting in inconsistent results and is prone to misjudgment due to fatigue or negligence. The shortage of professional personnel and their high cost also limit the widespread availability of testing services.

[0010] The difficulty in identifying early-stage micro-leakage makes preventative maintenance challenging. Passive detection methods produce very weak signals for early-stage micro-leakage with low moisture content and small leakage volume, often drowned out by background noise and undetectable. By the time the leakage develops to a detectable level, irreversible deterioration has often occurred inside the concrete, significantly increasing the difficulty and cost of repairs. Summary of the Invention

[0011] In order to solve the above-mentioned problems in the existing technology, the purpose of this invention is to provide an intelligent detection system and method for leakage of thick foundation slabs of existing buildings based on a multimodal sensor array, so as to achieve full thickness detection coverage from the surface to the deep layer, a leap from qualitative judgment to quantitative assessment, and a transformation from manual operation to intelligent automation.

[0012] The technical solution adopted in this invention is as follows: A smart detection system for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array includes a cloud analysis server and a field mobile detection platform. The cloud server performs deep learning inference, big data analysis, and digital twin modeling, and achieves data interaction and collaborative work with the field mobile detection platform through 5G or WiFi wireless communication. The field mobile detection platform, as the operation execution unit, integrates an intelligent mobile vehicle, an active excitation subsystem, a multimodal sensor array subsystem, and an edge computing unit.

[0013] As a preferred embodiment of the present invention, the multimodal sensor array subsystem includes an infrared thermal imager, a ground penetrating radar, an ultrasonic phased array, a microwave sensor, and a vibration sensor; the infrared thermal imager is used for surface detection, the ground penetrating radar achieves deep penetration, the ultrasonic phased array provides high-resolution focused imaging, the microwave sensor measures water content, and the vibration sensor acquires the dynamic response of the structure.

[0014] As a preferred embodiment of the present invention, the active excitation subsystem includes a heating device, an ultrasonic excitation source, and a water pressure simulation device. The active excitation subsystem improves the detection sensitivity of microleakage by applying energy perturbation.

[0015] As a preferred embodiment of the present invention, the edge computing unit is responsible for real-time processing and preliminary diagnosis of field data, converting raw sensor signals into structured feature data, and realizing rapid response at the edge and reasonable allocation of cloud computing load.

[0016] A smart detection method for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array includes the following steps: S1: Infrared detection: Passive multi-temporal detection: 5-6 AM, 2-3 PM, 6-7 PM; Active thermal excitation: heating for 15-30 min, cooling for 30-60 min; Feature extraction: temperature gradient. T, time derivative T / t, thermal time constant τ; S2: GPR detection: multi-frequency scanning: 800MHz coarse scan, 1.5GHz fine scan, 2.5GHz detailed scan; three-dimensional imaging: B-scan profile, C-scan plane, 3D reconstruction; water content inversion: dielectric constant εr, volumetric water content θ; S3: Ultrasonic testing: Phased array focusing: depth 100-600mm, electronic scanning; Echo analysis: time difference Δt, amplitude A, spectrum f; SAFT imaging: improved resolution; S4: Microwave humidity measurement: array measurement: 5×5 sensor, penetration 100-200mm; gradient analysis: spatial gradient θ, high-value region identification; S5: Vibration Analysis: Passive Environmental Vibration: Multi-point synchronous acquisition, 1-500Hz; Active Ultrasonic Excitation: 40-100kHz frequency sweep; Anomaly Identification: Damping α variation, transfer function H(ω). S6: Multimodal data fusion: Spatial registration: accuracy <10mm; Feature extraction: 41-dimensional vector, infrared 12-dimensional, GPR 8-dimensional, ultrasound 10-dimensional, microwave 5-dimensional, vibration 6-dimensional; Deep learning fusion: multi-head attention, Transformer encoder, multi-task output; Physical constraint correction: thermal diffusion equation, seepage mechanics, wave equation. S7: 3D Reconstruction: Point Cloud Generation: Voxel 50mm 3 Color mapping; Path tracing: A* algorithm, gradient tracing, BIM integration, AR display; S8: Quantitative Assessment: Comprehensive Index ISI: =Σwi•Ii; Leakage Classification: Class I: 0-25, Class II: 25-50, Class III: 50-75, Class IV: 75-100; Leakage Estimation: Q=k•i•A, accuracy ±15%; S9: Intelligent Diagnosis: Cause Diagnosis: Bayesian Network, Decision Tree; Repair Solutions: Technology Selection, Cost Estimation, and Schedule Planning.

[0017] As a preferred embodiment of the present invention, in step S1: Passive detection parameter settings: Acquisition time period: 5-6 AM, 2-3 PM, 6-7 PM; Acquisition distance: 1-2 meters; Number of frames per point ≥ 10; Temperature difference requirement > 5℃; Movement step: 1 meter / step; Active excitation parameter configuration: target temperature rise: +10-15℃; heating time: 15-30 minutes; acquisition interval: 30 seconds / time; cooling observation time: 30-60 minutes; acquisition interval: 60 seconds / time; heating power: 500-2000W.

[0018] As a preferred embodiment of the present invention, in step S2, a multi-frequency adaptive scanning strategy is adopted: Coarse scan: frequency 800MHz; detection depth >500mm; resolution ~150mm; grid spacing 0.5m; scan area / hour: 200-300m² 2 Application objectives: Rapid comprehensive survey and identification of large-scale anomalies; Fine scanning: frequency 1.5GHz; detection depth 300-400mm; resolution ~80mm; grid spacing 0.2m; scanning area / hour: 100-150m² 2 Application objective: Mid-level of abnormal areas, precise defect location; Fine scanning: frequency 2.5GHz; detection depth <300mm; resolution ~50mm; grid spacing 0.1m; scanning area / hour: 50-80m² 2 Application target: surface imaging of key areas, high-resolution imaging.

[0019] As a preferred embodiment of the present invention, in step S3, the phased array focusing parameters are configured as follows: 64-element configuration: element spacing 2mm; center frequency 100kHz; pulse width 2-5 cycles; depth of focus 100-600mm; focal length step 50mm; angle scan ±30°; sampling rate ≥10MHz; 128-element configuration: element spacing 1mm; center frequency 150kHz; pulse width 2-3 cycles; focusing depth 100-600mm; focal length step 50mm; angle scanning ±45°; sampling rate ≥20MHz.

[0020] As a preferred embodiment of the present invention, in step S4, the microwave sensor array is configured as follows: Standard configuration: Array size 5×5; sensor spacing 0.5m; measurement time / point 2 seconds; total measurement time 50 seconds; coverage area 4m² 2 Spatial resolution: 0.5m; Applicable scenarios: routine inspection; High-precision configuration: 7×7 array size; sensor spacing 0.3m; measurement time / point 3 seconds; total measurement time 147 seconds; coverage area 3.6m². 2 Spatial resolution: 0.3m; Applicable scenarios: critical areas; Fast scanning configuration: 3×3 array size; sensor spacing 0.8m; measurement time / point 1 second; total measurement time 9 seconds; coverage area 3.2m². 2 Spatial resolution 0.8m; Applicable scenario: preliminary screening.

[0021] As a preferred embodiment of the present invention, in step S5, the vibration test configuration parameters are as follows: Passive environmental vibration: Excitation source: environmental vibration; Sensor type: MEMS triaxial accelerometer; Arrangement: grid pattern, spacing 0.5-1m; Sampling rate 2kHz; Acquisition time 5-10 minutes; Sensitivity >1000mV / g; Dynamic range >60dB; Measurement frequency band 1-500Hz; Application objective: low-frequency overall response, damping identification; Active ultrasonic excitation: Excitation source: ultrasonic excitation source, 40-100kHz frequency sweep; Sensor type: MEMS triaxial accelerometer; Arrangement: grid pattern, spacing 0.5-1m; Sampling rate: 500kHz; Acquisition duration: 0.5 seconds per frequency point; Sensitivity >1000mV / g; Dynamic range >60dB; Measurement frequency band: 40-100kHz; Application target: high-frequency local defects, modal analysis.

[0022] The beneficial effects of this invention are as follows: 1. Regarding improved detection capabilities, this invention aims to achieve blind-spot-free detection of base plates with thicknesses ranging from 300 to 800 mm. By integrating five sensors based on different principles, it fully leverages the advantages of infrared thermal imaging's high sensitivity to the surface, ground-penetrating radar's strong penetration to medium and deep layers, ultrasonic phased array's high-resolution focusing, microwave sensors' precise moisture measurement, and vibration sensors' full-thickness fluctuation characteristics to construct a multimodal detection system that is deeply complementary and integrates information. This ensures that regardless of the depth at which leakage occurs in the base plate, it can be effectively detected by at least two types of sensors, eliminating detection blind spots.

[0023] 2. High-precision positioning is a key requirement for engineering applications. This invention requires a planar positioning accuracy of within 50 mm and a depth positioning error controlled within ±30 mm. A combined positioning technology of laser ranging and inertial navigation achieves centimeter-level precision calibration of the sensor position. Using ultrasonic phased array electronic focusing and SAFT synthetic aperture imaging technology, the depth resolution is improved to the 30 mm level. Combining ground-penetrating radar time-depth conversion and velocity field inversion, a precise three-dimensional coordinate system is established. Such positioning accuracy provides a reliable spatial reference for maintenance and construction, avoiding blind excavation and over-processing.

[0024] 3. The establishment of a quantitative assessment system is a key innovation of this invention. Traditional methods can only provide qualitative descriptions such as "minor," "moderate," and "severe." This invention aims to achieve quantitative assessment through physical models and mathematical algorithms. First, a Comprehensive Leakage Index (ISI) is established, based on weighted fusion of multimodal characteristics, mapped to a quantitative range of 0 to 100, clearly defining the dividing points for the four levels. More importantly, it must be able to estimate the actual leakage volume in liters per square meter per day, with an estimation accuracy within ±15%. This index has clear engineering significance, can be directly used for the design of maintenance plans and cost budgeting, and also provides a quantitative basis for insurance claims.

[0025] 5. In terms of functional expansion, the 3D reconstruction and path tracing of leakage channels are the outstanding features of this invention. Utilizing 3D volumetric data fused from multimodal data, voxel-based modeling is employed, assigning a leakage severity index to each 50 cubic millimeter voxel. A continuous path with a high ISI value is searched in 3D space, and a combination of the A* algorithm and gradient tracing is used to reconstruct the complete 3D channel from the leakage source to the surface manifestation point. Visualization employs color mapping and transparency control to clearly show the expansion pattern of the leakage within the base slab. This path tracing capability allows maintenance personnel to "see through" the concrete, find the true source of the leakage, and achieve a fundamental solution rather than merely treating the symptoms.

[0026] 5. Intelligent diagnosis is key to improving the level of intelligent detection. This invention aims to achieve automated reasoning from leakage phenomena to cause analysis. A Bayesian network model is established, using observed multimodal features as evidence. Through probabilistic reasoning, the most likely causes of leakage are derived, such as construction cold joints, structural cracks, pipe leaks, and aging of the waterproofing layer, and the probability distribution of each cause is provided. Decision trees assist in refining the diagnosis, quickly locating the specific cause type based on the leakage location, depth, and feature combinations. This automated diagnostic capability reduces reliance on expert experience and improves the objectivity and consistency of diagnosis.

[0027] 6. The automated generation of maintenance plans further enhances the system's practical value. Based on diagnostic results, matching maintenance plans are retrieved from the rule base and case base, including complete information such as technology selection, material list, construction steps, and quality control points. Material usage and project costs are automatically calculated, and the critical path method is used to plan the construction schedule, providing a time estimate. The generated plans are not simply template-based but customized according to actual inspection data, ensuring the plans' relevance and operability. This automated plan generation can significantly shorten the time from inspection to maintenance decision-making, improving response efficiency.

[0028] 7. Performance optimization emphasizes improved detection efficiency. Existing methods often require 20 to 30 minutes to detect a single measurement point. This invention, through automatic navigation of a mobile platform, parallel acquisition by multiple sensors, and real-time preprocessing via edge computing, reduces the time for a single measurement point to less than 10 minutes. The overall detection efficiency is 3 to 5 times higher than traditional methods. This efficiency improvement is crucial for scenarios such as subway stations where detection needs to be completed within a short time window at night.

[0029] 8. Enhanced environmental adaptability is reflected in several aspects. The introduction of active excitation technology is a key innovation. By artificially applying a temperature gradient through a controllable heating device, it no longer passively relies on diurnal temperature differences, enabling detection to be carried out at any time and under any environmental conditions. The multi-frequency adaptive scanning strategy allows the ground-penetrating radar to automatically adjust its operating frequency according to the detection depth, finding the optimal balance between depth and accuracy. The electronic focusing of phased array ultrasound overcomes the influence of the decorative layer, allowing direct detection of the substrate through materials such as tiles and flooring. The comprehensive application of these technologies gives the system stable detection capabilities in all weather conditions and under all operating conditions.

[0030] 9. Significantly improved automation is another key aspect of performance optimization. From data acquisition, quality control, feature extraction, fusion analysis to result presentation, the automation level of the entire testing process should reach over 90%. Operators only need to set the testing area and parameters; the system automatically completes path planning, sensor control, data acquisition, and real-time feedback. Abnormal situations during the testing process, such as sensor malfunctions, positioning drift, and environmental interference, can be automatically identified and addressed by the system, or an alarm can be issued to the operator. Upon completion of the testing, a structured report is automatically generated, including complete content such as testing data, analysis results, 3D model, diagnostic conclusions, and maintenance suggestions, eliminating the need for manual refactoring.

[0031] 10. Enhancing application value is the ultimate goal pursued by this invention. Non-destructive testing is a basic requirement; the entire testing process does not require core drilling or excavation, does not damage the decoration or structure at all, and the site can be immediately restored to use after the testing is completed, without affecting normal production and living order. This not only saves on decoration and repair costs, but more importantly, avoids interference with the owner and improves the acceptability of the testing service.

[0032] 11. The realization of preventative maintenance capabilities is of great significance. Through proactive stimulation and highly sensitive multimodal detection, early micro-leakage with a moisture content of only 3% to 5% can be identified. At this stage, the concrete has not yet undergone significant deterioration, and repair measures are simple and inexpensive. Regular testing can establish a health record for the foundation slab, track the development trend of leakage, and intervene in time before the problem worsens. This shift from reactive repair to preventative maintenance can significantly extend the service life of buildings and reduce total life cycle costs.

[0033] 12. The significant improvement in cost-effectiveness is reflected in multiple aspects. Increased testing efficiency by 3 to 5 times means a corresponding reduction in labor and equipment operating costs. Non-destructive testing avoids drilling and renovation repairs, saving substantial expenses. Precise positioning and quantitative assessment enable more rational repair plans, avoiding over-processing and rework, and reducing overall repair costs by 30% to 50%. Attached Figure Description

[0034] Figure 1 This is the overall system architecture diagram of the present invention; Figure 2 This is a transient thermal response curve diagram - a comparison diagram of the normal region and the water-bearing region; Figure 3 It is a leakage diagnosis decision tree diagram-feature-cause mapping diagram. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0036] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the invention can be combined with each other.

[0037] like Figure 1 As shown, the system adopts a layered distributed architecture, consisting of two core components: a cloud-based analysis server and a field mobile detection platform. The cloud server undertakes high-performance computing tasks such as deep learning inference, big data analysis, and digital twin modeling, and interacts with the field platform via 5G or WiFi wireless communication to achieve data exchange and collaborative work.

[0038] The on-site mobile inspection platform, serving as the operational unit, integrates multiple functional subsystems. Its core is a multimodal sensor array subsystem, equipped with a 640×512 resolution, 20mK sensitivity infrared thermal imager for surface inspection; a ground-penetrating radar operating at frequencies from 800MHz to 2.5GHz for deep penetration; a 64 or 128-channel ultrasonic phased array for high-resolution focused imaging; a 5×5 array microwave sensor operating at 2.45GHz for precise moisture content measurement; and a MEMS vibration sensor network covering the 1 to 500Hz frequency band to acquire structural dynamic responses.

[0039] To enhance the detection signal, the system is equipped with an active excitation subsystem, including a heating device with a controllable temperature range of 30 to 80°C, an ultrasonic excitation source with a frequency of 40 to 100 kHz, and a water pressure simulation device. This artificially applied energy perturbation improves the sensitivity for detecting micro-leaks. The intelligent mobile vehicle subsystem enables autonomous operation of the platform. Its autonomous navigation function supports adjustable speeds from 0.1 to 0.5 meters per second. A three-axis gimbal provides sensor attitude adjustment, and a precision positioning system combining laser ranging and an inertial measurement unit ensures centimeter-level accuracy in measurement position.

[0040] The edge computing unit deploys Jetson Xavier processors, responsible for real-time processing and preliminary diagnosis of on-site data. It transforms raw sensor signals into structured feature data, enabling rapid response at the edge and reasonable allocation of computing load to the cloud. The entire system, through data flow and collaborative control between its subsystems, forms a complete detection chain from data acquisition and edge processing to in-depth analysis in the cloud.

[0041] 4.1 Technical parameters and detection range of multimodal sensor Table 1 shows the technical parameters and detection range of the multimodal sensor.

[0042]

[0043] 4.4.1 Implementation of Passive Detection Technology Infrared thermal imaging detection utilizes the differences in thermophysical properties between leaking areas and normal concrete. Due to the high specific heat capacity (4200 J / kg·K) and thermal conductivity (0.6 W / m·K) of water, the thermal response behavior of water-containing areas differs significantly from that of dry concrete (specific heat capacity 840 J / kg·K, thermal conductivity 1.4 W / m·K). Under natural temperature changes or active heating conditions, this difference will create identifiable temperature field distribution characteristics on the surface.

[0044] Passive detection is based on the principle of diurnal temperature variation. The first data collection period is from 5-6 AM, when the ambient temperature has dropped to its lowest point after a night of heat dissipation, and the surface temperature field of the substrate reaches a relatively stable low temperature state. Due to their higher heat capacity, the water-bearing areas experience a slower temperature drop than normal areas, appearing as relatively high-temperature zones on the thermal image. The second period is from 2-3 PM, when the accumulated heating during the day causes the surface temperature to reach its peak. Because of the heat absorption effect of water evaporation, the surface temperature of the water-bearing areas is lower than that of normal areas, creating a temperature reversal phenomenon. The third period is from 6-7 PM, during the transition from heating to heat dissipation, when the temperature gradient change is most pronounced, and abnormal heat diffusion in the leakage area is most easily identified during this stage.

[0045] The infrared thermal imager is positioned 1-2 meters from the substrate surface. This distance allows the 640×512 pixel detector's field of view to cover an area of ​​approximately 3×2 meters, with each pixel corresponding to an actual area of ​​approximately 5mm×5mm, meeting the spatial resolution requirements for leak detection. Too close a acquisition distance (<0.5 meters) results in a small field of view and low efficiency; too far (>3 meters) increases the coverage area per pixel, decreasing spatial resolution and potentially smoothing out minute leak features. At least 10 frames are acquired at each measurement point and averaged. This time-domain averaging effectively suppresses random noise, increasing the thermal sensitivity (NETD) from 20mK per frame to approximately 6mK, significantly improving the ability to identify subtle temperature anomalies.

[0046] The mobile platform moves along a gridded path with 1-meter increments, ensuring approximately 50% overlap between the fields of view of adjacent measuring points. This overlap design avoids blind spots and allows for the generation of large-area continuous thermal images through image stitching technology. A diurnal temperature difference of >5℃ is a basic requirement for passive detection. When the temperature difference is less than 3℃, the temperature contrast between the leaking area and the normal area is less than 1℃, approaching the instrument's noise level, making reliable identification difficult.

[0047] 4.4.2 Passive Detection Parameter Configuration Table 2 is the passive detection parameter configuration table.

[0048]

[0049] 4.4.3 Implementation of Active Stimulus Detection Technology Active excitation detection amplifies the difference in response between the leaking area and the normal area by artificially applying energy disturbance. The controllable heating plate uses carbon fiber or resistance wire heating elements with an adjustable power range of 500-2000W. Temperature control employs a PID closed-loop system with an accuracy of ±1℃. The standard heating plate size is 0.5m × 0.5m, covering an area of ​​0.25m². 2 For large suspected areas, multiple heating plates can be connected in parallel or a large-size heating plate of 1m×1m can be used.

[0050] The target temperature rise is set at 10-15℃ above the ambient temperature; this temperature difference range has been verified as optimal through extensive experimentation. Too small a temperature rise (<5℃) results in insufficient temperature contrast after heating, making it difficult to distinguish between leaking and normal areas; too large a temperature rise (>20℃) may lead to localized overheating stress and a significant increase in energy consumption. The heating time of 15-30 minutes is based on the concrete's thermal diffusivity α≈0.5-0.8mm. 2 The time required to diffuse heat to a depth of 50mm is calculated in seconds. Taking a 500mm thick base plate as an example, the time t=d is approximately the time required for heat to diffuse to a depth of 50mm. 2 / (4α) ≈ 1000-1500 seconds (17-25 minutes), which is sufficient time to establish a stable temperature gradient on the surface.

[0051] Infrared images were acquired every 30 seconds during the heating process to record the dynamic evolution of the temperature field. In the initial heating phase (0-5 minutes), the surface temperature rose rapidly, with a heating rate of approximately 0.6-0.8℃ / min in normal areas and only 0.3-0.5℃ / min in water-bearing areas due to the high specific heat capacity of water. In the middle heating phase (5-20 minutes), the temperature rise rate gradually decreased, approaching thermal equilibrium. In the later heating phase (20-30 minutes), the surface temperature stabilized, with normal areas reaching the target temperature, while water-bearing areas still experienced a temperature deficit of 2-5℃.

[0052] The cooling observation phase after heating is stopped is equally important. After the heating power is turned off, the surface of the substrate begins to dissipate heat to the environment. Normal areas, due to their small heat capacity and rapid heat conduction, cool at a rate of approximately 0.4-0.6℃ / minute; areas containing water, due to their strong heat storage capacity, cool at a rate of only 0.2-0.3℃ / minute, exhibiting a "slow cooling" characteristic. In the initial cooling phase (0-10 minutes after heating stops), the temperature drops most rapidly, and the temperature difference between the two types of areas widens, making this the optimal window for identifying leaks. In the later stages of cooling (10-60 minutes), the temperature difference gradually decreases, eventually approaching ambient temperature. The data acquisition interval throughout the cooling process is extended to 60 seconds to balance data volume and temporal resolution.

[0053] The heating power is automatically adjusted based on the heating plate area and the target temperature rise. For a 0.5m × 0.5m heating plate, achieving a 15℃ temperature rise typically requires 800-1200W of power. Insufficient power will result in excessively long heating times (>45 minutes), reducing detection efficiency; excessive power will make temperature control difficult and prone to overshoot.

[0054] 4.4.4 Active Excitation Parameter Configuration Table 3 is the active excitation parameter configuration table.

[0055]

[0056] Figure 2 This is a transient thermal response curve diagram - a comparison diagram of the normal region and the water-bearing region.

[0057] 4.5 Step 2: Ground Penetrating Radar Deep Exploration (Detailed Parameters) 4.5.1 Ground Penetrating Radar Electromagnetic Wave Detection Principle and Frequency Selection Ground-penetrating radar (GPR) transmits high-frequency electromagnetic pulses into concrete and infers the internal structure and water content based on the time delay and amplitude changes of the reflected echoes. The propagation speed of electromagnetic waves in a medium is v = c / √εr, where c = 0.3 m / ns is the speed of light in a vacuum, and εr is the relative permittivity. Dry concrete has an εr of approximately 4-8 and a propagation speed of 0.11-0.15 m / ns; water has an εr as high as 80, and water-bearing concrete can have an εr as high as 15-30, with the propagation speed decreasing to 0.05-0.08 m / ns. This difference in speed leads to strong reflections and time delay anomalies in water-bearing areas, forming the physical basis for leak detection.

[0058] Frequency selection follows the principle of "low frequency for deep penetration, high frequency for precise detection." The 800MHz low-frequency antenna emits a long wave with a wavelength of λ=v / f≈0.15m, offering strong penetration and capable of detecting depths >500mm. However, limited by a half-wavelength resolution of δ≈λ / 2≈75mm, it can only identify large-scale anomalies. The 1.5GHz mid-frequency antenna has a wavelength of approximately 0.08m, a penetration depth of 300-400mm, and a resolution improved to around 40mm, suitable for precise location of mid-layer defects. The 2.5GHz high-frequency antenna has a wavelength of only 0.05m; although its penetration depth is <300mm, its resolution reaches 25mm, enabling the identification of small cracks and surface segregation.

[0059] The three-frequency adaptive strategy fully utilizes the advantages of each frequency. The first round involves a full-area coarse scan at 800MHz with a grid spacing of 0.5m. 512 sampling points are collected per second along a single survey line. At a moving speed of 0.3m / s, the spatial sampling interval is approximately 0.6mm, meeting data continuity requirements. The purpose of the coarse scan is to quickly cover the entire area and identify the approximate location and depth range of strong reflection anomalies. Under typical operating conditions, this covers an area of ​​200-300m. 2 The rough sweeping of the area takes 1-1.5 hours.

[0060] The second round of scanning focuses on finer scans of the anomalous areas identified by the coarse scan. The depth d = v·t / 2 is calculated based on the two-way time delay t of the anomalous reflection. When d < 200 mm, a high frequency of 2.5 GHz is selected, with a grid spacing of 0.1 m, increasing the resolution to 25 mm and enabling clear imaging of surface crack morphology. When 200 mm ≤ d ≤ 400 mm, a medium frequency of 1.5 GHz is used with a grid spacing of 0.2 m, balancing efficiency and accuracy. When d > 400 mm, 800 MHz is continued, but the anomalous areas are scanned 3-5 times, with multiple scans stacked to improve the signal-to-noise ratio. The fine scan area is typically 20-30% of the total area, and the time required is comparable to the coarse scan.

[0061] The third round is a verification scan of key areas (such as known pipeline crossings and historical maintenance areas), with multiple frequency cross-verifications to ensure no cases are missed.

[0062] Table 4 is the parameter table for the three-round scan.

[0063]

[0064] 4.5.2 Multi-frequency adaptive scanning strategy Table 5 is a table of multi-frequency adaptive scanning strategies.

[0065]

[0066] 4.5.3 Implementation of Adaptive Frequency Selection Algorithm Algorithm Summary Core idea: Anomalies are identified based on initial scanning; frequency is adaptively selected based on depth; normal areas are skipped to save time; and cluster analysis optimizes the scanning range.

[0067] Key parameters: Initial frequency: 800MHz; electromagnetic velocity: 0.12m / ns; threshold criterion: μ+3σ; buffer radius: 0.5m; clustering parameters: ε=1m, MinPts=3.

[0068] Deep classification strategy: Surface anomalies (d<200mm): Use 2.5GHz frequency and 0.1m grid density for detailed scanning; Mid-layer anomalies (200≤d≤400mm): Perform detailed scanning using a 1.5GHz frequency and a grid density of 0.2m; Deep anomalies (d>400mm): Continue using the 800MHz frequency, maintain the original mesh, and repeat the scan 5 times for verification.

[0069] Optimization strategy: Full-area fine scan: When the proportion of abnormal points is greater than 50%, the skip strategy is canceled and a uniform high-frequency scan is performed on the entire area; Sparse fine scanning: When outliers account for less than 5% and are scattered, only clustered regions are fine-scanned, and isolated points are skipped; Adaptive fine scanning: In other cases, the frequency is selected based on the depth, and the grid is adjusted based on the density; Expected results: Compared to full-area scanning, it can save 20-30% of time and cost.

[0070] 4.5.4 Dielectric constant inversion and quantitative calculation of moisture content Measuring the two-way time delay t of electromagnetic waves is fundamental to the inversion. The radar records the total time t from transmission to reception, which includes the round-trip time of the electromagnetic wave within the concrete. Given the thickness d of the base plate and the propagation speed v = 2d / t, there is an implicit assumption: the electromagnetic wave is incident perpendicularly, and its path length is equal to twice the thickness. In actual measurements, there is a slight tilt angle between the antenna and the base plate, which needs to be corrected using IMU attitude data. The correction formula is d' = d / cos(θ), where θ is the tilt angle.

[0071] The relative permittivity is inversely derived from velocity: εr = (c / v) 2 =(ct / (2d)) 2 The measured εr of dry C30 concrete is approximately 5-7, and that of C40 is approximately 4-6. εr increases significantly after moisture absorption; for every 1% increase in moisture content (by volume), εr increases by approximately 2-3. In areas of strong reflection, εr can reach 20-40, corresponding to a moisture content of 10-20%, indicating severe leakage.

[0072] Two models were used to verify the moisture content calculation. The CRIM model (Complex Refractive Index Model) is based on a volume-weighted average: √εr = θ√εw + (1-θ)√εs, where θ is the volumetric moisture content, εw = 80 is the dielectric constant of water, and εs = 5 is the dielectric constant of dry concrete. This model has a clear physical meaning but is sensitive to aggregate type. Topp's empirical formula is θ = -0.053 + 0.0292εr - 0.00055εr. 2 +0.0000043εr 3 It is a cubic polynomial fitted based on a large amount of measured soil and concrete data, which has high accuracy but limited extrapolation ability.

[0073] In practical applications, for low to medium moisture content regions (εr < 15), the results from the two models are similar, with an error of < ±2%. For high moisture content regions (εr > 20), the CRIM model tends to overestimate, while the Topp formula is more accurate, but still has an error of ±3%. The main sources of error are differences in concrete mix proportions, aggregate types, and pore structure. To improve accuracy, it is recommended to perform core sampling of similar concrete samples before testing, measure the moisture content at several points, and calibrate the model parameters.

[0074] Table 6 shows the calculation steps.

[0075]

[0076] 4.6 Step 3: Ultrasonic Phased Array Testing (Detailed Parameters) 4.6.1 Principle of Electronic Focusing in Ultrasonic Phased Array An ultrasonic phased array consists of 64 or 128 small piezoelectric crystals (elements) arranged in a linear or planar array, with each element independently emitting and receiving ultrasonic waves. By precisely controlling the excitation timing of each element, the sound waves emitted by different elements are superimposed in phase at a certain point in space, forming a focus. This is similar to the focusing of an optical convex lens, but the "lens" in an ultrasonic phased array is electronically implemented and requires no physical movement.

[0077] The focusing principle is based on Huygens' principle and Fermat's principle. Let the target focal point F be located at a depth zf in front of the array, and let the lateral distance xi from the i-th element to the array center be the same. The acoustic path Ri from element i to focal point F is √(xi² + zf²). If all elements emit simultaneously, the sound waves arrive at point F at different times and cannot be superimposed in phase. Therefore, elements farther away need to emit earlier, while those closer need to emit later, to compensate for the path difference.

[0078] The delay calculation formula is: Δti = (Rmax - Ri) / v, where Rmax = √(xmax² + zf²) is the distance from the farthest element to point F, and v is the speed of sound in concrete (approximately 4000 m / s). After delay control, the sound waves from each element arrive at point F simultaneously, maximizing the amplitude superposition, concentrating energy, and improving detection sensitivity. At locations far from point F, the phases of the sound waves from each element are random, canceling each other out, resulting in weak signals. This spatial selectivity achieves "focusing" on point F.

[0079] By changing the setting value of the depth of focus (zf) and recalculating the delay of each element, the focus can be moved to different depths without moving the probe. This dynamic depth focusing (DDF) technology can scan substrates with a thickness of 100-600mm, requiring only adjustment of the delay parameters, and the acquisition time for each depth is less than 1 second, which is much faster than mechanical scanning.

[0080] The lateral resolution of a focused acoustic beam is determined by the array aperture and frequency: δlateral≈λ·zf / D, where λ=v / f is the wavelength and D is the effective aperture of the array. A 64-element array with a 2mm element spacing has a total aperture D=128mm; operating at 100kHz, with a wavelength λ=40mm; at a focusing depth of 300mm, the lateral resolution δ≈40×300 / 128≈94mm. Doubling the aperture of a 128-element array improves the resolution to approximately 50mm. This is far superior to a traditional single-probe array, where the beam diverges, resulting in a spot diameter >150mm at 300mm.

[0081] The longitudinal (depth) resolution is determined by the pulse width: δaxial = v·τ / 2, where τ is the pulse width. Typically, short pulses of 2-5 cycles are used; at 100kHz, τ = 20-50μs, resulting in a longitudinal resolution of 40-100mm. At higher frequencies (200kHz), this can be reduced to 20-50mm, but the penetration depth decreases accordingly.

[0082] 4.6.2 Phased Array Focusing Parameter Configuration Table 7 shows the phased array focusing parameter configuration.

[0083]

[0084] 4.6.3 Precise Calculation of Focus Delay The three-dimensional distance Ri from the i-th element (position xi, 0) to the focal point F(xf, yf, zf) is √[(xi-xf)]. 2 +(0-yf) 2 +zf²]. For a linear array, simplifying to two dimensions: Ri = √[(xi - xf)]. 2 +zf²]. The sound velocity v is taken as the longitudinal wave velocity of concrete, approximately 4000 m / s for C30 concrete and approximately 4200 m / s for C40 concrete, which decreases slightly to 3800 m / s after water is added.

[0085] The transmission delay Δti ensures that the farthest element transmits first, followed by the nearest element, with a delay range of 0-Rmax / v. Taking a 64-element array with a 2mm element spacing and a 300mm focusing depth as an example: Rmax = √(64² + 300²) ≈ 306mm, and the maximum delay Rmax / 4000 ≈ 77μs. The delay accuracy requirement is <1μs (1 / 40 of the wavelength), corresponding to a path error <4mm, while also meeting phase accuracy requirements.

[0086] The principle of receiving delay is similar, but the calculation reference is the nearest array element. The received echo signals are summed after their respective delays Δti' to achieve receiving focus. The combination of transmitting focus and receiving focus forms dual focusing, further improving the signal-to-noise ratio and resolution.

[0087] Fan-shaped scanning is achieved through electronic deflection. When the focal point is not directly in front of the array but deflected to the side, the delay of each array element changes according to a certain pattern. Let the deflection angle be θ, the focal point position be (xf = zf·tanθ, zf), and Ri and Δti be recalculated, with the sound beam pointing in the θ direction. The scanning angle is ±30°, incrementing by 1°, for a total of 61 angles, synthesizing a fan-shaped image. This technique is particularly suitable for detecting inclined cracks and side leaks.

[0088] Table 8 is the calculation parameter table.

[0089]

[0090] 4.6.4 SAFT Synthetic Aperture Focusing Imaging Technology SAFT technology draws inspiration from radar imaging principles, using software algorithms to achieve focusing during the data processing stage, thus overcoming hardware aperture limitations. The basic idea is to move the probe to collect data at multiple locations, equivalent to receiving data simultaneously from a large-aperture array. This data is then coherently superimposed to synthesize the information from each location, reconstructing a high-resolution image.

[0091] Assume the probe is at N positions (xi, yi), i = 1, 2, ..., N, sequentially acquiring echoes A(xi, yi, t). For any point P(xp, yp, zp) in the imaging space, calculate the acoustic path from each probe position to point P: Ri = √[(xp - xi)] 2 +(yp-yi) 2 +zp²]. Two-way time ti = 2Ri / v. From the echo signal at position i, extract the amplitude ai = A(xi, yi, ti) corresponding to time ti. Coherently superimpose the amplitudes at N positions: I(P) = Σ[ai·wi], where wi is the weighting function, usually cosine weighted wi = cos(θi), θi is the incident angle, to reduce edge effects.

[0092] Repeat the above calculation for all points in the imaging space to obtain the three-dimensional image I(x, y, z). The computational load is enormous: N acquisition positions × M imaging points × T time sampling points, but it can be accelerated in parallel by GPU for real-time processing.

[0093] SAFT's advantage lies in the fact that its lateral resolution is no longer limited by the beam width of a single measurement, but is determined by the synthetic aperture. The moving distance L is equivalent to the aperture, and the resolution δ≈λ·z / L. Moving 1 meter to a depth of 300mm, the resolution can reach 12mm, nearly 8 times higher than a single measurement (94mm). The disadvantage is that it requires precise knowledge of the probe position, and the positioning accuracy of the moving platform (<10mm) directly affects the image quality.

[0094] 4.7 Step 4: Fine Measurement of Microwave Humidity Field (Detailed Parameters) 4.7.1 Physical Principles of Microwave Humidity Measurement Microwave humidity sensors utilize the polarity of water molecules. Water molecules are strongly polar molecules, undergoing rotational polarization under the influence of a microwave electric field, absorbing electromagnetic energy. 2.45 GHz is close to the relaxation frequency and absorption peak of water, making it extremely sensitive to changes in moisture content. When microwaves penetrate concrete, the power attenuation is P / P0 = exp(-α·d), and the attenuation coefficient α is proportional to the moisture content. Simultaneously, the relative permittivity εr is strongly correlated with the moisture content, affecting the microwave propagation speed and phase delay.

[0095] The sensor employs either transmission or reflection measurement. Transmission sensors place transmitting and receiving antennas on both sides of the base plate to measure transmitted power attenuation and phase delay; reflection sensors are arranged on one side to measure the reflection coefficient. Existing buildings typically only allow contact from one side, so reflection sensors are used. The reflection coefficient Γ = (Z - Z0) / (Z + Z0), where Z0 is the air impedance (377Ω), Z is the surface impedance of the water-containing concrete, Z = √(μ0 / ε0εr), and εr varies with the moisture content, causing Γ to change.

[0096] The penetration depth of 2.45 GHz in dry concrete (εr≈5) is δ=1 / (2πf√(με)·tanδ)≈150-200mm, where tanδ is the loss tangent (≈0.01). As the moisture content increases, εr increases, tanδ increases, and the penetration depth decreases to 100-150mm. Therefore, microwave sensors mainly sense humidity from the surface to the middle layer (0-200mm) and are not sensitive to deeper layers.

[0097] Array-based measurement uses multiple spatially distributed sensors to acquire a two-dimensional humidity distribution map. A standard 5×5 array, with a sensor spacing of 0.5m, covers a 2m×2m area. Each sensor measures a circular area approximately 0.3m in diameter below it, with about 40% overlap between the measurement areas of adjacent sensors. A continuous humidity field can be reconstructed through interpolation.

[0098] The measurement process is significantly affected by temperature. The temperature coefficient of water's dielectric constant is approximately -0.4% / ℃, meaning that for every 1℃ increase in temperature, εr decreases by about 0.3. To eliminate temperature interference, the sensor incorporates a temperature compensation chip that measures the surface temperature in real time and corrects the moisture content reading. The compensation formula is: θcorrected = θmeasured + k·(T - Tref), where k ≈ 0.1% / ℃ is the compensation coefficient, and Tref = 20℃ is the reference temperature.

[0099] 4.7.2 Microwave Sensor Array Configuration Table 9 is a microwave sensor array configuration table.

[0100]

[0101] 4.8 Step 5: Structural Vibration Response Analysis (Detailed Parameters) 4.8.1 Physical Mechanism of Vibration Detection Structural vibrations are highly sensitive to water content and defects, stemming from the following physical mechanisms: Damping effect: The damping ratio ζ of water-containing concrete increases significantly. Dry concrete has a ζ of approximately 0.03-0.05, while water-containing concrete can reach 0.08-0.12, an increase of 1-2 times. Damping originates from the viscous dissipation of water and the friction generated by the compression of pore water. Vibration decay accelerates, with the amplitude decaying exponentially with distance, α = 2πfζ / v, where f is the frequency and v is the wave velocity. The higher the frequency, the faster the decay; therefore, high-frequency vibrations are more sensitive to water content.

[0102] Wave velocity variation: Moisture content reduces the concrete's stiffness and density ratio E / ρ, leading to a decrease in wave velocity v = √(E / ρ). The elastic modulus E decreases by approximately 2-3% with every 1% increase in moisture content, while density ρ increases by approximately 1%. The combined effect is a wave velocity decrease of approximately 1.5-2% per 1% of moisture content. At 10% moisture content, the wave velocity decreases from 4000 m / s to approximately 3500 m / s.

[0103] Frequency shift: Natural frequency fn = √(K / M) / (2π), where K is stiffness and M is mass. Water content leads to a decrease in stiffness K and an increase in mass M, both of which reduce fn. Experimental measurements show that for every 1% increase in water content, the lower-order natural frequencies decrease by approximately 0.5-1%.

[0104] Discontinuous vibration transmission: Cracks and voids block the propagation of vibration waves, and the transfer function H(ω) = Xout / Xin drops sharply at the defect. In the intact region, |H|≈0.8-1.0, while cracks cause |H|<0.5, and the phase difference Δφ>30°.

[0105] Passive environmental vibration testing utilizes vibrations generated during normal building use as excitation sources. Human footsteps produce low-frequency vibrations of 1-10Hz, vehicle loads of 2-20Hz, mechanical equipment of 10-200Hz, and wind loads of 0.5-5Hz. These broadband random excitations contain rich frequency components, capable of exciting multiple modes of the structure. The advantages are that no artificial excitation is required and continuous monitoring is possible; the disadvantages are that the amplitude is uncontrollable, the frequency is not selectable, and data processing requires long-term accumulation.

[0106] 4.8.2 Vibration Test Configuration Parameters Table 10 shows the vibration test configuration parameters.

[0107]

[0108] 4.9 Summary of Technical Implementation for Steps 7 to 9 4.9.1 Key Points of Three-Dimensional Reconstruction Technology for Leakage Channels Voxelization discretizes a continuous space into a regular mesh, assigning an ISI value to each voxel. The choice of voxel size balances accuracy and computational cost: 25mm.3 High precision but high voxel count (×8 times), 100mm 3 Fast calculation, but loss of detail. Standard 50mm. 3 balance.

[0109] Table 11 is a table of key technical points for three-dimensional reconstruction of leakage channels.

[0110]

[0111] Starting from candidate leak points, the search proceeds along the ascending direction of the ISI gradient. The A* algorithm introduces a heuristic function h(n) = Euclidean distance to the suspected leak source, accelerating the search.

[0112] The evaluation function is f(n) = g(n) + h(n), where g(n) is the cumulative cost from the starting point to the current node. The cost is defined as the path length plus ISI weighting: cost = Σ[Δd·(1+ISI / 100)]. The higher the ISI, the more likely it is to be a leakage channel, and the greater the weight, guiding the path to grow along the high ISI region.

[0113] The gradient pursuit method calculates the ISI gradient for each voxel. ISI = ( ISI / x, ISI / y, ISI / z). Starting from the low ISI point at the boundary (ISI<30), iteratively move along the gradient direction: r(k+1) = r(k) + α· ISI / | ISI|, step size α = voxel size. When a local maximum is reached (| Stop at ISI|90), forming a tracing path. The convergence point of multiple paths is the inferred core of the leakage source.

[0114] Post-processing of the path includes smoothing filtering, bifurcation detection, and trunk extraction. Smoothing uses cubic B-spline fitting to eliminate jagged trajectories. Bifurcation detection identifies nodes with multiple incident paths for a single voxel and marks them as branch points. Trunk extraction retains the path with the largest ISI integral as the main leakage channel, and branches as secondary channels.

[0115] The voxelized 3D model is stored in PLY or STL format. Color mapping: Blue (ISI=0) → Green (ISI=25) → Yellow (ISI=50) → Orange (ISI=75) → Red (ISI=100), using linear interpolation of RGB values. Transparency is set according to confidence level: High confidence (>80%) completely opaque alpha=1.0, medium confidence (50-80%) semi-transparent alpha=0.6, low confidence (<50%) highly transparent alpha=0.3.

[0116] BIM integration imports inspection results into BIM platforms such as Revit or ArchiCAD. Utilizing the IFC (Industry Foundation Classes) standard data interface, leak points, paths, and isosurfaces are added as BIM components to the building model. Each leak object includes attributes: type (point / line / surface), location coordinates, ISI value, leak volume, inspection date, and confidence level. Within the BIM environment, users can query leak information for any component, and clash detection (leak path and pipeline intersection analysis), schedule simulation (repair and construction sequencing), and cost estimation (material quantity statistics) are supported.

[0117] Augmented Reality (AR) displays a virtual 3D leak model overlaid on a real-world scene using mobile devices or AR glasses. How it works: The device's camera captures a live video stream, and a SLAM (Simultaneous Localization and Mapping) algorithm identifies spatial feature points to determine the device's position and orientation within the room. Based on these positional relationships, the virtual model's projection onto the screen is calculated and rendered in real-time. Users can "see through" the leak channels inside the floor slab using their mobile devices, aiding in repair and location. AR annotations display ISI (Inspection and Retention Index) values, depth information, and recommended measures. Interactive features allow users to click on virtual objects to access detailed inspection data.

[0118] 4.9.2 Implementation of Intelligent Diagnostic System Causal diagnosis employs Bayesian network reasoning. Nodes include: observed evidence (ISI, location, depth, moisture content, etc.), intermediate variables (existence of cracks, condition of waterproofing layer, backfill quality, etc.), and root causes (construction defects, material aging, external force damage, improper design, etc.). Edges represent causal relationships, and the conditional probability table (CPT) is learned from historical cases.

[0119] Bayesian inference formula: P(cause|evidence) = P(evidence|cause) · P(cause) / P(evidence). Input observed evidence and network propagation probability; output the posterior probability of each cause. For example: Observed ISI = 85, depth 400mm, location near an exterior wall, water content 18%. Inference results: exterior wall penetration probability 65%, pipe rupture 25%, structural crack 10%. Select the cause with the highest probability as the principal cause, and those with a probability > 20% as possible causes.

[0120] Decision trees aid in refined diagnosis. The root node represents the leak location (internal / edge / corner), the second layer represents the depth (surface / middle / deep), and the third layer represents feature combinations (high / low temperature, high / low humidity, sound velocity changes, etc.). Leaf nodes represent the specific cause and confidence level. Training data consists of 200 expert-annotated cases, and the tree is constructed using the CART algorithm, with a maximum depth of 6 layers and a minimum of 5 leaf node samples. Figure 3 It is a leakage diagnosis decision tree diagram - feature - cause mapping.

[0121] Repair solutions are generated based on a rule engine and case-based reasoning. The rule base contains 100+ IF-THEN rules, such as: "IF Leakage Level = IV AND Depth > 300mm AND Cause = Structural Crack THEN Recommended Solution = Pressure Grouting + Flexible Waterproofing Layer". The case base stores historical repair projects, including problem descriptions, solution selections, implementation results, costs, and timelines. For new problems, the 3-5 most similar cases are retrieved, and a solution is recommended based on a similarity-weighted approach.

[0122] This invention is not limited to the above-described optional embodiments. Anyone can derive other various forms of products under the guidance of this invention. However, regardless of any changes made in their shape or structure, any technical solution that falls within the scope of the claims of this invention shall be protected by this invention.

Claims

1. A smart detection system for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array, characterized in that: It includes a cloud-based analysis server and a field mobile detection platform; the cloud server performs deep learning inference, big data analysis, and digital twin modeling, and achieves data interaction and collaborative work with the field mobile detection platform through 5G or WiFi wireless communication; the field mobile detection platform, as the operation execution unit, integrates an intelligent mobile vehicle, an active excitation subsystem, a multimodal sensor array subsystem, and an edge computing unit.

2. The intelligent detection system for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array as described in claim 1, characterized in that: The multimodal sensor array subsystem includes an infrared thermal imager, a ground-penetrating radar, an ultrasonic phased array, a microwave sensor, and a vibration sensor. The infrared thermal imager is used for surface detection, the ground-penetrating radar achieves deep penetration, the ultrasonic phased array provides high-resolution focused imaging, the microwave sensor measures water content, and the vibration sensor acquires the dynamic response of the structure.

3. The intelligent detection system for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array as described in claim 1, characterized in that: The active excitation subsystem includes a heating device, an ultrasonic excitation source, and a water pressure simulation device. The active excitation subsystem improves the detection sensitivity of microleakage by applying energy perturbation.

4. The intelligent detection system for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array as described in claim 1, characterized in that: The edge computing unit is responsible for real-time processing and preliminary diagnosis of field data, converting raw sensor signals into structured feature data, enabling rapid response at the edge and reasonable allocation of cloud computing load.

5. A method for intelligent detection of leakage in thick foundation slabs of existing buildings based on a multimodal sensor array, using the intelligent detection system for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array as described in any one of claims 1 to 4, characterized in that: Includes the following steps: S1: Infrared detection: Passive multi-temporal detection: 5-6 AM, 2-3 PM, 6-7 PM; Active thermal excitation: heating for 15-30 min, cooling for 30-60 min; Feature extraction: temperature gradient. T, time derivative T / t, thermal time constant τ; S2: GPR detection: multi-frequency scanning: 800MHz coarse scan, 1.5GHz fine scan, 2.5GHz detailed scan; three-dimensional imaging: B-scan profile, C-scan plane, 3D reconstruction; water content inversion: dielectric constant εr, volumetric water content θ; S3: Ultrasonic testing: Phased array focusing: depth 100-600mm, electronic scanning; Echo analysis: time difference Δt, amplitude A, spectrum f; SAFT imaging: improved resolution; S4: Microwave humidity measurement: array measurement: 5×5 sensor, penetration 100-200mm; gradient analysis: spatial gradient θ, high-value region identification; S5: Vibration Analysis: Passive Environmental Vibration: Multi-point synchronous acquisition, 1-500Hz; Active Ultrasonic Excitation: 40-100kHz frequency sweep; Anomaly Identification: Damping α variation, transfer function H(ω). S6: Multimodal data fusion: Spatial registration: accuracy <10mm; Feature extraction: 41-dimensional vector, infrared 12-dimensional, GPR 8-dimensional, ultrasound 10-dimensional, microwave 5-dimensional, vibration 6-dimensional; Deep learning fusion: multi-head attention, Transformer encoder, multi-task output; Physical constraint correction: thermal diffusion equation, seepage mechanics, wave equation. S7: 3D Reconstruction: Point Cloud Generation: Voxel 50mm 3 Color mapping; Path tracing: A* algorithm, gradient tracing, BIM integration, AR display; S8: Quantitative Assessment: Comprehensive Index ISI: =Σwi•Ii; Leakage Classification: Class I: 0-25, Class II: 25-50, Class III: 50-75, Class IV: 75-100; Leakage Estimation: Q=k•i•A, accuracy ±15%; S9: Intelligent Diagnosis: Cause Diagnosis: Bayesian Network, Decision Tree; Repair Solutions: Technology Selection, Cost Estimation, and Schedule Planning.

6. The intelligent detection method for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array according to claim 5, characterized in that: In step S1: Passive detection parameter settings: Acquisition time period: 5-6 AM, 2-3 PM, 6-7 PM; Acquisition distance: 1-2 meters; Number of frames per point ≥ 10; Temperature difference requirement > 5℃; Movement step: 1 meter / step; Active excitation parameter configuration: target temperature rise: +10-15℃; heating time: 15-30 minutes; acquisition interval: 30 seconds / time; cooling observation time: 30-60 minutes; acquisition interval: 60 seconds / time; heating power: 500-2000W.

7. The intelligent detection method for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array according to claim 5, characterized in that: In step S2, a multi-frequency adaptive scanning strategy is adopted: Coarse scan: frequency 800MHz; detection depth >500mm; resolution ~150mm; grid spacing 0.5m; scan area / hour: 200-300m² 2 Application objectives: Rapid comprehensive survey and identification of large-scale anomalies; Fine scanning: frequency 1.5GHz; detection depth 300-400mm; resolution ~80mm; grid spacing 0.2m; scanning area / hour: 100-150m² 2 Application objective: Mid-level of abnormal areas, precise location of defects; Fine scanning: frequency 2.5GHz; detection depth <300mm; resolution ~50mm; grid spacing 0.1m; scanning area / hour: 50-80m² 2 ; Application target: Surface imaging of key areas, high resolution imaging.

8. The intelligent detection method for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array according to claim 5, characterized in that: In step S3, the phased array focusing parameters are configured: 64-element configuration: element spacing 2mm; center frequency 100kHz; pulse width 2-5 cycles; depth of focus 100-600mm; focal length step 50mm; angle scan ±30°; sampling rate ≥10MHz; 128-element configuration: element spacing 1mm; center frequency 150kHz; pulse width 2-3 cycles; focusing depth 100-600mm; focal length step 50mm; angle scanning ±45°; sampling rate ≥20MHz.

9. The intelligent detection method for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array according to claim 5, characterized in that: In step S4, the microwave sensor array is configured as follows: Standard configuration: Array size 5×5; sensor spacing 0.5m; measurement time / point 2 seconds; total measurement time 50 seconds; coverage area 4m² 2 ; Spatial resolution: 0.5m; Applicable scenarios: routine testing; High-precision configuration: 7×7 array size; sensor spacing 0.3m; measurement time / point 3 seconds; total measurement time 147 seconds; coverage area 3.6m². 2 Spatial resolution 0.3m; Applicable scenarios: critical areas; Fast scanning configuration: 3×3 array size; sensor spacing 0.8m; measurement time / point 1 second; total measurement time 9 seconds; coverage area 3.2m². 2 Spatial resolution 0.8m; Applicable scenario: preliminary screening.

10. The intelligent detection method for leakage in thick foundation slabs of existing buildings based on a multimodal sensor array according to claim 5, characterized in that: In step S5, the vibration test configuration parameters are as follows: Passive environmental vibration: Excitation source: environmental vibration; Sensor type: MEMS triaxial accelerometer; Arrangement: grid pattern, spacing 0.5-1m; Sampling rate 2kHz; Acquisition time 5-10 minutes; Sensitivity >1000mV / g; Dynamic range >60dB; Measurement frequency band 1-500Hz; Application objective: low-frequency overall response, damping identification; Active ultrasonic excitation: Excitation source: ultrasonic excitation source, 40-100kHz frequency sweep; Sensor type: MEMS triaxial accelerometer; Arrangement: grid pattern, spacing 0.5-1m; Sampling rate: 500kHz; Acquisition duration: 0.5 seconds per frequency point; Sensitivity >1000mV / g; Dynamic range >60dB; Measurement frequency band: 40-100kHz; Application target: high-frequency local defects, modal analysis.

Images