Building waterproof construction quality monitoring and early warning system based on Internet of Things

Abstract

The invention discloses a building waterproof construction quality monitoring and early warning system based on the Internet of Things, relates to the technical field of building engineering, and realizes omnibearing and real-time monitoring and intelligent early warning of a building waterproof construction area through a multi-layer architecture formed by a data acquisition layer, a data transmission layer, a data processing layer and an application layer. The data processing layer analyzes data features from a time sequence and a spatial distribution dimension through normalization, denoising processing and a multi-head attention mechanism in a Transform model, and accurately recognizes the abnormality and risk trend of a weak point in combination with dynamic threshold and trend analysis, and the application layer determines the risk trend of the weak point through multi-dimensional data visualization, historical playback and real-time interaction. And the early warning module is combined with intelligent reasoning to generate dynamic maintenance suggestions, a maintenance scheme is optimized, the construction quality and the maintenance efficiency are improved, the intelligent level of building waterproof monitoring is improved, risk prediction and accurate maintenance are realized, the management cost is reduced, and the system has wide application value.

Description

Technical Field

[0001] The present invention relates to the technical field of construction engineering, and specifically to an Internet of Things-based building waterproof construction quality monitoring and early warning system. Background Art

[0002] The quality of building waterproof construction is directly related to the service life and safety of buildings. However, the existing waterproof construction quality monitoring methods still have certain limitations, especially in terms of real-time performance, accuracy, and intelligence, which need to be improved.

[0003] Chinese invention patent CN117169086B discloses a method for detecting the construction quality of the underground waterproof layer of a building. It collects air humidity data and reflective image matrices through a hygrometer, and calculates the leakage position and water penetration volume using a water penetration model as an evaluation index for the waterproof layer construction quality. This method has the following advantages: wide coverage: it can simultaneously detect the waterproof quality of basement walls, floors, and ceilings; non-destructive detection: the waterproof layer structure will not be damaged during the detection process. However, this patent also has the following deficiencies: lack of real-time performance: due to relying on the data calculation of the humidity matrix and reflective matrix that change over time, the detection has a certain delay and cannot respond in a timely manner to sudden waterproof risks; single analysis dimension: this system mainly relies on humidity and reflective data, and does not combine other environmental parameters (such as temperature, pressure, etc.) to comprehensively evaluate waterproof weak points, which may cause some potential risks to be ignored.

[0004] Chinese invention patent CN114232868B discloses a BIPV building roof waterproof system and its waterproof detection method. Through the combined design of waterproof coiled materials and support structures, the waterproof ability of the roof is improved, and the detection and management of the waterproof system are realized through an electric control box. Its main advantages include: strong structural stability: the strength and durability of the roof are ensured through prefabricated design; double-layer waterproof structure: effectively reduces the risk of roof leakage and water seepage at the corner joints. However, this patent still has the following limitations: limited detection range: this system mainly focuses on the waterproof detection of the roof and does not cover the waterproof monitoring of complex parts of the building (such as internal and external corners, pipe roots, etc.); lack of intelligent analysis ability: the system relies more on the optimization of the hardware structure, lacks intelligent monitoring and analysis of environmental data, and is difficult to predict risks in advance or provide maintenance suggestions. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and propose an Internet of Things-based building waterproof construction quality monitoring and early warning system to solve the above-mentioned problems.

[0006] The object of the present invention is achieved by the following technical solutions: An Internet of Things-based building waterproof construction quality monitoring and early warning system, comprising: a data acquisition layer, which is used to collect image data of the building waterproof construction area through image acquisition devices, and to monitor the environmental parameters and physical state changes of waterproof weak points in real time through humidity sensors, temperature sensors and pressure sensors; A data transmission layer, which is used to transmit the collected data to the data processing layer by using wireless communication technology, wherein the image data is transmitted through Wi-Fi, the sensor data is transmitted through LoRa or Bluetooth, and the data is encrypted to ensure data security and integrity; The data processing layer includes: An image recognition module, which uses a deep learning algorithm based on a convolutional neural network (CNN) to analyze the image data, identify the building waterproof weak points, and uniquely number the weak points according to a preset numbering rule; A data conversion and analysis module: Data conversion: Standardize the continuous data collected by the sensor, calculate the zero mean and unit variance, and perform normalization processing to map the data to an interval; at the same time, use a discretization method to convert the continuous numerical data into a Token sequence suitable for processing by the Transformer model, where different intervals correspond to unique Tokens; Multi-head attention analysis: Use the multi-head attention mechanism of the Transformer model to analyze the Token sequence, and each attention head analyzes the data from the dimensions of time series features and spatial distribution features respectively: Time series feature analysis: Some attention heads focus on the time dimension change of the sensor data. For example, analyze the Token sequence of the temperature sensor to identify the trend of temperature change over time, and judge the risk of performance degradation of the waterproof material caused by sharp temperature fluctuations; Spatial distribution feature analysis: Some attention heads analyze the Token sequence of the humidity sensor, and judge the spatial distribution of the humidity abnormal area by comparing the humidity values in different areas of the building to locate the possible leakage risk; Feature fusion and optimization: Generate a deep feature vector through splicing and linear transformation of the output of the multi-head attention mechanism to provide optimization support for anomaly detection and trend analysis; Anomaly detection and trend analysis module: Anomaly detection: Calculate the mean and standard deviation of the sensor feature values based on historical data, set a threshold range, and when the Token sequence feature value exceeds the threshold range, the system marks it as an anomaly and locates the anomaly position; Trend analysis: Use the moving average method and the exponential smoothing method to process the time series data, and predict the risk change trend of the weak points. For example, the continuous rise of humidity or temperature may indicate waterproof failure; The application layer includes: A digital twin platform for constructing a three-dimensional model of the building waterproof area and real-time mapping the status and changes of the waterproof system; An early warning and maintenance recommendation module that provides predictive maintenance recommendations based on data analysis results, including recommendations on repair time, materials, and methods, and issues early warnings through text messages, platform pop-ups, and email pushes.

[0007] The image acquisition device includes a high-resolution infrared night vision camera and has the function of dynamically adjusting the shooting angle to adapt to different lighting conditions and complex construction environments.

[0008] The sampling frequency of the humidity sensor and the temperature sensor is not less than once per second, and it can maintain the accuracy and stability of data acquisition in high humidity or high temperature environments.

[0009] The data conversion module normalizes the data using the min-max normalization formula and eliminates high-frequency noise in the sensor data through a denoising algorithm.

[0010] The multi-head attention analysis module includes at least one time series feature attention head and one spatial distribution feature attention head. The time series feature attention head analyzes the long-term data trend using a time window sliding mechanism, and the spatial distribution feature attention head optimizes the spatial accuracy of anomaly detection based on the sensor position relationship.

[0011] The anomaly detection module dynamically adjusts the threshold range according to the historical mean and standard deviation of humidity and temperature data, and uses an adaptive learning algorithm to update the threshold in real time according to environmental changes to improve the sensitivity of anomaly detection.

[0012] The trend analysis module performs multi-level smoothing processing on time series data by combining the moving average method and the exponential smoothing method to improve prediction accuracy, and generates a probability distribution of the future weak point state based on historical trend data.

[0013] The digital twin platform supports multi-dimensional data visualization functions, can label the risk levels of weak points with different colors or graphics, and supports real-time interactive queries and historical data playback.

[0014] The early warning module provides dynamic maintenance recommendations based on the crack propagation speed and the abnormal growth trend of humidity, and can recommend specific repair strategies and priorities through the intelligent reasoning module.

[0015] The maintenance recommendation module supports generating repair plans, including recommended repair times, applicable materials, and construction methods, and optimizes the economy and feasibility of the plan by comparing with historical repair records.

[0016] The beneficial effects of the present invention are: 1. The present invention makes full use of an image acquisition device (including a high-resolution infrared night vision camera) and a variety of sensors (humidity, temperature, pressure, etc.), covering all-round perception from visual detection (such as cracks, bulges, color changes, etc.) to environmental parameter acquisition (dynamic changes in temperature, humidity, pressure). It greatly improves the comprehensive control ability of construction quality and later use status. Through the high-resolution infrared night vision camera, clear image information can be obtained even under low-light or night construction conditions, realizing continuous monitoring of key construction details and potential hazards, and effectively reducing the phenomenon of missed inspections.

[0017] 2. The sampling frequency of sensor data can reach no less than once per second and is transmitted through low-power wireless communication methods such as LoRa or Bluetooth; the image data is transmitted at high speed using Wi-Fi, taking into account the balance between real-time performance and energy consumption. The data is encrypted to ensure that it is not easily tampered with or stolen during wireless transmission or cloud storage, enhancing the security of the system in complex construction environments.

[0018] 3. The convolutional neural network (CNN) is used to automatically identify cracks and locate weak waterproof points in the image data, and the Transformer multi-head attention mechanism is combined to perform spatio-temporal feature analysis on continuous sensor data. The fusion processing method of image + sensor data can better capture the potential risks of the building waterproof layer: it can not only identify surface defects (such as cracks, bulges), but also monitor the internal environment (such as abnormal changes in humidity and temperature), forming a monitoring closed-loop.

[0019] 4. Through preprocessing and deep learning means such as data standardization, min-max normalization, denoising algorithms, and multi-head attention analysis, noise can be effectively filtered and key features can be refined. With the help of dynamic threshold adjustment and adaptive learning algorithms, the anomaly detection module of the present invention can automatically update the threshold range when the construction environment changes or the seasons alternate, greatly reducing the situation of false alarms and missed detections. The combination of the moving average method and the exponential smoothing method for multi-level smoothing processing can more accurately predict long-term trends and short-term fluctuations, generating the probability distribution of future weak point states, helping engineering managers to timely formulate maintenance or prevention plans.

[0020] 5. Through the digital twin platform, the three-dimensional model of the building waterproof area can real-time map the sensor and image recognition results, thus realizing the visual superposition of multi-sensor readings and historical data playback. Different risk levels (such as crack expansion speed, humidity anomaly growth rate, etc.) will be marked with different colors or icons, allowing managers to quickly locate the problem area and conduct interactive queries. This visualization and interaction method is particularly efficient when the building scale is large or the number of sensors deployed is numerous, enabling management decisions to be "what you see is what you get".

[0021] 6. Based on the results of real-time risk analysis, the system makes a comprehensive judgment on the abnormal growth trends of crack propagation speed, humidity, or temperature, and gives maintenance suggestions at different levels to prevent small faults from evolving into major problems. The warning methods are flexible and diverse, such as text messages, platform pop-ups, email push, etc., which helps to notify relevant responsible persons in a timely manner and shorten the response time. The intelligent reasoning module can make intelligent recommendations and rankings for maintenance materials and construction methods based on expert rules, historical case libraries, or machine learning algorithms, and give priorities, reducing blind decisions and repeated construction, and significantly reducing maintenance costs.

[0022] 7. The maintenance suggestion module can call historical maintenance records, and comprehensively consider the performance of existing materials, construction period, environmental adaptability, and historical construction costs to automatically generate multiple sets of maintenance plans for comparison and selection. By comparing the costs, construction periods, and effect evaluations of different plans, managers can quickly find the optimal or most cost-effective plan, greatly improving the scientific nature and economic benefits of the maintenance process.

[0023] 8. Adopting the modular design concept, data interaction between functional modules is carried out through standardized interfaces. Users can freely add, subtract, or combine modules according to different scales of building types or different levels of monitoring requirements. In terms of networking environment, device adaptation, on-site layout, etc., the present invention also has strong flexibility and can be deployed in scenarios such as bridges, tunnels, subways, underground garages, and other scenarios that require anti-leakage monitoring, providing a wide range of application spaces for multiple fields such as civil engineering, construction, and maintenance.

[0024] 9. By establishing a complete "pre-warning - mid-term monitoring - post-maintenance" full-cycle management system, it is possible to comprehensively, dynamically, and traceably monitor and maintain the building waterproof system, greatly extending the service life of the waterproof layer. The management concept of preventing problems before they occur reduces the damage caused by leakage to the building structure and even the overall use function, and also effectively reduces the high costs or safety hazards brought by large-scale renovations in the later stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is the system architecture diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] Next, the technical solutions of the present invention will be clearly and completely described in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present invention.

[0027] It should be noted that the orientation concepts of "left", "right", "up", "down", "front", "back", "inside", and "outside" in the following solutions are all relative directions, and will not be listed one by one here.

[0028] Example 1 The Internet of Things-based building waterproof construction quality monitoring and early warning system provided in this example mainly includes four levels: a data acquisition layer, a data transmission layer, a data processing layer, and an application layer. The structural and functional relationships can be summarized as follows: Data acquisition layer: Image data and environmental parameters and physical states such as humidity, temperature, and pressure in the building waterproof construction area are obtained through image acquisition devices and various sensors. Data transmission layer: The collected data is securely and reliably transmitted to the data processing layer using wireless communication technology. Data processing layer: Data analysis and processing are carried out relying on an image recognition module and a Transformer architecture, including image recognition, data conversion, attention mechanism analysis, and anomaly detection and trend analysis modules. Application layer: A digital twin platform and an early warning and maintenance advice module are provided to realize the visual display of waterproof risks and intelligent early warning.

[0029] Data acquisition layer Image acquisition device: To better adapt to complex construction environments and different lighting conditions, a high-resolution infrared night vision camera is selected in this example, which has the function of dynamically adjusting the shooting angle. In this way, regardless of the weak light or night construction environment, the state changes and construction details of the waterproof layer can be captured in a timely manner. This device has a high resolution and can accurately obtain key image information such as fine cracks and material joints on the surface of the waterproof layer.

[0030] Sensor module, humidity sensor and temperature sensor: To effectively monitor the temperature and humidity changes near the waterproof layer, the sampling frequency is set to no less than once per second in this example. At the same time, the selected sensors can maintain the accuracy and stability of data acquisition in high humidity or high temperature environments, so as to capture more timely the factors that may lead to the deterioration of waterproof weak points. Pressure sensor: It is mainly used to monitor the pressure changes caused by external forces, wind pressure, etc. during or after construction, and can assist in judging whether there are potential hazards such as local deformation, bulging, and cracking.

[0031] Data transmission layer Wireless communication technology, Wi-Fi: It is mainly used to transmit large-capacity image data. When the network environment permits, images can be uploaded to the server for analysis in real time or regularly. LoRa or Bluetooth: It is mainly used for the transmission of sensor data. These protocols have characteristics such as low power consumption, long-distance transmission, or convenient networking, and can ensure stable and reliable data acquisition even in construction sites with a wide distribution of sensors and complex wiring. Data security and integrity: During the data transmission process, the system will perform encryption processing to ensure that the key data at the construction site is not tampered with or stolen. The encryption method can select symmetric or asymmetric encryption algorithms according to actual needs, or a higher level of protection can be achieved by combining security authentication means.

[0032] Data processing layer Image recognition module (based on CNN) The system inputs the real-time images obtained from the image acquisition device into a deep learning model (such as a convolutional neural network, CNN) for automatic recognition. The model can detect the weak points of building waterproofing (such as cracks, bulges, discolored areas, etc.) and assign a unique number according to the preset numbering rules. After analyzing the weak points, construction workers install the required sensors at the corresponding positions, so that the weak points can be accurately located in subsequent monitoring and their evolution trends can be analyzed.

[0033] Data conversion and analysis module Data conversion: First, standardize the continuous data from humidity, temperature, and pressure sensors; use the method of zero mean and unit variance to make subsequent analysis more stable.

[0034] Continue to normalize the data and map it to the interval ([0, 1]) or the corresponding interval, and use the discretization method to convert these numerical data into a Token sequence suitable for processing by the Transformer model. Each interval corresponds to a unique Token.

[0035] Multi-head attention analysis (based on Transformer): Time series feature analysis: Some attention heads focus on the trends of sensor readings such as temperature and humidity over time. For example, in the Token sequence of temperature, if the model finds an abnormal rapid increase or decrease, it may judge that there is a risk of attenuation of the performance of the waterproof material. Spatial distribution feature analysis: Another part of the attention heads compares the differences in humidity values in different regions. If the humidity in some positions is significantly higher than that in other positions, it can be preliminarily determined that there is a leakage risk in that region. Feature fusion and optimization: Synthesize the above analysis outputs into a deep feature vector through concatenation and linear transformation for use by the subsequent anomaly detection and trend analysis modules.

[0036] Anomaly detection and trend analysis module Anomaly detection: Compare the feature vector output by the multi-head attention with historical statistics (mean, standard deviation, etc.) to determine whether the current environmental and physical parameters deviate significantly from the normal state. Once the feature value in the Token sequence exceeds the set threshold, the system marks it as an anomaly and gives the specific location.

[0037] Trend analysis: The moving average method and exponential smoothing method are introduced to further process time series data for predicting the evolution trend of waterproof weak points. For example, if the temperature or humidity continues to rise in a certain area, the model will determine that the potential risk of leakage or material aging is increasing, and then recommend early preventive maintenance.

[0038] Application layer Digital twin platform, which is based on 3D modeling technology to virtually present the building waterproof area and realize real-time mapping of parameters such as humidity, temperature, and pressure on site. When the system detects an outlier at a certain location or monitors signs of cracks and leakage, it can mark the risk points on the 3D model to facilitate quick positioning and follow-up processing by engineering personnel.

[0039] Early warning and maintenance recommendation module. After the system completes data analysis, it will comprehensively evaluate the risk level, possible causes, and future evolution trend of waterproof weak points. For potential risk points, the system automatically generates predictive maintenance recommendations, including recommended repair time, materials, and methods, and notifies relevant personnel through text messages, platform pop-ups, and email push. For example, when the humidity shows an upward trend for several hours or days continuously and the temperature shows abnormal fluctuations, the system will give an emergency warning and recommend arranging a construction team for local inspection in advance to avoid larger-area leakage or damage.

[0040] Working process Data collection and transmission During the building waterproof construction or completion stage, multiple high-resolution infrared night vision cameras and humidity, temperature, and pressure sensors are arranged on site. The sensors collect environmental data at a frequency of no less than once per second and transmit digital signals to the data processing layer through LoRa or Bluetooth; the cameras upload images to the server in real-time / at regular intervals using Wi-Fi. The data undergoes encryption processing during transmission to ensure security and integrity.

[0041] Data analysis After the server receives the data, it first performs defect detection and weak point marking in the image recognition module. Next, the sensor data is input into the Transformer model through data conversion (standardization, normalization, discretization) for multi-head attention analysis of time series features and spatial distribution features. The analysis results are comprehensively processed in the anomaly detection and trend analysis module to obtain a judgment on whether there are anomalies (such as the appearance of abnormal humidity clusters or sharp temperature fluctuations) and the trend of risk evolution in the future period.

[0042] Result display and early warning The results obtained from the analysis will be displayed in real time on the digital twin platform, and 3D visual annotations will be made on the corresponding waterproof weak points. When the system detects an anomaly and the trend analysis indicates an increasing risk, the warning and maintenance advice module will push text messages, emails, or platform notifications to provide decision-making support for subsequent maintenance and repair work.

[0043] With a sensor data acquisition frequency of once per second, high-frequency dynamic monitoring of the waterproof area is achieved. Any significant changes in humidity, temperature, pressure, etc. can be captured immediately, greatly reducing the probability of missed detections and delayed discoveries.

[0044] The high-resolution infrared night vision camera enables the system to maintain high-quality image acquisition capabilities at night or in low-light conditions. Combined with the CNN algorithm, it can quickly detect extremely fine cracks or material defects to ensure the recognition accuracy.

[0045] Through the comprehensive analysis of temporal and spatial distribution features by the multi-head attention mechanism, abnormal changes or risk points can be quickly captured from a large amount of data, realizing the early prediction of leakage risks. Data encryption processing further enhances the security of the system and ensures the reliability of long-distance wireless transmission.

[0046] The system not only detects anomalies but also uses trend analysis models (moving average method, exponential smoothing method) to provide references for the subsequent risk evolution and gives targeted maintenance strategies. This predictive maintenance reduces the costs of blind maintenance and repeated construction, maximizing benefits.

[0047] The three-dimensional visualization and real-time interaction functions of the digital twin platform enable project managers, construction parties, and even property owners to observe the real state of the waterproof area through the network at any location. Combined with the warning system, text message, and email push functions, it is convenient for all parties to collaborate in handling problems and improves the overall management efficiency.

[0048] In summary, this embodiment realizes an integrated waterproof construction quality monitoring and warning process from perception - transmission - analysis - visualization - decision-making. Especially in complex construction environments and changing meteorological conditions, it can still maintain high-precision acquisition and high-efficiency processing of key data, greatly improving the guarantee level of building waterproof safety. Through all-round and intelligent monitoring and warning, it can help relevant personnel take maintenance measures more timely and accurately, extend the service life of the building waterproof layer and reduce maintenance costs, with significant practical value and promotion prospects.

[0049] Embodiment 2 Based on Embodiment 1, in this embodiment, the minimum-maximum normalization formula is further adopted to normalize the sensor data to ensure the comparability of data under different dimensions and magnitudes. Specifically, for the measured value (x) of a certain sensor over a period of time, its normalized result (x') can be defined as: Among them, (min(X)) and (max(X)) respectively represent the minimum and maximum values in the current data batch or historical statistical data. By mapping the sensor values to the interval ([0, 1]), it can make the subsequent multi-head attention analysis more robust when processing different sensor data.

[0050] Before data conversion in this embodiment, a denoising algorithm for high-frequency noise elimination is added. This algorithm can be one of methods such as wavelet transform denoising, Kalman filtering, or adaptive filtering. The specific selection can be flexibly configured in combination with the actual noise characteristics of the construction site.

[0051] By removing high-frequency noise and random pulse interference, the sensor data can be made smoother and more stable before entering the multi-head attention analysis link, effectively reducing the false alarm rate.

[0052] Improvement of the multi-head attention analysis module Time series feature attention head (time window sliding mechanism) In Embodiment 1, the multi-head attention analysis mainly focuses on data based on time series and spatial distribution features. In this embodiment, a time window sliding mechanism is further introduced for the time series feature attention head. Specifically, the system will automatically slide a set time window (such as 30 minutes, 1 hour, or other custom durations) in the historical records of sensor data, aggregate and extract features from the data within each window, so as to analyze the long-term data trend.

[0053] The advantage of this is that when the humidity or temperature at a certain place in the building shows a slow and continuous rise or fall (rather than a short-term violent fluctuation), the system can also timely capture long-term risks, such as potential problems like material aging or micro-cracks in construction joints.

[0054] Spatial distribution feature attention head (optimizing the anomaly detection accuracy based on positional relationships) In Embodiment 2, the spatial distribution feature attention head further considers the physical position relationship of each sensor at the construction site to optimize the spatial accuracy of anomaly detection. For example, the system uses the coordinate information of the sensors and cameras in the three-dimensional space to calculate the distance or enclosed area between adjacent sensors, and conducts a more accurate differential comparison of the humidity or temperature distribution. When there is a leakage in a local area or a large-scale humidity anomaly occurs, this attention head can quickly locate the area around the sensor most likely to have problems, thus providing a more accurate spatial reference for anomaly detection and subsequent maintenance.

[0055] Improvement of the Anomaly Detection Module Dynamically Adjusting the Threshold Range To improve the sensitivity of anomaly detection, in this embodiment, based on Embodiment 1, a method of dynamically adjusting the threshold according to the historical mean and standard deviation of humidity and temperature data is introduced. That is to say, the system will regularly or real-time calculate the mean and standard deviation of humidity and temperature of the sensors in the recent period (such as the past 24 hours or the past week). When environmental factors (such as seasons, outdoor climate) or construction conditions change, the system can adaptively update the anomaly threshold. Through this dynamic adjustment, it is possible to reduce "false positive" alarms caused by external climate factors or normal construction activities, and it can also avoid the lag response to sudden anomaly situations.

[0056] Adaptive Learning Algorithm In addition to dynamically adjusting the threshold range based on historical mean and standard deviation, this embodiment also performs real-time threshold updates through an adaptive learning algorithm (such as online learning, reinforcement learning, or autoregressive model, etc.). When the sensor data pattern shows a new distribution form with the progress of construction, material properties, or seasonal changes, the system can automatically update the judgment criteria, improving the robustness and adaptability of anomaly detection to environmental changes. For example, in the plum rain season or extremely high-temperature weather, the humidity and temperature benchmarks may be generally higher or fluctuate more. The adaptive learning algorithm can quickly correct these external factors and improve the accuracy of anomaly detection.

[0057] Working Process Data Preprocessing and Denoising After the temperature, humidity, pressure and other data collected by the sensors are transmitted to the data processing layer, they first enter the newly added denoising algorithm module in this embodiment. By filtering out high-frequency noise and random pulse interference, a smoother and more reliable data sequence is obtained. Subsequently, the min-max normalization is applied to uniformly scale the sensor data of each channel to the same dimension, facilitating the comprehensive analysis of multi-dimensional information by the Transformer model.

[0058] Multi-Head Attention and Time Window Sliding After the preprocessing is completed, the data is discretized into a token sequence suitable for the Transformer model, and the system creates at least two attention heads: Time series feature attention head: Adopting a time window sliding mechanism, it focuses on the changes in sensor readings over multiple time periods, especially the slow drift in long cycles and the sharp jumps in short periods. Spatial distribution feature attention head: Based on the analysis of the sensor position relationship, it detects anomalies in humidity or temperature gradients between different areas of the building. The outputs of these two attention heads are fused into a deep feature vector through concatenation and linear transformation, providing richer and more accurate context information for the subsequent anomaly detection module.

[0059] Dynamic threshold anomaly detection After receiving the deep feature vector, the anomaly detection module combines the historical mean, standard deviation, and the thresholds maintained by the system to update the model, and makes a judgment on whether each time window or each batch of data is abnormal. If it is detected that the current humidity has risen sharply above the average level in the past week and exceeds several times the standard deviation, the system will trigger an anomaly warning, and at the same time, it will make a comprehensive judgment based on real-time environmental factors (such as recent continuous rainfall, sudden drop or rise in external temperature) to avoid false alarms.

[0060] Adaptive learning and threshold update When the anomaly detection module repeatedly discovers a trend shift in the sensor data distribution (for example, the humidity continuously decreases due to the gradual drying of new materials), the system incorporates the latest data distribution into the online learning algorithm, continuously correcting the threshold range or model parameters. In this way, the system's identification of potential future anomalies will be more accurate, further reducing the situations of "missed alarms" and "false alarms".

[0061] Through the dual improvement of the min-max normalization and denoising algorithms, the system's ability to handle noisy data and dimensional differences is significantly enhanced, effectively reducing the sensitivity and misjudgment of the model to abnormal data.

[0062] The time window sliding mechanism enables the system to not only capture short-term abnormal fluctuations but also identify long-term, small-amplitude humidity or temperature changes, predicting potential signs of waterproof failure in advance.

[0063] Incorporating the sensor position relationship into the spatial distribution feature attention head makes the positioning of leakage risks more accurate, especially applicable to large building complexes or construction sites with a wide waterproof operation surface, which can shorten the inspection time and reduce the unnecessary disassembly and inspection scope.

[0064] In this embodiment, through the adaptive learning algorithm and dynamic threshold adjustment method, the system can automatically adjust the threshold with the change of the environment, avoiding excessive ineffective alarms. At the same time, for anomalies occurring at critical moments (such as the rainy season or high-temperature season), it can also issue warnings in a timely manner.

[0065] As the on-site data continues to accumulate, the system model will continue to iterate and improve, better adapting to the actual construction environment and climate characteristics, thus becoming more intelligent over time and playing a greater role in subsequent maintenance and early warning.

[0066] By accurately locating potential hidden dangers and precisely predicting long-term trends, construction units can take more targeted maintenance measures, reducing the risks of blind maintenance or large-scale rework.

[0067] In summary, based on Example 1, Example 2 greatly improves the adaptability, accuracy, and intelligence level of the present invention in actual building waterproof monitoring by introducing key technical means such as min-max normalization, denoising algorithms, time window sliding mechanisms, and adaptive learning algorithms. The technical improvements brought by this example can not only effectively reduce the false alarm and missed alarm rates but also provide more accurate data support for subsequent maintenance and management decisions, possessing significant economic value and application prospects for promotion.

[0068] Example 3 In Examples 1 and 2, it has been mentioned that the trend analysis of time series data mainly uses the moving average method and the simple exponential smoothing method. This example further emphasizes the combined application of the two methods and introduces a multi-level smoothing processing strategy: Moving Average: Conduct multiple sliding averages on the sensor readings to eliminate short-term random fluctuations, thereby reflecting medium- and long-term change trends. Exponential Smoothing: By assigning greater weights to newer data, the system can more sensitively capture recent changes, facilitating the rapid detection of sudden anomalies.

[0069] After combining the two, the system will respectively obtain a long-term trend curve and a short-term response curve, and then generate a comprehensive smoothing curve through a multi-level weighted fusion method. This can not only balance the reference degree of historical data but also retain the sensitivity to the latest abnormal fluctuations, significantly improving the accuracy and stability of prediction.

[0070] After obtaining the comprehensive smoothing curve, the system uses methods based on probability statistics or time series prediction models (such as ARIMA, LSTM, etc.) to generate the probability distribution of the future weak point states.

[0071] For example, for a weak point where the humidity shows a continuous upward trend, the model may give "the probability that the humidity exceeds a certain threshold within the next week is X%", so that relevant personnel can more intuitively understand the risk level and estimate the time window.

[0072] Enhancement of Digital Twin Platform Multi-Dimensional Data Visualization Function To better display the information obtained from sensors and image recognition, in this embodiment, multi-dimensional data visualization capabilities are introduced on the digital twin platform. The platform not only supports displaying building structures and sensor positions on a 3D model but also can overlay layer information such as humidity, temperature, pressure, and crack detection. Different weak points will be marked with different colors or graphics according to the risk level. For example: Green: Within the safe range Yellow: Mild warning Orange: Moderate warning Red: Severe warning, immediate maintenance required This intuitive visualization method enables construction workers, inspectors, and management to quickly locate anomalies or potential risks and make targeted decisions.

[0073] Real-Time Interactive Query and Historical Data Playback In this embodiment, the digital twin platform also provides a real-time interactive query function, supporting users to click on a certain sensor or area on the 3D model at any time to view its latest humidity, temperature, or image information. At the same time, historical data playback is supported, and the state changes of a certain weak point can be traced back along the time axis, allowing users to clearly understand the process evolution of leakage from small to large or cracks from thin to wide, providing sufficient basis for subsequent analysis.

[0074] Optimization of the Early Warning Module Dynamic maintenance suggestions based on crack propagation speed and abnormal humidity growth trend. Based on Embodiments 1 and 2, this embodiment further incorporates dynamic monitoring of crack propagation speed and abnormal humidity growth trend. For example: The system obtains the length or width information of cracks at different times from the image recognition module, compares the change amount between adjacent times to estimate the expansion speed. If the expansion speed exceeds the preset threshold, the system will determine that there are major potential safety hazards in the structure. At the same time, if the humidity sensor also shows a rapid increase in this area, it indicates that the crack may have caused a leakage channel. Based on these comprehensive indicators, the system will generate more targeted maintenance suggestions, such as: "The crack width has increased by X mm / day, and the humidity increase rate exceeds Y%. It is recommended to arrange maintenance in the near future." Intelligent Inference Module Recommends Specific Repair Strategies and Priorities Based on the dynamically maintained suggestions, the system is also equipped with an intelligent reasoning module that can perform rule reasoning or data mining by combining expert experience or based on a case library. This module can comprehensively consider various information such as crack size, location, material aging degree, environmental factors, etc., and then recommend specific repair strategies (such as whether local reinforcement is needed, replacement of waterproof materials, or re-spraying of waterproof coatings) and give corresponding priorities.

[0075] Expansion of the Maintenance Suggestion Module Automatic Generation of Repair Plans In this embodiment, the maintenance suggestion module realizes the function of automatically generating repair plans based on Embodiments 1 and 2: Recommended repair time: The system will give a relatively reasonable time period that does not affect the overall construction according to the currently monitored danger level and the construction progress of the building, avoiding secondary construction conflicts or excessive excavation. Applicable materials: Through comprehensive analysis of the material database and environmental data, materials more suitable for the local climate and the current characteristics of the waterproof layer are selected (such as special waterproof coatings, leak repair adhesives, sealing strips, etc.). Construction methods: According to the differences in crack morphology or leakage location, the system proposes corresponding construction processes, such as specific processes like local grinding, pasting rolls, and spraying waterproof coatings.

[0076] Compare with Historical Repair Records to Optimize the Economy and Feasibility of the Plan In this embodiment, the maintenance suggestion module will also retrieve the platform's historical repair records (including information such as the types of materials used in the past, construction personnel arrangements, repair cycles, and costs), and compare them with the currently recommended plan. If it is found that a certain material had poor effects in a similar environment in the past, or a certain construction method had high costs or long construction periods, the system will give optimization suggestions, thereby continuously improving the economy and feasibility of the plan. In this way, the management side can select the repair path with the greatest benefit, the shortest cycle, and the highest adaptability among many candidate plans, achieving scientific and efficient waterproof management.

[0077] Working Process Data Acquisition and Processing Similar to Embodiments 1 and 2, the system still uses high-resolution infrared night vision cameras and various sensors to capture the status of the building's waterproof layer in real time. After the data is transmitted and processed, it enters the enhanced trend analysis module for multi-level smoothing and probability distribution prediction.

[0078] Trend Analysis and Risk Judgment The system uses the moving average method and the exponential smoothing method to hierarchically process the data of the past and current time periods to obtain a stable trend curve, and combines statistical or deep learning methods to generate risk prediction values (such as the probability of leakage within the next week, the crack expansion speed, etc.).

[0079] Digital Twin and Multi-Dimensional Visualization The processed data is immediately mapped to the digital twin platform, and visual markings are given to the risk levels of weak points. Construction managers can understand the instantaneous changes of sensors through the real-time interaction interface and can also trace back the evolution of any time period through the historical data playback function.

[0080] Early warning module and dynamic maintenance suggestions When the crack propagation speed or the abnormal growth rate of humidity continuously exceeds the standard, the system automatically triggers an early warning and generates brief dynamic maintenance suggestions. If the risk further intensifies, the intelligent reasoning module will give a more detailed repair strategy, including the urgency level, material selection, and construction technology, etc.

[0081] Automatic generation and optimization of repair plans If the user confirms that repair measures need to be taken, the maintenance suggestion module will call the historical records and material database to automatically generate multiple sets of repair plans. The user can select a better plan according to the budget, construction period, or construction site conditions, or make secondary modifications and comparisons on the platform.

[0082] By combining the moving average method and the exponential smoothing method and generating the probability distribution of the future state of weak points, the system's time prediction and risk grading of potential risks are more accurate, reducing the blindness of construction management regarding uncertainties.

[0083] The color, graphic risk markings, and real-time interactive queries on the digital twin platform greatly improve the efficiency of on-site monitoring and problem location. At the same time, the historical data playback provides evidence support for tracing the cause of faults and determining responsibilities.

[0084] From the crack propagation speed, abnormal humidity growth to the leakage probability assessment, the early warning module of this embodiment can give hierarchical maintenance suggestions for different stages and different severities, enabling the construction party to better grasp the timing of maintenance and avoid small problems evolving into major failures.

[0085] The system combines historical maintenance records and material databases to automatically recommend maintenance plans that are both economical and feasible, avoiding repeated trial and error and cost waste, and significantly improving the scientific nature of construction management decisions.

[0086] Through accurate risk prediction, timely maintenance processing, and optimized suggestions from the intelligent reasoning module, the construction party can more effectively reduce the building damage caused by leakage, significantly extend the service life of the building waterproof layer, and reduce the cost of large-scale repairs in the later stage.

[0087] In summary, based on the foundation laid by the first two examples, Example 3 forms a more complete, accurate, and expandable building waterproof construction quality monitoring and early warning system by strengthening the trend prediction algorithm, digital twin visualization, and intelligent maintenance suggestion functions. Whether it is the identification of short-term sudden risks or the control of long-term trends, this example can provide more comprehensive technical guarantees and management support, helping relevant parties achieve better results in terms of project quality, cost control, and construction efficiency, and having great practical application value and promotion prospects.

[0088] The above are only the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications, and environments, and can be changed within the scope of the concept described herein through the above teachings or the technology or knowledge in related fields. As long as the changes and variations made by those skilled in the art do not depart from the spirit and scope of the present invention, they should all be within the protection scope of the appended claims of the present invention.

Claims

1. A building waterproof construction quality monitoring and early warning system based on the Internet of Things, characterized in that: include: The data acquisition layer is used to collect image data of the building waterproofing construction area through image acquisition equipment, and to monitor the environmental parameters and physical state changes of waterproofing weak points in real time through humidity sensors, temperature sensors and pressure sensors; The data transmission layer is used to transmit the collected data to the data processing layer using wireless communication technology, where image data is transmitted via Wi-Fi, and sensor data is transmitted via LoRa or Bluetooth, and the data is encrypted to ensure data security and integrity; Data processing layer, including: The image recognition module uses a deep learning algorithm based on a convolutional neural network to analyze image data, identify building waterproofing weak points, and uniquely number the weak points according to a preset numbering rule; Data conversion and analysis module: Data conversion: Standardize the continuous data collected by the sensor, calculate the zero mean and unit variance, and perform normalization to map the data to intervals. At the same time, use the discretization method to convert the continuous numerical data into a Token sequence suitable for Transformer model processing, where different intervals correspond to unique Tokens. Multi-head attention analysis: The multi-head attention mechanism of the Transformer model is used to analyze the token sequence. Each attention head analyzes the data from the dimensions of time series features and spatial distribution features: Time series feature analysis: Part of the attention is focused on the changes in the time dimension of sensor data; Spatial distribution feature analysis: Some attention heads analyze the token sequence of the humidity sensor, and by comparing the humidity values ​​in different areas of the building, determine the spatial distribution of abnormal humidity areas and locate possible leakage risks; Feature fusion and optimization: The output of the multi-head attention mechanism is concatenated and linearly transformed to generate a deep feature vector, providing optimization support for anomaly detection and trend analysis; Anomaly detection and trend analysis module: Anomaly detection: Calculate the mean and standard deviation of sensor feature values ​​based on historical data, set the threshold range, and when the Token sequence feature value exceeds the threshold range, the system marks it as an anomaly and locates the anomaly; Trend analysis: Use moving average and exponential smoothing methods to process time series data and predict the risk trend of weak points. For example, a continuous increase in humidity or temperature may indicate waterproof failure. Application layer, including: Digital twin platform, used to build a 3D model of the building’s waterproofing area, mapping the status and changes of the waterproofing system in real time; The early warning and maintenance recommendation module provides predictive maintenance recommendations based on data analysis results, including recommendations on maintenance time, materials, and methods, and issues early warnings via text messages, platform pop-ups, and email push notifications.

2. According to the Internet of Things-based building waterproof construction quality monitoring and early warning system according to claim 1, it is characterized by: The image acquisition device includes a high-resolution infrared night vision camera and has the function of dynamically adjusting the shooting angle to adapt to different lighting conditions and complex construction environments.

3. The building waterproof construction quality monitoring and early warning system based on the Internet of Things according to claim 1 is characterized by: The sampling frequency of the humidity sensor and the temperature sensor is not less than once per second, and can maintain the accuracy and stability of data collection in a high humidity or high temperature environment.

4. The building waterproof construction quality monitoring and early warning system based on the Internet of Things according to claim 1 is characterized by: The data conversion module uses a minimum-maximum normalization formula to normalize the data and eliminates high-frequency noise in the sensor data through a denoising algorithm.

5. The building waterproof construction quality monitoring and early warning system based on the Internet of Things according to claim 1 is characterized by: The multi-head attention analysis module includes at least one time series feature attention head and one spatial distribution feature attention head, wherein the time series feature attention head adopts a time window sliding mechanism to analyze long-term data trends, and the spatial distribution feature attention head optimizes the spatial accuracy of anomaly detection based on the sensor position relationship.

6. The building waterproof construction quality monitoring and early warning system based on the Internet of Things according to claim 1 is characterized by: The anomaly detection module dynamically adjusts the threshold range according to the historical mean and standard deviation of humidity and temperature data, and uses an adaptive learning algorithm to update the threshold in real time according to environmental changes, thereby improving the sensitivity of anomaly detection.

7. The building waterproof construction quality monitoring and early warning system based on the Internet of Things according to claim 1 is characterized by: The trend analysis module performs multi-level smoothing on the time series data by combining the moving average method and the exponential smoothing method to improve the prediction accuracy, and generates the probability distribution of the future weak point status based on the historical trend data.

8. The building waterproof construction quality monitoring and early warning system based on the Internet of Things according to claim 1 is characterized by: The digital twin platform supports multi-dimensional data visualization functions, can mark the risk levels of weak points with different colors or graphics, and supports real-time interactive query and historical data playback.

9. The building waterproof construction quality monitoring and early warning system based on the Internet of Things according to claim 1 is characterized by: The early warning module provides dynamic maintenance suggestions based on the crack expansion rate and abnormal humidity growth trend, and can recommend specific repair strategies and priorities through the intelligent reasoning module.

10. The building waterproof construction quality monitoring and early warning system based on the Internet of Things according to claim 1 is characterized by: The maintenance suggestion module supports the generation of maintenance plans, including recommended maintenance time, applicable materials and construction methods, and optimizes the economy and feasibility of the plan by comparing with historical maintenance records.

Images