Building outer wall facing layer hollowing defect monitoring device and monitoring method thereof

Abstract

The invention provides a building outer wall facing layer hollowing defect monitoring device and a monitoring method thereof, and belongs to the technical field of building outer wall facade facing layer defect monitoring. The monitoring method comprises a temperature data acquisition process, a heat flow intelligent analysis process, a hollowing defect adaptive identification process, a temperature field dynamic optimization process, an environmental interference self-learning correction process, a multi-source data fusion diagnosis process, a defect trend prediction process, a visual interaction process and a resource intelligent distribution process. All the processes of the monitoring method are executed through a temperature collecting and processing end, and aiming at the problems of insufficient precision, large environmental interference, low intelligence, prediction deficiency and low resource efficiency in building outer wall facing layer hollowing defect detection, through a temperature sensor matrix, an intelligent data processing process and a multi-source data fusion design, the building outer wall facing layer hollowing defect monitoring method is provided. And the detection precision, the automation degree and the maintenance efficiency are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of defect monitoring of building exterior wall facade finishing layers, and specifically to a device and method for monitoring hollowing defects in building exterior wall finishing layers. Background Art

[0002] The quality of building exterior wall finishes (such as cement mortar, tiles, and paint) directly affects the building's safety, durability, and aesthetics, as they are an important component of the building's appearance and protective structure. According to industry research, hollowing defects in exterior wall finishes are a common problem during a building's service life, and users prioritize hollowing detection in the order of high-precision positioning, rapid diagnosis, non-destructive testing, environmental adaptability, and risk prediction. However, actual testing methods are primarily based on tapping, infrared thermal imaging, and ultrasonic testing, with significant differences in performance and requirements. Although hollowing detection is one of the core demands of building maintenance, it is limited by the insufficient accuracy of testing equipment, significant environmental interference, complex operation, and a lack of predictive capabilities. Current technology struggles to meet users' expectations for efficient and intelligent hollowing management.

[0003] Modern buildings (such as high-rise residential buildings, commercial complexes, and historically protected buildings) place higher demands on exterior wall quality management, especially the need for dynamic monitoring and preventive maintenance during their service life. However, existing hollowing defect detection technologies face the following key technical challenges: Inadequate detection accuracy and non-destructiveness: Traditional percussion methods rely on manual experience, judging hollows by the sound of the tapping sound. This can result in positional errors of up to ±50mm and area errors of ±10-15%. Furthermore, tapping can damage the finish, making it difficult to meet non-destructive requirements. Infrared thermal imaging methods (such as the Fluke Ti400), while non-contact, have a spatial resolution of only 15mm and a temperature accuracy of ±2°C. This can easily miss small hollow areas (<0.1m²), making it difficult to meet high-precision requirements.

[0004] Limited environmental adaptability: Existing detection methods are significantly affected by environmental factors (such as high temperatures, rainfall, and strong winds). For example, infrared thermal imaging can have a temperature error of up to ±1.8°C in high humidity or strong sunlight, resulting in a false alarm rate as high as 12%. Tapping and ultrasonic detection methods lack environmental correction mechanisms and are difficult to adapt to cross-seasonal (winter and summer) or extreme weather conditions (such as heavy rain and strong winds), limiting their reliability and applicability.

[0005] Low levels of intelligence and automation: Current inspection equipment is mostly single-function modules (such as infrared thermal imagers and ultrasonic probes), lacking a unified control system architecture. Data is isolated and processing relies on manual analysis. For example, infrared thermal imaging requires professional interpretation of the thermal image, and the tapping method requires point-by-point inspection, resulting in lengthy inspection times (60-150 minutes per square meter) and the inability to achieve automation. Existing systems lack real-time data fusion and intelligent diagnostic capabilities, requiring frequent manual operation and inefficient maintenance.

[0006] Lack of predictive capabilities: Existing technologies (such as percussion, infrared thermal imaging, and ultrasonic testing) can only detect the current hollowing state and cannot predict the risk of hollowing expansion over the next 96 hours or longer. This lack of trend prediction leads to delayed maintenance, increased repair costs, and safety hazards. For example, hollowing expansion can cause the finish layer to fall off, threatening building safety.

[0007] Inefficient resource utilization: Existing testing equipment consumes high power (for example, infrared thermal imagers consume 88% of their rated power) and lacks resource optimization mechanisms. The testing process doesn't dynamically adjust based on environmental data (such as humidity and wind speed), resulting in energy waste. Ultrasonic testing equipment is expensive (approximately 200,000 yuan per unit), making it unsuitable for large-scale deployment and limiting its economic viability.

[0008] Therefore, a device and method for monitoring hollow defects in the exterior wall finishing layer of a building are needed to solve the above problems. Summary of the Invention

[0009] Technical issues solved: In view of the deficiencies in the prior art, the present invention provides a device and method for monitoring hollowing defects in a building exterior wall finishing layer, which solve the problems mentioned in the above background technology.

[0010] Technical solution: To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for monitoring hollowing defects in building exterior wall finishing layers, comprising a monitoring method, the monitoring method comprising a temperature data acquisition process, a heat flow intelligent analysis process, a hollowing defect adaptive identification process, a temperature field dynamic optimization process, an environmental interference self-learning correction process, a multi-source data fusion diagnosis process, a defect trend prediction process, a visualization interaction process, and a resource intelligent allocation process, all processes of the monitoring method being executed by a temperature acquisition and processing end; The temperature data acquisition process collects the surface temperature data of the main structure of the building in real time through the temperature sensor in the temperature measuring unit. The temperature measuring units are distributed in a rectangular array with a spacing of 200mm×200mm. The temperature data is transmitted to the temperature acquisition processing end through the wire at a frequency of milliseconds, and the sampling frequency is dynamically adjusted according to the changes in ambient light, temperature and wind speed to optimize the data acquisition efficiency; the heat flow intelligent analysis process receives the temperature data of the temperature data acquisition process, adopts the heat flow distribution analysis method optimized by neural network, combines the thermal conductivity characteristics of the finishing layer and the main structure of the building, generates a heat flux density distribution map with a resolution of 20mm, identifies potential hollowing areas, and transmits the results to the air Drum defect adaptive identification process and multi-source data fusion diagnosis process; the hollowing defect adaptive identification process receives the heat flux density distribution map from the heat flow intelligent analysis process, adopts a deep learning-based cliff temperature change detection method, dynamically identifies temperature anomaly patterns through multi-scale time series analysis, generates preliminary hollowing defect diagnosis results, and optimizes the detection threshold through a self-learning mechanism to improve accuracy; the temperature field dynamic optimization process receives temperature data from the temperature data acquisition process, adopts three-dimensional spatial interpolation and heat flow balance methods to construct a real-time temperature field model of the main building structure surface with a resolution of 50mm, and automatically optimizes the monitoring coverage of the temperature measurement unit to reduce blind spots; The environmental interference self-learning correction process collects ambient temperature, humidity, wind speed, sunshine intensity, and air pressure data through external sensors, adopts an adaptive interference filtering method, dynamically learns the law of environmental changes, corrects the interference factors in the temperature data, and controls the error within ±0.08°C. The correction results are then transmitted to the multi-source data fusion diagnosis process; The multi-source data fusion diagnosis process receives the heat flux density distribution map of the heat flux intelligent analysis process, the preliminary diagnosis results of the hollowing defect adaptive identification process, and the correction data of the environmental interference self-learning correction process. It uses a multi-feature dynamic weighted fusion method to comprehensively consider temperature, heat flux, environmental factors, and time series stability to calculate the confidence score of the hollowing defect and generate a diagnosis report containing the defect location, area, and confidence level. The defect trend prediction process receives the diagnosis report from the multi-source data fusion diagnosis process, uses a trend prediction method based on a generative adversarial network to analyze the temperature and heat flow data of the past 21 days, predicts the risk of hollowing defect expansion in the next 96 hours, and generates a multi-level risk warning; The visualization interaction process receives the diagnostic report of the multi-source data fusion diagnosis process and the prediction results of the defect trend prediction process, and generates an interactive visualization interface including a three-dimensional temperature field heat map, a hollowing defect distribution map, and a risk trend curve. It supports users to adjust display parameters through a touch screen or mobile phone application, and provides remote real-time monitoring, multi-user collaboration, and data backtracking functions. The intelligent resource allocation process receives the real-time power consumption and computing load data of each process, and adopts a dynamic resource scheduling method to optimize the computing resource allocation of each process in the temperature acquisition and processing end based on the real-time load and historical operation data, ensuring that the power consumption does not exceed 80% of the rated power, and supports fault self-diagnosis and automatic recovery functions.

[0011] Preferably, the heat flux distribution analysis method of the heat flux intelligent analysis process is based on a convolutional neural network, combined with the 200mm×200mm grid distribution of the temperature measurement unit and the material properties of the finishing layer such as cement mortar, tiles, paint, and stone, to generate a heat flux density distribution map with a resolution of 20mm in real time; The heat flux distribution analysis method is trained using 45 days of historical temperature data and a material database to automatically identify areas of heat flux anomaly and generate a feature vector containing the heat flux anomaly intensity, location, timestamp, material type, and defect depth estimation. It supports online updating of the analysis model to adapt to different building exterior wall materials and construction processes.

[0012] Preferably, the cliff-like temperature change detection method of the hollowing defect adaptive identification process adopts multi-scale time series analysis, sets five analysis windows of 3 days, 7 days, 14 days, 21 days and 30 days, and dynamically detects abnormal events with a temperature drop rate greater than 0.25°C / hour or a temperature fluctuation standard deviation greater than 1.2°C; the cliff-like temperature change detection method is combined with a reinforcement learning mechanism, and dynamically optimizes the detection threshold based on historical false alarm data, cross-regional temperature comparison and material properties to generate preliminary diagnostic results including hollowing position coordinates, temperature change amplitude, confidence score, defect severity and potential expansion direction.

[0013] Preferably, the three-dimensional spatial interpolation and heat flow balance method of the temperature field dynamic optimization process is based on the grid distribution of the temperature measuring unit 2 and adopts Gaussian process regression technology to construct a dynamic temperature field model of the surface of the main structure 8 of the building with a resolution of 50mm; the heat flow balance method supports adaptive grid adjustment. When a local temperature anomaly is detected, the grid resolution is automatically encrypted to 15mm, and the heat flow distribution prediction is optimized in combination with the building geometry, wind direction, and wall material to reduce monitoring blind spots, and the results are transmitted to the multi-source data fusion diagnosis process.

[0014] Preferably, the environmental interference self-learning correction process collects ambient temperature, humidity, wind speed, sunshine intensity, air pressure, and rainfall data through external sensors, adopts an adaptive interference filtering method, and dynamically updates the weights of interference factors based on Bayesian reasoning. The initial weights are 0.45 for ambient temperature, 0.25 for humidity, 0.15 for sunshine intensity, 0.05 for wind speed, 0.03 for air pressure, and 0.02 for rainfall; the adaptive interference filtering method optimizes the weight distribution through a 21-day learning cycle, supports environmental adaptation across seasons, day and night, and extreme weather such as heavy rain and strong winds, and the temperature data error after correction is controlled within ±0.08°C, and the correction results are transmitted to the multi-source data fusion diagnosis process.

[0015] Preferably, the multi-feature dynamic weighted fusion method of the multi-source data fusion diagnosis process constructs a feature vector including temperature change rate, heat flux change rate, environmental correction factor, time series stability, and material thermal conductivity, and adopts a random forest classification model to dynamically adjust the feature weights. The initial weights are 0.35 for temperature change rate, 0.3 for heat flux change rate, 0.2 for environmental correction factor, 0.1 for time series stability, and 0.05 for material thermal conductivity; the multi-feature dynamic weighted fusion method generates a confidence score for hollowing defects. When the score is greater than 0.9, a diagnosis report is generated including the hollowing location, area error ±1.5%, material type inference, repair priority, and construction risk assessment, and the report is transmitted to the visual interactive process through an encrypted protocol.

[0016] Preferably, the generative adversarial network trend prediction method of the defect trend prediction process receives temperature, heat flux, confidence score, and environmental data of the past 21 days to construct a hollowing defect expansion risk model; The proposed generative adversarial network trend prediction method optimizes prediction accuracy through adversarial training and generates a risk probability curve for the next 96 hours. When the risk probability is greater than 85%, it triggers multi-level warnings (low, medium, high, and emergency). It also generates a warning report containing a repair time window, material selection recommendations, and construction safety tips, and transmits it to the visual interactive process.

[0017] Preferably, the interactive visualization interface of the visualization interaction process includes the following six parts: 3D temperature field heat map dynamically displays the temperature distribution on the surface of the main building structure. Hollow areas are highlighted in red, and edges use a gradient transition to identify potential defects. It supports 360° rotation and 3D cross-section view switching. Hollow defect distribution map, with hollow location, area, confidence level, defect depth estimation, and material type marked, supporting dynamic zoom, area screening, and defect evolution animation; The risk trend curve shows the change in hollowing risk probability in the next 96 hours, and supports user-defined prediction time range from 24 hours to 30 days; The user interaction module supports adjusting the heat map perspective, setting risk warning thresholds, and exporting diagnostic reports through the touch screen or mobile application. The interface supports voice control, multi-language switching between Chinese, English, Japanese, and German, and gesture operation; The historical data backtracking module displays the temperature, heat flow, and defect change trends over the past 60 days, and supports abnormal event annotation, playback, and comparative analysis; The intelligent suggestion module generates suggestions including repair process, material recommendation, and construction time optimization based on the diagnosis report and prediction results, and automatically matches the local supplier database; Interface data is transmitted to user terminals in real time via 5G networks and Wi-Fi, supporting multi-user simultaneous access, hierarchical permission management, and offline data caching.

[0018] Preferably, the dynamic resource scheduling method of the intelligent resource allocation process receives the real-time power consumption, computing delay, data throughput and memory usage data of each process, and uses deep reinforcement learning to optimize resource allocation; the dynamic resource scheduling method predicts the resource demand for the next 3 hours based on the operating data of the past 21 days, and dynamically adjusts the allocation ratio of each process, with the temperature data acquisition process allocated 26%, the heat flow intelligent analysis process allocated 24%, and the hollow defect adaptive identification process allocated 20%; when a hardware fault is detected, the dynamic resource scheduling method automatically isolates the fault module, reallocates resources, generates a diagnostic report including the fault type, impact range, recovery suggestions, estimated recovery time and backup plans, and transmits it to the visual interactive process.

[0019] A device for monitoring hollowing defects in a building exterior wall finishing layer, comprising a temperature sensor matrix, a temperature measurement unit, and a building main structure; The main structure of the building is the outer surface of the main structure of the building. The rectangular array of temperature measuring units on the surface of the main structure of the building is provided. The temperature measuring units are distributed at a spacing of 200mm×200mm. The temperature measuring units include aluminum foil insulation cotton and a temperature sensor. The temperature sensor is fixed to the aluminum foil insulation cotton by gluing. The surface of the aluminum foil insulation cotton is provided with a fixing hole. The aluminum foil insulation cotton is fixed to the surface of the main structure of the building after an ST38 air nail with a length of 15mm is driven through the fixing hole by an air nail gun; wires are connected between the temperature measuring units, and the temperature measuring units and wires distributed in a matrix constitute a complete temperature sensor matrix; a temperature acquisition and processing end is clamped on the side of the surface of the main structure of the building close to the temperature sensor matrix, and the temperature acquisition and processing end A monitor is installed at the bottom, and the monitor includes a display screen, which is used to display the temperature data and diagnostic results collected by the temperature sensor matrix; the surface of the main structure of the building is covered with a finishing layer, and the finishing layer is constructed on the temperature sensor matrix through traditional construction technology; the temperature data collected by the temperature sensor matrix is ​​transmitted to the temperature collection and processing end through a wire and displayed by the monitor. When the temperature monitored by the temperature measuring unit in the temperature sensor matrix drops sharply, it is determined that a hollowing defect occurs in the finishing layer on the surface of the main structure of the building; the temperature collection and processing end is integrated with a data processing module, and the above-mentioned monitoring method is executed by the data processing module to process the temperature data, generate a diagnostic report and prediction results, and output them through a transmission chip.

[0020] Beneficial effects: The present invention provides a device and method for monitoring hollowing defects in building exterior wall finishing layers. It has the following beneficial effects: 1. This invention addresses the challenges of insufficient precision, significant environmental interference, low intelligence, prediction deficiencies, and low resource efficiency in hollowing defect detection in building exterior wall finishes. By utilizing a temperature sensor matrix, intelligent data processing, and multi-source data fusion, it significantly improves detection accuracy, automation, and maintenance efficiency. Firstly, high-precision non-destructive testing and environmental adaptability are achieved. The temperature data acquisition process utilizes high-precision temperature sensors (accuracy ±0.05°C) distributed in a 200mm×200mm grid to collect real-time surface temperature data from the building's main structure. Combined with a convolutional neural network (CNN) in the intelligent heat flow analysis process, this generates a 20mm resolution heat flux density map, accurately identifying hollowing locations (to an accuracy of ±3-4mm) and areas (to an accuracy of ±1.0-1.2%) without damaging the finish layer by tapping. This method outperforms traditional tapping methods (to an accuracy of ±50mm) and infrared thermal imaging methods (to an accuracy of ±25mm). The environmental interference self-learning correction process uses Bayesian reasoning to dynamically adjust the weights of environmental factors (temperature 0.45, humidity 0.25, etc.), controlling the correction error to ±0.08°C, adapting to cross-seasonal (winter and summer) and extreme weather (such as heavy rain and strong winds), and overcoming the high false alarm rate (12%) of infrared thermal imaging methods.

[0021] 2. The multi-source data fusion diagnostic process in this invention utilizes a random forest model, integrating temperature, heat flow, environmental factors, time series stability, and material thermal conductivity to generate a diagnostic report with a confidence score (>0.9). It automatically outputs the hollowing location, area, and repair priority, reducing detection time to 18-25 minutes per 1m², a 3-8 times improvement in efficiency compared to infrared thermal imaging (70 minutes) and tapping methods (150 minutes). The adaptive hollowing defect identification process uses deep learning to detect abrupt temperature drops (>0.25°C / hour), combined with reinforcement learning to optimize thresholds, reducing manual intervention and achieving full process automation. This meets the dynamic monitoring needs of high-rise residential buildings, commercial complexes, and historical structures.

[0022] 3. The defect trend prediction process in this invention utilizes a generative adversarial network to predict the 96-hour risk of hollow expansion (with an accuracy rate of 86-92%) based on 21 days of historical data. This process generates multi-level warnings (low, medium, high, and urgent) and repair recommendations (such as local grouting and tile re-tiling). This prevents the risk of finish peeling caused by hollow expansion, reduces repair costs by 30%, and compensates for the lack of predictive capabilities of traditional methods (such as percussion and infrared thermal imaging). The interactive visualization process generates 3D temperature field maps, defect distribution maps, and risk trend curves. It supports remote monitoring via a mobile app, voice control (in Chinese, English, Japanese, and German), and multi-user collaboration, significantly improving user convenience and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is the monitoring flow chart of the present invention; Figure 2 It is a diagram of the framework of the present invention; Figure 3 It is the overall structural diagram of the present invention.

[0024] Legend: 1. Temperature sensor matrix; 2. Temperature measurement unit; 3. Wire; 4. Temperature sensor; 5. Aluminum foil insulation cotton; 6. Fixing hole; 7. Monitor; 8. Main building structure; 9. Finishing layer; 10. Temperature acquisition and processing end. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Specific embodiment one: like Figures 1 to 3 As shown, the present invention provides a method for monitoring hollowing defects in a building exterior wall finishing layer, including a monitoring method, which includes a temperature data acquisition process, a heat flow intelligent analysis process, a hollowing defect adaptive identification process, a temperature field dynamic optimization process, an environmental interference self-learning correction process, a multi-source data fusion diagnosis process, a defect trend prediction process, a visual interaction process, and a resource intelligent allocation process. All processes of the monitoring method are executed by the temperature acquisition and processing terminal 10; The temperature data acquisition process uses the temperature sensor 4 in the temperature measurement unit 2 to collect the surface temperature data of the main building structure 8 in real time. The temperature measurement unit 2 is distributed in a rectangular array with a spacing of 200mm×200mm. The temperature data is transmitted to the temperature acquisition processing end 10 through the wire 3 at a millisecond frequency, and the sampling frequency is dynamically adjusted according to the changes in ambient light, temperature and wind speed to optimize the data acquisition efficiency. The heat flow intelligent analysis process receives the temperature data of the temperature data acquisition process, adopts the heat flow distribution analysis method optimized by the neural network, and combines the thermal conductivity characteristics of the finishing layer 9 and the main building structure 8 to generate a heat flux density distribution map with a resolution of 20mm, identify potential hollowing areas, and generate the results. The data is transmitted to the hollowing defect adaptive identification process and the multi-source data fusion diagnosis process. The hollowing defect adaptive identification process receives the heat flux density distribution map from the heat flow intelligent analysis process, adopts a deep learning-based cliff temperature change detection method, dynamically identifies temperature anomaly patterns through multi-scale time series analysis, generates preliminary hollowing defect diagnosis results, and optimizes the detection threshold through a self-learning mechanism to improve accuracy. The temperature field dynamic optimization process receives the temperature data from the temperature data acquisition process, uses three-dimensional spatial interpolation and heat flow balance methods to construct a real-time temperature field model of the surface of the building's main structure 8 with a resolution of 50mm, and automatically optimizes the monitoring coverage of the temperature measurement unit 2 to reduce blind spots. The environmental interference self-learning correction process collects ambient temperature, humidity, wind speed, sunshine intensity, and air pressure data through external sensors. It uses an adaptive interference filtering method to dynamically learn the laws of environmental changes and correct interference factors in the temperature data. The error is controlled within ±0.08°C and the correction results are transmitted to the multi-source data fusion diagnosis process. The multi-source data fusion diagnosis process receives the heat flux density distribution map from the heat flux intelligent analysis process, the preliminary diagnosis results from the hollowing defect adaptive identification process, and the correction data from the environmental interference self-learning correction process. It uses a multi-feature dynamic weighted fusion method to integrate temperature, heat flux, environmental factors, and time series stability to calculate the confidence score of the hollowing defect and generate a diagnosis report that includes the defect location, area, and confidence level. The defect trend prediction process receives the diagnostic report from the multi-source data fusion diagnostic process and uses a trend prediction method based on a generative adversarial network to analyze the temperature and heat flow data of the past 21 days to predict the risk of hollowing defect expansion in the next 96 hours and generate a multi-level risk warning. The interactive visualization process receives the diagnostic reports from the multi-source data fusion diagnosis process and the prediction results from the defect trend prediction process, generating an interactive visualization interface that includes a three-dimensional temperature field heat map, a hollowing defect distribution map, and a risk trend curve. It supports users to adjust display parameters through a touch screen or mobile phone application, and provides remote real-time monitoring, multi-user collaboration, and data backtracking capabilities. The intelligent resource allocation process receives the real-time power consumption and computing load data of each process, and adopts a dynamic resource scheduling method to optimize the computing resource allocation of each process in the temperature acquisition and processing end 10 based on the real-time load and historical operation data, ensuring that the power consumption does not exceed 80% of the rated power, and supports fault self-diagnosis and automatic recovery functions.

[0027] The heat flow distribution analysis method of the heat flow intelligent analysis process is based on a convolutional neural network. It combines the 200mm×200mm grid distribution of the temperature measurement unit 2 and the material characteristics of the finishing layer 9 (such as cement mortar, tiles, paint, and stone) to generate a heat flux density distribution map with a resolution of 20mm in real time. The heat flux distribution analysis method is trained using 45 days of historical temperature data and a material database to automatically identify areas of heat flux anomaly and generate a feature vector containing heat flux anomaly intensity, location, timestamp, material type, and defect depth estimation. It supports online updating of the analysis model to adapt to different building exterior wall materials and construction processes.

[0028] The cliff-like temperature change detection method in the adaptive hollowing defect identification process uses multi-scale time series analysis, setting five analysis windows of 3 days, 7 days, 14 days, 21 days, and 30 days to dynamically detect abnormal events such as a temperature drop rate greater than 0.25°C / hour or a temperature fluctuation standard deviation greater than 1.2°C. The cliff-like temperature change detection method combines a reinforcement learning mechanism to dynamically optimize the detection threshold based on historical false alarm data, cross-regional temperature comparisons, and material properties, generating preliminary diagnostic results that include hollowing location coordinates, temperature change amplitude, confidence score, defect severity, and potential expansion direction.

[0029] The three-dimensional spatial interpolation and heat flow balance method of the temperature field dynamic optimization process uses Gaussian process regression technology based on the grid distribution of the temperature measurement unit 2 to construct a dynamic temperature field model of the surface of the building's main structure 8 with a resolution of 50mm. The heat flow balance method supports adaptive grid adjustment. When local temperature anomalies are detected, the grid resolution is automatically refined to 15mm. The heat flow distribution prediction is optimized in combination with the building geometry, wind direction, and wall material to reduce monitoring blind spots. The results are transmitted to the multi-source data fusion diagnosis process.

[0030] The environmental interference self-learning correction process collects ambient temperature, humidity, wind speed, sunshine intensity, air pressure, and rainfall data through external sensors, and adopts an adaptive interference filtering method to dynamically update the weights of interference factors based on Bayesian reasoning. The initial weights are 0.45 for ambient temperature, 0.25 for humidity, 0.15 for sunshine intensity, 0.05 for wind speed, 0.03 for air pressure, and 0.02 for rainfall. The adaptive interference filtering method optimizes the weight distribution through a 21-day learning cycle to support environmental adaptation across seasons, day and night, and extreme weather such as heavy rain and strong winds. The temperature data error after correction is controlled within ±0.08°C, and the correction results are transmitted to the multi-source data fusion diagnosis process.

[0031] The multi-feature dynamic weighted fusion method of the multi-source data fusion diagnosis process constructs a feature vector including temperature change rate, heat flux change rate, environmental correction factor, time series stability, and material thermal conductivity. The random forest classification model is used to dynamically adjust the feature weights. The initial weights are 0.35 for temperature change rate, 0.3 for heat flux change rate, 0.2 for environmental correction factor, 0.1 for time series stability, and 0.05 for material thermal conductivity. The multi-feature dynamic weighted fusion method generates a confidence score for hollowing defects. When the score is greater than 0.9, a diagnostic report is generated that includes the hollowing location, area error of ±1.5%, material type inference, repair priority, and construction risk assessment. The report is transmitted to the visual interactive process via an encrypted protocol.

[0032] The defect trend prediction process uses a generative adversarial network trend prediction method, which receives temperature, heat flux, confidence score, and environmental data from the past 21 days to build a hollowing defect expansion risk model. The generative adversarial network trend prediction method optimizes prediction accuracy through adversarial training and generates a risk probability curve for the next 96 hours. When the risk probability is greater than 85%, it triggers multi-level warnings of low, medium, high, and emergency. It also generates an early warning report containing a repair time window, material selection recommendations, and construction safety tips, and transmits it to the visual interactive process.

[0033] The interactive visualization interface of the visualization interaction process consists of the following six parts: 3D temperature field heat map dynamically displays the temperature distribution on the surface of the main building structure. Hollow areas are highlighted in red, and edges use a gradient transition to identify potential defects. It supports 360° rotation and 3D cross-section view switching. Hollow defect distribution map, with hollow location, area, confidence level, defect depth estimation, and material type marked, supporting dynamic zoom, area screening, and defect evolution animation; The risk trend curve shows the change in hollowing risk probability in the next 96 hours, and supports user-defined prediction time range from 24 hours to 30 days; The user interaction module supports adjusting the heat map perspective, setting risk warning thresholds, and exporting diagnostic reports through the touch screen or mobile application. The interface supports voice control, multi-language switching between Chinese, English, Japanese, and German, and gesture operation; The historical data backtracking module displays the temperature, heat flow, and defect change trends over the past 60 days, and supports abnormal event annotation, playback, and comparative analysis; The intelligent suggestion module generates suggestions including repair process, material recommendation, and construction time optimization based on the diagnosis report and prediction results, and automatically matches the local supplier database; Interface data is transmitted to user terminals in real time via 5G networks and Wi-Fi, supporting multi-user simultaneous access, hierarchical permission management, and offline data caching.

[0034] The dynamic resource scheduling method for the intelligent resource allocation process receives real-time power consumption, computing latency, data throughput, and memory usage data for each process, and uses deep reinforcement learning to optimize resource allocation. Based on the operating data of the past 21 days, the dynamic resource scheduling method predicts resource requirements for the next three hours and dynamically adjusts the allocation ratio of each process, allocating 26% to the temperature data collection process, 24% to the intelligent heat flow analysis process, and 20% to the adaptive hollowing defect identification process. When a hardware fault is detected, the dynamic resource scheduling method automatically isolates the faulty module, reallocates resources, and generates a diagnostic report that includes the fault type, impact range, recovery suggestions, estimated recovery time, and backup plans, and transmits it to the visual interactive process.

[0035] A device for monitoring hollowing defects in a building exterior wall finishing layer, comprising a temperature sensor matrix 1, a temperature measuring unit 2 and a building main structure 8; The main structure 8 of the building is the outer surface of the main structure of the building. The rectangular array of temperature measuring units 2 is arranged on the surface of the main structure 8. The temperature measuring units 2 are distributed at a spacing of 200mm×200mm. The temperature measuring units 2 include aluminum foil insulation cotton 5 and temperature sensors 4. The temperature sensors 4 are fixed to the aluminum foil insulation cotton 5 by gluing. The surface of the aluminum foil insulation cotton 5 is provided with fixing holes 6. The aluminum foil insulation cotton 5 is fixed to the surface of the main structure 8 after passing through the fixing holes 6 with an ST38 15mm long air nail shot by an air nail gun. Wires 3 are connected between the temperature measuring units 2. The matrix-distributed temperature measuring units 2 and wires 3 constitute a complete temperature sensor matrix 1. A temperature acquisition and processing terminal 10 is clamped on one side of the surface of the main structure 8 close to the temperature sensor matrix 1. The bottom of the temperature acquisition and processing terminal 10 is fixed to the main structure 8. A monitor 7 is installed on the inside, and the monitor 7 includes a display screen, which is used to display the temperature data and diagnostic results collected by the temperature sensor matrix 1; the surface of the building main structure 8 is covered with a finishing layer 9, and the finishing layer 9 is constructed on the temperature sensor matrix 1 through traditional construction technology; the temperature data collected by the temperature sensor matrix 1 is transmitted to the temperature collection and processing end 10 through the wire 3, and is displayed by the monitor 7. When the temperature monitored by the temperature measuring unit 2 in the temperature sensor matrix 1 drops sharply, it is determined that a hollowing defect occurs in the finishing layer 9 on the surface of the building main structure 8; the temperature collection and processing end 10 is integrated with a data processing module, and all processes in the monitoring method are executed through the data processing module, temperature data is processed, a diagnostic report and prediction results are generated, and output through a transmission chip. Specific embodiment two: like Figures 1 to 3 As shown, the method for monitoring the hollowing defect of the exterior wall finishing layer 9 is as follows: During the building's service life, when sunlight shines on the outer surface of the finishing layer 9, heat is transferred from the outer surface of the finishing layer 9 to the main building structure 8 through heat conduction. When the finishing layer 9 does not have hollowing defects, heat is transferred only through a single thermal resistance, namely the cement mortar material of the finishing layer 9. According to the heat conduction formula, the amount of heat transferred over a period of time under steady-state conditions is shown in formula (1): (1); in: : Heat (unit: joule, ), which represents the amount of heat transferred over a period of time; : Thermal conductivity of the finishing layer 9 (unit: ); : The cross-sectional area of ​​the finishing layer 9 perpendicular to the heat flow direction (unit: square meters, ); : Temperature difference between the inner and outer surfaces of the finishing layer 9 (unit: or ),Right now ; : Finishing layer 9 thickness (unit: meter, );: time (unit: seconds, ).

[0037] When the finishing layer 9 has a hollow defect, the finishing layer + air is used as a series composite material, and heat transfer is transferred through the series composite material. The thermal resistance of the series composite structure is The calculation formula is shown in formula (2), the thermal resistance of each material It can be solved by formula (3).

[0038] (2); (3); in: : Effective thermal resistance of the series composite structure (unit: ); : Thermal resistance of finishing material (unit: ); : Thermal resistance of air (unit: ); : Material thickness (unit: ); : Heat transfer area of ​​the material (unit: ).

[0039] Substituting formula (3) into formula (2) yields formula (4).

[0040] (4); in: : Thickness of the finishing layer (Unit: ); : The thickness of the air layer (unit: ); : Thickness of the tandem composite material (unit: ), .

[0041] By rearranging formula (4), the effective thermal conductivity of the series composite material is obtained: , see formula (5).

[0042] (5); According to formula (5), , when the same amount of heat is transferred in the same time, the temperature difference between the inside and outside of the series composite material can be obtained. According to formula (1), after the defect occurs, the temperature difference Get bigger.

[0043] Theoretical calculation: The thickness of the building facade finishing layer is 14 , the thermal conductivity of cement mortar is 0.9 , the area is 0.04 The temperature difference between the inner and outer surfaces of the finishing layer is 3 , heat transfer time is 3600 , according to formula (1) the heat transferred .

[0044] When hollowing occurs between the finishing layer and the main body of the building, an air gap of 1mm is generated, forming a composite material of finishing layer + air in series. The thermal conductivity of air is known to be 0.026. , then according to formula (5) the effective thermal conductivity of the composite material is , then the same amount of heat is transferred in the same time. According to formula (1), the temperature difference on both sides of the composite material is .

[0045] Through the above example, before and after the hollowing between the finishing layer and the building body occurs, the heat transferred to one side of the building body decreases significantly, resulting in a decrease in the temperature on that side. Specific embodiment three: like Figures 1 to 3 As shown, the following is a complete implementation of the content described in the above embodiment: The nine processes of the first embodiment (temperature data acquisition, intelligent heat flow analysis, adaptive identification of hollowing defects, dynamic temperature field optimization, self-learning correction of environmental interference, multi-source data fusion diagnosis, defect trend prediction, visual interaction, and intelligent resource allocation) are executed in the temperature acquisition and processing terminal 10, and the basis for determining hollowing defects is explained by using the heat conduction principle of the second embodiment.

[0047] Equipment Installation and Preparation: Before exterior wall renovation, select the main building structure 8 as the installation surface. Ensure the surface is clean and flat, free of cracks or loose materials. Fabricate the temperature measurement units 2: Use a 15mm x 5mm temperature sensor 4 (accuracy ±0.05°C) and secure it to the center of a 30mm diameter, 5mm thick aluminum foil insulation 5 using high-temperature-resistant glue. Preserve fixing holes 6 (2mm diameter) on the insulation surface. Use an ST38 nail gun (15mm length) to secure the temperature measurement units 2 to the installation surface 8 in a rectangular array with 200mm x 200mm spacing. Drive the nails through the fixing holes 6 and into the wall 10mm to ensure a secure fit without damaging the sensors. Connect adjacent temperature measurement units 2 with high-temperature-resistant PVC wire 3 (1mm diameter) to form a temperature sensor matrix 1, covering the monitoring area (e.g., 100m², in a 4x4 or 8x8 grid). A temperature acquisition and processing terminal 10 (200mm × 150mm × 50mm, IP65 protection) is mounted on one side of the matrix and secured with embedded bolts. A monitor 7 (150mm × 100mm, 7-inch HD touchscreen, IP54 protection) with a built-in data processing module (ARM processor, 4GB RAM, 128GB storage) is mounted on the bottom. Matrix cables 3 connect to the 32-channel signal input port of the processing terminal 10 and connect to the monitor 7 via RS485. An environmental sensor module (including sensors for temperature ±0.1°C, humidity ±2%RH, wind speed ±0.3m / s, solar intensity ±5W / m², air pressure ±0.5hPa, and rainfall ±0.1mm) is installed on top of mounting surface 8. Data is transmitted to the processing terminal 10 via Wi-Fi. A finishing layer 9 (cement mortar, tiles, or paint, 14mm thick) is applied after the matrix is ​​installed and cured for 28 days to ensure stability. During this period, the cable signals are tested for normal operation.

[0048] Implementation of the monitoring method: All processes are executed in the data processing module of the temperature acquisition and processing terminal 10. The specific steps are as follows: Temperature data collection process: Temperature sensor matrix 1 is activated to collect surface temperature data on mounting surface 8 at a 10ms frequency. Based on real-time data from environmental sensors (light intensity 0-2000 W / m², temperature -20°C to 50°C, wind speed 0-20 m / s), the sampling frequency is dynamically adjusted: increasing to 5ms for light intensity >1000 W / m² or temperature >30°C, and decreasing to 20ms at night or when temperature <10°C), optimizing collection efficiency. Data is transmitted via wire 3 to processing terminal 10 and stored in a local database. The default sampling period is 24 hours per day, running continuously.

[0049] The intelligent heat flow analysis process receives temperature data and uses a convolutional neural network (CNN)-optimized heat flow distribution analysis method. This method combines the thermal conductivity of the finishing layer (e.g., 0.9 W / (m·K) for cement mortar and 1.1 W / (m·K) for ceramic tiles) with the thermal conductivity characteristics of the mounting surface (e.g., 8) to generate a 20mm resolution heat flux density distribution map. The CNN model is trained using 45 days of historical temperature data and a material database (including cement mortar, ceramic tiles, paint, and stone). It automatically identifies areas of abnormal heat flow and generates feature vectors (heat flow intensity, location, timestamp, material type, and defect depth estimate). The model supports online model updates to adapt to different construction techniques. The results are then transmitted to the adaptive hollowing defect identification and multi-source data fusion diagnosis process.

[0050] The adaptive hollowing defect identification process receives a heat flux distribution map and uses a deep learning-based cliff-like temperature change detection method. Five time windows are set: 3, 7, 14, 21, and 30 days. The method detects abnormal events with a temperature drop rate greater than 0.25°C / hour or a fluctuation standard deviation greater than 1.2°C. Combined with reinforcement learning, the method dynamically optimizes thresholds based on historical false alarm data, cross-region temperature comparisons (temperature differences between adjacent measurement units), and material properties. This generates preliminary diagnostic results (including hollowing location coordinates, temperature change amplitude, confidence score, defect severity, and expansion direction), achieving accuracy exceeding 95%.

[0051] The dynamic temperature field optimization process receives temperature data and uses Gaussian process regression technology to construct a dynamic temperature field model with a 50mm resolution across the eight surfaces of the installation plane, based on a 200mm grid distribution. When a local temperature anomaly is detected (deviation > 1°C), the grid is automatically refined to 15mm. Heat flow prediction is optimized based on building geometry (wall curvature), wind direction, and material properties, reducing blind spots (coverage > 98%). The optimization results are then transmitted to the multi-source data fusion diagnostic process.

[0052] Environmental interference self-learning correction process: Temperature, humidity, wind speed, sunshine intensity, air pressure, and rainfall data are collected through environmental sensors. A Bayesian inference adaptive interference filtering method is used, with initial weights of 0.45 for temperature, 0.25 for humidity, 0.15 for sunshine, 0.05 for wind speed, 0.03 for air pressure, and 0.02 for rainfall. Weights are optimized over a 21-day learning cycle to support cross-seasonal (winter and summer), day and night, and extreme weather conditions (heavy rain and strong winds). Temperature data errors are corrected to within ±0.08°C, and the correction results are transmitted to the multi-source data fusion diagnostic process.

[0053] The multi-source data fusion diagnostic process receives heat flux distribution maps, preliminary diagnostic results, and corrected temperature data. A random forest classification model is used to construct a feature vector (temperature change rate, heat flux change rate, environmental correction factor, time series stability, and material thermal conductivity, with weights of 0.35, 0.3, 0.2, 0.1, and 0.05). A hollowing defect confidence score is calculated. When the score is >0.9, a diagnostic report is generated (including hollowing location, area error ±1.5%, material type inference, repair priority, and construction risk assessment) and transmitted via AES-256 encryption to the visual interactive process.

[0054] Defect trend prediction process: After receiving the diagnosis report, a generative adversarial network (GAN) is used to analyze 21 days of temperature, heat flow, confidence, and environmental data to predict the 96-hour risk of hollowing expansion. GAN optimizes accuracy through adversarial training and generates a risk probability curve. When the probability exceeds 85%, a multi-level warning (low, medium, high, and emergency) is triggered. An early warning report is generated, including the repair time window, material selection (such as high-strength mortar), and construction safety tips, and transmitted to the visual interactive process.

[0055] The interactive visualization process receives diagnostic reports and prediction results and generates a six-part interactive interface: 1) A 3D temperature field heat map, displaying the temperature distribution across the installation plane. Hollow areas are highlighted in red with a gradient edge, supporting 360° rotation and 3D cross-sectional views; 2) A hollow defect distribution map, annotated with location, area (±1.5%), confidence level, depth estimate, and material type, with dynamic zoom and animation; 3) A risk trend curve, displaying 96-hour risk probability, with customizable ranges from 24 hours to 30 days; 4) A user interaction module, supporting touchscreen or mobile app viewing, view adjustment, warning threshold setting, and PDF report export, with voice control (in Chinese, English, Japanese, and German) and gesture control; 5) A historical data backtracking module, displaying 60-day temperature, heat flow, and defect trends, with anomaly annotation and playback support; 6) An intelligent suggestion module, generating repair techniques (such as local grouting), material recommendations, construction time optimization, and matching local supplier databases. This interface is transmitted to user terminals (mobile phones and PCs) via 5G / Wi-Fi, supporting simultaneous multi-user access (with administrator and engineer permissions) and offline caching.

[0056] The intelligent resource allocation process receives data on power consumption, computing latency, data throughput, and memory usage for each process, then uses deep reinforcement learning (DRL) to optimize resource allocation. It predicts three-hour demand based on 21 days of operational data and adjusts the resource allocation ratio (temperature data collection 26%, heat flow analysis 24%, and hollowing detection 20%). When hardware faults are detected (such as sensor disconnection or processor overheating), the system automatically isolates the faulty module, reallocates resources, and generates a diagnostic report (including fault type, impact scope, recovery recommendations, estimated recovery time, and backup plans) that is transmitted to the visual interactive process. Power consumption is kept below 80% of rated power.

[0057] Principle of hollowing defect detection (see Example 2): The monitoring method is based on thermal conductivity differences. Sunlight shines on the finishing layer 9, transferring heat from the outer surface to the mounting surface 8. Under normal circumstances, heat is transferred through a single finishing layer (such as cement mortar, with a thermal conductivity of 0.9 W / (m·K)). However, the hollowing area forms a series structure of finishing layer + air (thermal conductivity of 0.026 W / (m·K)), increasing thermal resistance and causing a significant temperature drop on the mounting surface 8 side (the temperature difference increases by approximately 10 times, as calculated in Example 2). Temperature sensor 4 detects a sudden temperature drop (>0.25°C / hour) and, combined with multi-scale analysis and fusion diagnosis, determines a hollowing defect.

[0058] Implementation Results: This method, leveraging modular processes, CNN, deep learning, GAN, and DRL algorithms, achieves high-precision (confidence level > 0.9), non-destructive, automated monitoring. It is adaptable to cross-seasonal and extreme weather conditions, predicts 96-hour hollowing risk, and generates repair recommendations. It outperforms percussion testing (lower accuracy), ultrasonic testing (costly and complex), infrared thermal imaging (high environmental interference), mortar pull-off (destructive), and drone aerial photography (requiring professional analysis). The interface supports remote monitoring and multi-user collaboration, reducing maintenance costs and making it suitable for dynamic management during service life. Specific embodiment four: like Figures 1 to 3 As shown, based on the above content, the following control system use cases are provided: Use Case 1: Hollowing Monitoring of Exterior Walls of High-Rise Residential Buildings: Scenario: A new high-rise residential building (30 stories, with a wall area of ​​approximately 5,000 m²) was constructed in a certain city. The exterior walls were finished with cement mortar (14 mm thick, with a thermal conductivity of 0.9 W / (m·K)). Hollowing defect monitoring was required during the initial service period (six months after completion) to ensure long-term safety.

[0060] Equipment Installation: After 28 days of construction and curing, a temperature sensor matrix 1 was installed on the exterior surface of the building's main structure (installation plane 8), covering the main exterior wall area (approximately 1,000 m²). Temperature measurement units 2 were arranged at 200 mm x 200 mm intervals (approximately 25,000 units). Temperature sensors 4 (15 mm x 5 mm) were adhered to aluminum foil insulation 5 (30 mm diameter, 5 mm thickness) and secured using ST38 air nails. Wires 3 were connected to form a matrix and connected to temperature acquisition and processing units 10 (IP65 protection, four units located at the four corners of the wall). Each unit was snapped onto a monitor 7 (7-inch touch screen) at the bottom. Environmental sensor modules (temperature, humidity, wind speed, sunlight intensity, air pressure, and rainfall) were installed on the roof, with data transmitted via Wi-Fi.

[0061] Monitoring method implementation: Temperature data collection: In summer (ambient temperature 30°C, light intensity 1200W / m²), temperature data is collected at a frequency of 5ms, which is reduced to 20ms at night. The collection period is 30 days.

[0062] Intelligent heat flow analysis: The CNN model (trained on 45 days of historical data) generates a 20mm resolution heat flux density distribution map and identifies five areas of abnormal heat flow (abnormal intensity > 0.5W / m²).

[0063] Adaptive identification of hollowing defects: Detected three locations with abrupt temperature drops (rate 0.3°C / hour, standard deviation 1.5°C), optimized the threshold using reinforcement learning, and generated a preliminary diagnosis (locations: southeast facade, 3rd, 7th, and 15th floors, confidence level 0.92).

[0064] Dynamic optimization of the temperature field: Construct a 50mm resolution temperature field, encrypt local abnormal areas to 15mm, achieve a coverage rate of 99%, and confirm the abnormal location.

[0065] Environmental interference self-learning correction: Corrects for high temperature and high humidity (80% RH) interference with an error of ±0.08°C. It learns over 21 days to adapt to summer environments.

[0066] Multi-source data fusion diagnosis: A random forest model (weights: temperature 0.35, heat flow 0.3, environment 0.2, stability 0.1, material 0.05) generated a diagnostic report, confirming three hollow areas (areas 0.2m², 0.3m², and 0.5m², with an error of ±1.5%). It was recommended to prioritize repairing the 15th floor hollow area.

[0067] Defect trend prediction: GAN predicts a 96-hour risk, with a 15-layer hollowing risk probability of 88% (high). Repair is recommended within 48 hours.

[0068] Visual interaction: Monitor 7 displays a 3D heat map (red highlights hollows), a distribution map (annotated area), and a risk curve (96 hours). It also offers remote access via a mobile app, voice control to switch between Chinese and English, and the ability to export PDF reports.

[0069] Intelligent resource allocation: DRL optimizes resources (temperature acquisition 26%, heat flow analysis 24%), controls power consumption at 80%, and automatically isolates and generates a report if a sensor is detected to be disconnected.

[0070] Results Output and Processing: The diagnostic report revealed the most severe hollowing on the 15th floor of the southeast facade (0.5 m², confidence level 0.95). The intelligent suggestion module recommended local grouting repair (high-strength mortar), matched with a local supplier, and optimized the construction time to nighttime. The property management team monitored the situation in real time via a mobile app and arranged for repairs within 48 hours, mitigating safety risks.

[0071] Theoretical Basis (Example 2): The thermal resistance of the hollow area increases (thermal conductivity of air is 0.026 W / (m·K)), and the temperature drops significantly (the temperature difference increases by about 10 times), which is consistent with the cliff-like drop feature.

[0072] Use Case 2: Ceramic hollowing monitoring of exterior wall tiles in commercial complexes: Scenario: A commercial complex (five stories, 2000 m² wall area) with exterior ceramic tiles (10 mm thick, 1.1 W / (m·K) thermal conductivity) experienced complaints of localized hollowing after three years of service. Accurately locate and predict any further risks.

[0073] Equipment Installation: Temperature measurement units 2 were drilled into the existing wall (200mm spacing, approximately 10,000 units). Temperature sensors 4 were fixed to aluminum foil insulation 5, secured with air nails to avoid damaging the tiles. A temperature sensor matrix 1 covered a key area (the south facade, 500m²). Wires 3 connected to two temperature acquisition and processing terminals 10, with monitors 7 attached at the bottom. An environmental sensor module was installed on the roof, collecting temperature, humidity, wind speed, sunlight intensity, air pressure, and rainfall, using 4G data transmission.

[0074] Monitoring method implementation: Temperature data collection: In spring (temperature 15°C, humidity 70%), the frequency of data collection is 10 ms, which drops to 15 ms during rainfall. The collection period is 60 days.

[0075] Intelligent heat flow analysis: The CNN model (adapted to the characteristics of the tiles) generates a 20mm heat flux density distribution map and identifies four heat flow anomalies (intensity > 0.6W / m²).

[0076] Adaptive identification of hollow defects: Detected two locations with a cliff-like drop (rate 0.28°C / hour), used reinforcement learning combined with tile thermal conductivity to optimize the threshold and generate a diagnosis (south facade, floors 2 and 4, confidence level 0.90).

[0077] Dynamic optimization of temperature field: 50mm resolution temperature field, with abnormal areas encrypted to 15mm, combined with wind direction optimization, with a coverage rate of 98.5%.

[0078] Environmental interference self-learning correction: Corrects for rainfall and wind speed interference, with an error of ±0.08°C, and learns over 21 days to adapt to the rainy spring environment.

[0079] Multi-source data fusion diagnosis: The random forest model confirmed two hollow areas (areas of 0.15m² and 0.25m²), and it was recommended to prioritize the repair of the hollow on the fourth floor (near the main entrance).

[0080] Defect trend prediction: GAN predicts a 96-hour risk, with a high risk of 83% for four-layer hollowing. Repair is recommended within 72 hours.

[0081] Visual interaction: 3D heat maps display hollow areas, distribution maps mark areas, risk curves support 30-day forecasts, and mobile app gestures allow zooming in on four layers of hollow areas and exporting reports.

[0082] Intelligent resource allocation: DRL optimizes resources, reduces power consumption to 79%, detects processor overheating, automatically reduces frequency, and generates recovery suggestions.

[0083] Result Output and Processing: A diagnostic report confirmed a hollowing problem on the fourth floor of the south facade (0.25 m², confidence level 0.93). The intelligent suggestion module recommended tile re-laying and grouting, matched a supplier, and scheduled the work for a weekend night. Management monitored progress through multi-user access to ensure no disruption to business operations.

[0084] Theoretical basis (Example 2): The thermal resistance of the tile + air series structure increases, and the temperature drops significantly. The monitoring method accurately locates the temperature through the cliff-like characteristics.

[0085] Use Case 3: Hollowing Monitoring of Exterior Wall Paint of Historical Buildings: Scenario description: A historical building (2 stories, 800m² wall area) with exterior paint finishes (5mm thickness, thermal conductivity 0.5W / (m·K)) requires non-destructive hollowing monitoring to protect the value of the cultural relics. A comprehensive inspection is required after 10 years of service.

[0086] Equipment Installation: Removable adhesive was used to secure the temperature measurement units 2 (200mm spacing, approximately 4,000 units) without damaging the coating. Temperature sensors 4 were affixed to aluminum foil insulation 5, and air nails were replaced with high-strength adhesive. The temperature sensor matrix 1 covered the front wall (200m²). Wires 3 were connected to a temperature acquisition and processing terminal 10, which was then snapped onto a monitor 7. An environmental sensor module was installed under the eaves, collecting multi-factor data and transmitting it over 5G.

[0087] Monitoring method implementation: Temperature data collection: In winter (0°C, 500 W / m² of light), the frequency of temperature data collection is 15 ms, which is reduced to 25 ms at night. The collection period is 90 days.

[0088] Intelligent heat flow analysis: The CNN model (adapted to the low thermal conductivity of the coating) generates a 20mm heat flow distribution map and identifies three anomalies (intensity > 0.4W / m²).

[0089] Adaptive identification of hollowing defects: Detected one abrupt temperature drop (rate 0.26°C / hour), using reinforcement learning combined with cross-region comparison to generate a diagnosis (positive layer 1, confidence level 0.91).

[0090] Dynamic optimization of temperature field: 50mm temperature field, encrypted to 15mm, combined with architectural surface optimization, coverage rate of 99.2%.

[0091] Environmental interference self-learning correction: Corrects low temperature, high humidity (90%RH), and rainfall interference with an error of ±0.08°C, adapting to winter environments.

[0092] Multi-source data fusion diagnosis: The random forest model confirmed one hollowing site (area 0.1m²) and recommended low-invasive repair.

[0093] Defect trend prediction: GAN predicts a 96-hour risk with a probability of 60% (medium), and recommends repair within 7 days.

[0094] Visual interaction: heat maps show hollows, distribution maps mark depth, 3D cross-section views assist in cultural relic protection assessments, voice control switches between Chinese, English and German, and offline data caching.

[0095] Intelligent resource allocation: DRL optimization, power consumption reduced by 78%, and detection of a sensor failure due to low temperature, automatic isolation, and generation of a backup plan.

[0096] Results Output and Processing: The diagnostic report confirmed a single hollowing defect (0.1 m², confidence level 0.94) on the facade. The intelligent suggestion module recommended minimally invasive spray repair, using an environmentally friendly material supplier and scheduling the repair during rain-free weather. The cultural relics preservation team used a mobile app to review 60 days of data and confirmed that the hollowing defect had not expanded, ensuring minimal impact from the repair.

[0097] Theoretical basis (Example 2): The increase in thermal resistance of paint + air leads to a temperature drop, and the monitoring method protects the cultural relic wall through high-precision analysis. Specific embodiment five: like Figures 1 to 3 The following are the specific experimental data: Experimental purpose: To verify the hollowing detection accuracy, environmental adaptability, prediction accuracy and diagnostic reliability of the monitoring method, and compare it with infrared thermal imaging and tapping methods.

[0099] condition: Test objects: Three groups of 1m² exterior wall samples, with the finishing layers consisting of cement mortar (14mm, thermal conductivity 0.9W / (m·K)), ceramic tile (10mm, 1.1W / (m·K)), and paint (5mm, 0.5W / (m·K)). Each group was pre-set with 1-3 hollows (area 0.05-0.3m², air layer 1mm).

[0100] Equipment: Temperature sensor matrix 1 (200mm x 200mm spacing, 25 temperature measurement units 2), temperature sensor 4 (accuracy ±0.05°C) fixed to aluminum foil insulation 5, wires 3 connected to temperature acquisition and processing terminal 10, monitor 7 (7-inch touch screen) displays the results. Environmental sensors collect temperature (0-35°C), humidity (50-90% RH), wind speed (0-10m / s), sunlight intensity (0-1500W / m²), air pressure (980-1020hPa), and rainfall (0-5mm / h).

[0101] Environment: Laboratory conditions simulate summer (35°C, 80% RH, 1000W / m²), spring (15°C, 90% RH, 2mm / h rainfall), and winter (0°C, 60% RH, 500W / m²) for 30 days.

[0102] Comparison methods: infrared thermal imaging method (accuracy ±2°C, Fluke Ti 400 thermal imager) and tapping method (manual judgment, relying on experience).

[0103] Process: Run 9 processes (temperature data acquisition, intelligent heat flow analysis, adaptive identification of hollow defects, dynamic optimization of temperature field, self-learning correction of environmental interference, multi-source data fusion diagnosis, defect trend prediction, visual interaction, and intelligent resource allocation).

[0104] Indicators: detection accuracy (area error ±1.5%), correction error (±0.08°C), prediction accuracy (>85%), confidence level (>0.9), power consumption (<80%), false alarm rate, detection time, and non-destructiveness.

[0105] Theoretical Basis (Example 2): The hollowing area is identified through multi-scale analysis and fusion diagnosis due to the increased thermal resistance of the air layer (thermal conductivity 0.026 W / (m·K)), resulting in a drastic temperature drop (temperature difference increases approximately 10 times).

[0106] index Sample 1: Cement mortar Sample 2: Ceramic tiles Sample 3: Paint Infrared thermography Percussion Experimental environment Summer (35°C, 80%RH, 1000W / m²) Spring (15°C, 90% RH, 2 mm / h rainfall) Winter (0°C, 60%RH, 500W / m²) Same environment Same environment Preset number of hollow drums 3 (0.1m², 0.2m², 0.3m²) 2 (0.05m², 0.15m²) 1（0.1m²） Same as above Same as above Detection of hollow quantity 3 2 1 2 (1 missed) 2 (1 missed) Hollowing position error ±4mm ±3mm ±3mm ±25mm ±60mm Area error ±1.1% (0.099m²,0.198m², 0.296m²) ±1.2% (0.049m², 0.148m²) ±1.0%(0.099m²) ±6.0% ±12.0% Confidence score 0.96、0.94、0.93 0.93、0.92 0.95 0.78 0.65 Temperature correction error ±0.07°C ±0.08°C ±0.06°C ±1.8°C No correction 96-hour risk prediction accuracy 92% (2 / 3 expansion, high alert) 88% (1 / 2 expansion, medium warning) 86% (not expanded, low alert) No forecast No forecast False alarm rate 0.8% (1 / 125 area) 0.6% (0 / 80 area) 0.4% (0 / 25 areas) 12% 18% Detection time 25 minutes (automated) 20 minutes (automated) 18 minutes (automated) 70 minutes 150 minutes (manual) Power consumption ratio 77% 78% 76% 88% None (manual) Non-destructive yes yes yes yes no Data source: 1. Data sources for this method: Experimental data collection: Data is derived from laboratory simulation experiments using 1m² exterior wall panels (cement mortar, ceramic tiles, and paint). Temperature data was collected using a temperature sensor matrix (25 temperature measurement units, ±0.05°C accuracy). Environmental parameters were collected using environmental sensors (accuracy: temperature ±0.1°C, humidity ±2%RH, wind speed ±0.3m / s, solar intensity ±5W / m², air pressure ±0.5hPa, rainfall ±0.1mm). The experiment was conducted in a controlled environmental chamber (simulating summer, spring, and winter) for 30 days. Each panel group was tested five times, and the average value was calculated.

[0107] Algorithm processing: The temperature acquisition and processing terminal 10 (ARM processor, 4GB memory, 128GB storage) runs nine processes. The specific algorithms include: Intelligent heat flow analysis: The CNN model is trained based on 45 days of historical temperature data and a material database (thermal conductivity: cement mortar 0.9 W / (m·K), ceramic tile 1.1 W / (m·K), paint 0.5 W / (m·K)), generating a 20mm resolution heat flow map.

[0108] Adaptive hollowing defect identification: Deep learning multi-scale analysis (3-30 day window), detection of cliff drops (>0.25°C / hour), and reinforcement learning threshold optimization.

[0109] Environmental correction: Bayesian inference, 21-day learning period, correction error ±0.08°C.

[0110] Fusion diagnosis: Random forest model, 5 feature weights (temperature 0.35, heat flow 0.3, environment 0.2, stability 0.1, material 0.05), confidence > 0.9.

[0111] Trend prediction: GAN analyzes 21 days of data and predicts 96 hours of risk with an accuracy rate of >85%.

[0112] Resource allocation: DRL optimization, power consumption <80%.

[0113] Verification method: The location and area of ​​the pre-set hollowing are calibrated using an X-ray imager (accuracy ±1mm) and used as the true value. The detection results are output via monitor 7, and the error is calculated by comparing the true value with the detected value. The prediction accuracy is verified by actual expansion data over the subsequent 30 days.

[0114] Reliability: The experimental equipment is calibrated in accordance with ISO 17025 standards. The data processing module logs each run, and the standard deviation of five replicates is less than 5%, ensuring data consistency. The scientific validity of the precipitous drop was verified by referring to the heat conduction calculations in Example 2 (increasing the thermal resistance of the air layer and increasing the temperature difference by approximately 10 times).

[0115] 2. Infrared thermal imaging data source: Experimental data collection: A Fluke Ti400 thermal imager (spatial resolution 15mm, temperature accuracy ±2°C) was used to test the same sample panels in the same environments (summer, spring, and winter). A 1m² sample panel was scanned, the temperature distribution was recorded, and the thermal image was manually analyzed to identify hollows. The test was repeated five times, and the average value was calculated.

[0116] Data Processing: Fluke SmartView software analyzes thermal imaging data, annotates areas of temperature anomaly (>2°C difference), and calculates hollowing location and area. False alarm rate is based on the deviation of the anomaly area from the X-ray calibration value.

[0117] Recording limitations: Temperature error is ±1.8°C in high temperature and rainfall environments, and small hollow areas (<0.1m²) are missed. No prediction function is available. Data is derived from the device's internal log and manual records and conforms to manufacturer specifications.

[0118] Reliability Note: The device is calibrated in accordance with ASTM E1934 standards, but is affected by environmental interference (rainfall, sunlight), resulting in a 12% false alarm rate (caused by high temperature and humidity).

[0119] 3. Data source of the tapping method: Experimental data collection: Two experienced inspectors (with more than 10 years of experience in building inspection) used a hammer (standard 5N force) to knock on the sample, listen to the sound changes to determine the hollowness, record the location and area, repeat the test 5 times, and take the average value.

[0120] Data processing: Strike points were manually recorded (on a 50mm grid). Hollows were annotated by sound differences (hollow areas sound hollow), and their area was estimated using a ruler. False positive rates were based on deviations from X-ray calibration values.

[0121] Limitations: Position error ±60mm, area error ±12.0%, missed detection of small hollow areas (<0.1m²), inspection time 150 minutes, minor surface damage may occur due to knocking, no predictive function. Data is derived from the inspector's record sheet and relies on subjective judgment.

[0122] Reliability description: Data is affected by personnel experience, with a repeatability deviation of 10%-15%, which complies with the building inspection industry standard (GB / T50344-2019), but has poor non-destructive properties.

[0123] Data Analysis and Conclusions: This method detects all hollows (3 / 3, 2 / 2, 1 / 1) with a position error of ±3-4mm, an area error of ±1.0-1.2%, a confidence level of 0.92-0.96, a correction error of ±0.06-0.08°C, a prediction accuracy of 86-92%, a false alarm rate of 0.4-0.8%, a detection time of 18-25 minutes, a non-destructive method, and power consumption of 76-78%. Advantages include intelligent technology (CNN, GAN, DRL), environmental adaptability (Bayesian correction), prediction capabilities (96 hours), and a low false alarm rate.

[0124] Infrared thermal imaging method: missed detection of 1 small hollow area, position error ±25mm, area error ±6.0%, confidence level 0.78, correction error ±1.8°C, false alarm rate 12%, detection time 70min, no prediction function, affected by high temperature and rainfall.

[0125] Tapping method: missed 1 hollow drum, position error ±60mm, area error ±12.0%, confidence level 0.65, false alarm rate 18%, detection time 150min, poor non-destructiveness, no correction and prediction functions.

[0126] Theoretical verification: The heat conduction principle of Example 2 (the temperature difference in the hollow area increases by about 10 times) is consistent with the cliff-like drop in the experiment (>0.25°C / hour), proving the scientific nature of the method.

[0127] The experimental data of this method is verified through high-precision equipment, standardized processes and multiple scenarios. It is superior to traditional methods and is suitable for dynamic monitoring during the service period. Specific embodiment six: like Figures 1 to 3 As shown, the following are supplementary implementation details of the monitoring method and device, which improve the hardware parameters, algorithm implementation, exception handling, maintenance process and compliance description, based on specific embodiments one to five, to ensure the integrity and operability of the technical solution.

[0129] Hardware parameter refinement: Temperature sensor 4: Model DS18B20, size 15mm × 5mm, accuracy ±0.05°C, operating temperature -55°C to 125°C, power consumption 0.5mW, protection grade IP68, suitable for high temperature and high humidity environments.

[0130] Aluminum foil insulation cotton 5: diameter 30mm, thickness 5mm, thermal conductivity 0.03W / (m·K), fixing hole 6 diameter 2mm, temperature resistance -40°C to 150°C, in line with GB / T20473-2021 building insulation material standard.

[0131] Wire 3: PVC insulated high-temperature resistant wire, 1mm diameter, 300V withstand voltage, compliant with IEC60227 standard, supports 32-channel signal transmission.

[0132] Temperature acquisition and processing terminal 10: ARM Cortex-A53 processor, 4GB DDR4 memory, 128GB eMMC storage, 20W power consumption, IP65 protection grade (passes IEC60529 water and dust resistance test), supports RS485 and Wi-Fi (802.11ac) communications.

[0133] Environmental sensor module: Contains six sensors (temperature: LM35, ±0.1°C; humidity: HTU21D, ±2%RH; wind speed: FS3000, ±0.3m / s; sunlight intensity: BH1750, ±5W / m²; air pressure: BMP280, ±0.5hPa; rainfall: RG-11, ±0.1mm). Installed on the roof or wall, the number of sensors is optimized based on the area (one group per 500m²).

[0134] Monitor 7: 7-inch IPS touch screen, resolution 1024×600, power consumption 5W, IP54 protection, built-in data processing module, supports 5G (SA / NSA mode) and Wi-Fi transmission.

[0135] Algorithm implementation details: Thermal Flow Intelligent Analysis (CNN): This model uses a 5-layer convolutional neural network (3 convolutional layers, 2 pooling layers) and takes 45 days of temperature data (approximately 100 million points, 200mm grid, 10ms sampling) and a material database (100 materials, thermal conductivity 0.1-2.0W / (m·K)). Training takes 72 hours (GPU: NVIDIA RTX3060) to generate a 20mm resolution thermal flow map. The computational complexity is O(n²). The model is updated online monthly.

[0136] Adaptive hollowing defect recognition (deep learning): Based on an LSTM network, it analyzes a 3-30 day time window (approximately 50 million points) and detects abrupt temperature drops (>0.25°C / hour). Reinforcement learning uses the Q-learning algorithm, with a reward function based on false alarm rate and confidence level. After 30 days of training, the accuracy reaches 95% after threshold optimization.

[0137] Environmental correction (Bayesian reasoning): 21-day learning cycle, 6-factor data input (approximately 20 million points), initial weights (temperature 0.45, humidity 0.25, etc.) dynamically adjusted via a Bayesian network, error ±0.08°C, computational complexity O(n).

[0138] Fusion diagnosis (random forest): 100 decision trees, 5 eigenvectors (temperature change rate, heat flux density change rate, etc.), 21 days of training data (approximately 150 million points), confidence level > 0.9, error ±1.5%.

[0139] Trend prediction (GAN): Generator (3-layer fully connected network) and Discriminator (4-layer convolutional network), trained on 21 days of data, predicting 96 hours of risk, with an accuracy rate of >85%, and computational complexity of O(n³).

[0140] Resource Reduction and Restriction (DRL): Based on DeepQ-Network, the reward function is power consumption (<80%) and latency (<100ms). We trained on 21 days of run logs and optimized the proportions of nine processes (such as temperature acquisition (26%) and heat flow analysis (24%)).

[0141] Abnormal situation handling: Network interruption: The processing terminal 10 has a built-in 4GB cache that stores 24 hours of data. It automatically retransmits after detecting an interruption (retry 3 times, every 5 seconds), and ensures connection through 5G / Wi-Fi dual-mode switching.

[0142] Data loss: Redundant backup is used (cross-validation of 4 sensors per group). Automatic interpolation (Gaussian process regression) is used when the loss rate is <0.1%. An alarm is triggered when the loss rate is >5%.

[0143] Sensor failure: When the processing end 10 detects a signal interruption (<0.01V), it automatically switches to an adjacent sensor and generates a failure report (location, time, and replacement recommendations). The failure rate is <0.5%.

[0144] In extreme weather conditions: blizzards (-20°C) and typhoons (wind speed > 20m / s), the sensor module enters low-power mode (sampling frequency drops to 50ms), and the processing end 10 uses a backup battery (12 hours of battery life).

[0145] Maintenance and calibration procedures: Sensor calibration: The temperature sensor 4 is calibrated every 6 months using a standard thermocouple (accuracy ±0.01°C) in accordance with ISO17025 standards, with a calibration deviation of <0.05°C.

[0146] System update: Update the material database (10 new materials) and CNN / GAN model (5% accuracy optimization) annually, and upgrade the processing terminal 10 via OTA firmware, which takes 2 hours.

[0147] Hardware maintenance: Check the connection of wire 3 and the fixation of aluminum foil insulation cotton 5 every three months, clean the dust on the environmental sensor, and ensure IP65 / IP54 protection.

[0148] Data management: Stores 60 days of data (approximately 10TB) and automatically archives it to the cloud (with AES-256 encryption), complying with the Data Security Act and GDPR privacy requirements.

[0149] Compliance Notes: Building standards: The methods and equipment comply with GB / T50344-2019 "Technical Standards for Building Structure Inspection" and GB50204-2015 "Concrete Structure Engineering Construction Quality Acceptance Code".

[0150] Data privacy: Diagnostic reports and prediction data are transmitted via AES-256 encryption, with access rights graded (administrators, engineers), and log storage compliant with ISO27001 information security standards.

[0151] Environmental adaptability: The equipment has passed IEC60529 (IP65 / IP54) and IEC60068-2 (temperature, humidity, and vibration tests), and is suitable for coastal high-salinity and desert high-temperature environments.

[0152] Regarding hardware compatibility, the temperature sensor matrix undergoes adhesion testing during installation to verify the long-term compatibility of the aluminum foil insulation with the finishing layer, ensuring that the air-nail fixing method does not affect the quality of the finishing layer. For irregular structures such as curved walls, a flexible grid array and adaptive interpolation algorithm can be used to adjust the distribution of temperature measurement units. For hardware weatherability, the sensors must pass a 5000-hour UV exposure performance degradation test. The conductors are made of double-shielded PVC to withstand temperatures ranging from -40°C to 150°C. Sensors are calibrated every two years, and the conductors are replaced every five years. In the algorithm implementation, the CNN model input is a 100×100 spatiotemporal matrix (24-hour time step × 200mm spatial grid). The reinforcement learning reward function is set to R = 0.6 × (1-false alarm rate) + 0.4 × confidence level. When fusion is performed on multiple features, the temperature change rate is calculated as ΔT / Δt (Δt = 1 hour). For extreme environments, sensors are equipped with anti-corrosion coatings and heating insulation sleeves. The algorithm uses Kalman filtering to process data from strong electromagnetic interference. When multiple environmental factors are coupled, dynamic adjustments are made based on a temperature weight of 0.45 + 0.1 × (correction value for humidity > 80% RH). Data management utilizes wavelet transform compression technology, compressing 10TB of data to 3TB of storage. A distributed processing framework enables parallel computation of 1.25 million points per second. The risk warning threshold is based on 1,000 sets of historical data (with an actual expansion rate of 90% when the risk probability is >85%). The repair window is calculated based on a defect expansion rate of v = 0.1 m² / day and a construction preparation time of 48 hours. For network security, VPN encrypted transmission and intrusion detection systems are implemented. Third-party inspection reports must include type inspection data from the China Academy of Building Research.

[0153] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further restrictions, an element defined by the statement "comprising a reference structure" does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0154] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring hollowing defects in a building exterior wall finishing layer, comprising: The monitoring method includes a temperature data acquisition process, a heat flow intelligent analysis process, a hollowing defect adaptive identification process, a temperature field dynamic optimization process, an environmental interference self-learning correction process, a multi-source data fusion diagnosis process, a defect trend prediction process, a visual interaction process, and a resource intelligent allocation process. All processes of the monitoring method are executed through the temperature acquisition processing end (10); The temperature data acquisition process collects the surface temperature data of the main building structure (8) in real time through the temperature sensor (4) in the temperature measuring unit (2). The temperature measuring unit (2) is distributed in a rectangular array with a spacing of 200mm×200mm. The temperature data is transmitted to the temperature acquisition processing end (10) at a frequency of milliseconds through the wire (3). The sampling frequency is dynamically adjusted according to the changes in ambient light, temperature and wind speed to optimize the data acquisition efficiency. The heat flow intelligent analysis process receives the temperature data of the temperature data acquisition process, adopts the heat flow distribution analysis method optimized by the neural network, combines the heat conduction characteristics of the finishing layer (9) and the main building structure (8), generates a heat flow density distribution map with a resolution of 20mm, and identifies potential hollow areas. domain, and transmit the results to the hollowing defect adaptive identification process and the multi-source data fusion diagnosis process; the hollowing defect adaptive identification process receives the heat flux density distribution map of the heat flux intelligent analysis process, adopts the cliff-type temperature change detection method based on deep learning, dynamically identifies the temperature abnormality pattern through multi-scale time series analysis, generates the preliminary diagnosis result of the hollowing defect, and optimizes the detection threshold through the self-learning mechanism to improve the accuracy; the temperature field dynamic optimization process receives the temperature data of the temperature data acquisition process, adopts the three-dimensional space interpolation and heat flux balance method, constructs the real-time temperature field model of the surface of the main structure (8) of the building, with a resolution of 50mm, and automatically optimizes the monitoring coverage of the temperature measurement unit (2) to reduce the blind area; The environmental interference self-learning correction process collects ambient temperature, humidity, wind speed, sunshine intensity, and air pressure data through external sensors, adopts an adaptive interference filtering method, dynamically learns the law of environmental changes, corrects the interference factors in the temperature data, and controls the error within ±0.08°C. The correction results are then transmitted to the multi-source data fusion diagnosis process; The multi-source data fusion diagnosis process receives the heat flux density distribution map of the heat flux intelligent analysis process, the preliminary diagnosis results of the hollowing defect adaptive identification process, and the correction data of the environmental interference self-learning correction process. It uses a multi-feature dynamic weighted fusion method to comprehensively consider temperature, heat flux, environmental factors, and time series stability to calculate the confidence score of the hollowing defect and generate a diagnosis report containing the defect location, area, and confidence level. The defect trend prediction process receives the diagnosis report from the multi-source data fusion diagnosis process, uses a trend prediction method based on a generative adversarial network to analyze the temperature and heat flow data of the past 21 days, predicts the risk of hollowing defect expansion in the next 96 hours, and generates a multi-level risk warning; The visualization interaction process receives the diagnostic report of the multi-source data fusion diagnosis process and the prediction results of the defect trend prediction process, and generates an interactive visualization interface including a three-dimensional temperature field heat map, a hollowing defect distribution map, and a risk trend curve. It supports users to adjust display parameters through a touch screen or mobile phone application, and provides remote real-time monitoring, multi-user collaboration, and data backtracking functions. The resource intelligent allocation process receives the real-time power consumption and computing load data of each process, adopts a dynamic resource scheduling method, optimizes the computing resource allocation of each process in the temperature acquisition and processing end (10) based on the real-time load and historical operation data, and supports fault self-diagnosis and automatic recovery functions.

2. A method for monitoring hollowing defects in a building exterior wall finishing layer according to claim 1, characterized in that: The heat flux distribution analysis method of the heat flux intelligent analysis process is based on a convolutional neural network, combined with the 200mm×200mm grid distribution of the temperature measurement unit (2) and the material properties of the finishing layer (9), to generate a heat flux density distribution map with a resolution of 20mm in real time; The heat flux distribution analysis method is trained using 45 days of historical temperature data and a material database to automatically identify areas of heat flux anomaly and generate a feature vector containing the heat flux anomaly intensity, location, timestamp, material type, and defect depth estimation. It supports online updating of the analysis model to adapt to different building exterior wall materials and construction processes.

3. The method for monitoring hollowing defects in a building exterior wall finishing layer according to claim 1, wherein: The cliff-like temperature change detection method of the hollowing defect adaptive identification process adopts multi-scale time series analysis and sets five analysis windows of 3 days, 7 days, 14 days, 21 days and 30 days. The cliff-like temperature change detection method is combined with a reinforcement learning mechanism to dynamically optimize the detection threshold based on historical false alarm data, cross-regional temperature comparisons and material properties, and generate preliminary diagnostic results including hollowing location coordinates, temperature change amplitude, confidence score, defect severity and potential expansion direction.

4. The method for monitoring hollowing defects in a building exterior wall finishing layer according to claim 1, wherein: The three-dimensional spatial interpolation and heat flow balance method of the temperature field dynamic optimization process is based on the grid distribution of the temperature measurement unit (2) and adopts Gaussian process regression technology to construct a dynamic temperature field model of the surface of the building main structure (8); the heat flow balance method supports adaptive grid adjustment. When a local temperature anomaly is detected, the grid resolution is automatically encrypted to 15mm, and the heat flow distribution prediction is optimized in combination with the building geometry, wind direction, and wall material to reduce monitoring blind spots. The results are transmitted to the multi-source data fusion diagnosis process.

5. The method for monitoring hollowing defects in a building exterior wall finishing layer according to claim 1, wherein: The environmental interference self-learning correction process collects ambient temperature, humidity, wind speed, sunshine intensity, air pressure, and rainfall data through external sensors, adopts an adaptive interference filtering method, and dynamically updates the weights of interference factors based on Bayesian reasoning. The initial weights are 0.45 for ambient temperature, 0.25 for humidity, 0.15 for sunshine intensity, 0.05 for wind speed, 0.03 for air pressure, and 0.02 for rainfall. The adaptive interference filtering method optimizes the weight distribution through a 21-day learning cycle, supports environmental adaptation across seasons, day and night, and extreme weather. The temperature data error after correction is controlled within ±0.08°C, and the correction results are transmitted to the multi-source data fusion diagnosis process.

6. The method for monitoring hollowing defects in a building exterior wall finishing layer according to claim 1, wherein: The multi-feature dynamic weighted fusion method of the multi-source data fusion diagnosis process constructs a feature vector including the temperature change rate, the heat flux change rate, the environmental correction factor, the time series stability, and the material thermal conductivity. The random forest classification model is used to dynamically adjust the feature weights, with the initial weights being 0.35 for the temperature change rate, 0.3 for the heat flux change rate, 0.2 for the environmental correction factor, 0.1 for the time series stability, and 0.05 for the material thermal conductivity. The multi-feature dynamic weighted fusion method generates a hollowing defect confidence score. When the score is greater than 0.9, a diagnostic report is generated including the hollowing location, area, material type inference, repair priority, and construction risk assessment. The report is transmitted to the visual interactive process via an encrypted protocol.

7. The method for monitoring hollowing defects in a building exterior wall finishing layer according to claim 1, wherein: The generative adversarial network trend prediction method of the defect trend prediction process receives temperature, heat flux, confidence score, and environmental data from the past 21 days to build a hollowing defect expansion risk model; The proposed generative adversarial network trend prediction method optimizes prediction accuracy through adversarial training and generates a risk probability curve for the next 96 hours. When the risk probability is greater than 85%, a multi-level warning is triggered and an early warning report containing a repair time window, material selection recommendations, and construction safety tips is generated and transmitted to the visual interactive process.

8. The method for monitoring hollowing defects in a building exterior wall finishing layer according to claim 1, wherein: The interactive visualization interface of the visualization interaction process includes the following six parts: 3D temperature field heat map, dynamically displays the temperature distribution on the surface of the main building structure (8), with hollow areas highlighted in red and edges using a gradient transition to distinguish potential defects. It supports 360° viewing angle rotation and 3D cross-section view switching; Hollow defect distribution map, with hollow location, area, confidence level, defect depth estimation, and material type marked, supporting dynamic zoom, area screening, and defect evolution animation; The risk trend curve shows the change in the probability of hollowing risk in the next 96 hours and supports user-defined prediction time range; The user interaction module supports adjusting the heat map perspective, setting risk warning thresholds, and exporting diagnostic reports through the touch screen or mobile application. The interface supports voice control, multi-language switching, and gesture operation; The historical data backtracking module displays the temperature, heat flow, and defect change trends over the past 60 days, and supports abnormal event annotation, playback, and comparative analysis; The intelligent suggestion module generates suggestions including repair process, material recommendation, and construction time optimization based on the diagnosis report and prediction results, and automatically matches the local supplier database; Interface data is transmitted to user terminals in real time via 5G networks and Wi-Fi, supporting multi-user simultaneous access, hierarchical permission management, and offline data caching.

9. The method for monitoring hollowing defects in a building exterior wall finishing layer according to claim 1, wherein: The dynamic resource scheduling method for the intelligent resource allocation process receives real-time power consumption, computational latency, data throughput, and memory usage data for each process and uses deep reinforcement learning to optimize resource allocation. Based on the past 21 days' operating data, the dynamic resource scheduling method predicts resource requirements for the next three hours and dynamically adjusts the allocation ratios for each process, allocating 26% to the temperature data acquisition process, 24% to the intelligent heat flow analysis process, and 20% to the adaptive hollowing defect identification process. When a hardware fault is detected, the dynamic resource scheduling method automatically isolates the faulty module, reallocates resources, and generates a diagnostic report containing the fault type, impact range, recovery recommendations, estimated recovery time, and backup plans, which is then transmitted to the visual interactive process.

10. A method for monitoring hollowing defects in a building exterior wall finishing layer according to any one of claims 1 to 9, and a device for monitoring hollowing defects in a building exterior wall finishing layer corresponding to the monitoring method, characterized in that: It includes a temperature sensor matrix (1), a temperature measurement unit (2) and a main building structure (8); The main building structure (8) is the outer surface of the main building structure. The main building structure (8) has a rectangular array of temperature measuring units (2) on its surface. The temperature measuring units (2) are distributed at a spacing of 200 mm × 200 mm. The temperature measuring units (2) include aluminum foil insulation cotton (5) and a temperature sensor (4). The temperature sensor (4) is fixed to the aluminum foil insulation cotton (5) by gluing. The surface of the aluminum foil insulation cotton (5) is provided with a fixing hole (6). The aluminum foil insulation cotton (5) is fixed to the surface of the main building structure (8) after being nailed through the fixing hole (6) by an air nail gun with an ST38 length of 15 mm. Wires (3) are connected between the temperature measuring units (2). The matrix-distributed temperature measuring units (2) and wires (3) constitute a complete temperature sensor matrix (1). A temperature acquisition and processing end (10) is connected to the side of the surface of the main building structure (8) close to the temperature sensor matrix (1). A monitor (7) is installed at the bottom of the integrated processing end (10), and the monitor (7) includes a display screen, and the display screen is used to display the temperature data and diagnostic results collected by the temperature sensor matrix (1); the surface of the building main structure (8) is covered with a finishing layer (9), and the finishing layer (9) is constructed on the temperature sensor matrix (1) through a traditional construction process; the temperature data collected by the temperature sensor matrix (1) is transmitted to the temperature collection and processing end (10) through a wire (3) and displayed by the monitor (7); when the temperature monitored by the temperature measuring unit (2) in the temperature sensor matrix (1) drops sharply, it is determined that a hollowing defect occurs in the finishing layer (9) on the surface of the building main structure (8); a data processing module is integrated in the temperature collection and processing end (10), and the monitoring method described in claims 1 to 9 is executed by the data processing module to process the temperature data, generate a diagnostic report and a prediction result, and configure a transmission chip for output.

Images