Ai-driven long-term monitoring and analysis system for road maintenance technology

Abstract

The application provides an AI-driven long-term monitoring and analysis system for pavement maintenance technology, relates to the field of pavement maintenance monitoring, and comprises a multi-source perception monitoring terminal deployed at a key road location, an edge computing processing module, an AI intelligent analysis platform, an early warning pushing module and a maintenance management terminal. The multi-source perception monitoring terminal collects pavement crack images, environmental parameters and road load data in real time. The edge computing processing module pre-processes and preliminarily analyzes the collected data. The AI intelligent analysis platform accurately identifies and grades pavement cracks based on a deep learning algorithm, and constructs a crack development trend prediction model in combination with historical data and real-time data. The early warning pushing module pushes early warning information and maintenance suggestions to the maintenance management terminal. The maintenance management terminal receives early warning information, checks monitoring data and trend analysis results, and overall arranges maintenance work. The application improves the monitoring efficiency and accuracy of pavement cracks, realizes early detection, early monitoring and early repair, and reduces maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of transportation infrastructure maintenance technology; in particular, to the field of road maintenance monitoring technology; specifically, to an AI-driven long-term monitoring and analysis system for road maintenance technology. Background Technology

[0002] With the rapid development of my country's transportation infrastructure, the mileage of highways, bridges, and other transportation facilities is constantly increasing. Intersections and bridges, as transportation hubs, are key nodes for vehicle passage, and their road surface quality directly affects road safety and traffic efficiency. Over long-term use, road surfaces are susceptible to damage such as cracks, potholes, and subsidence due to various factors including vehicle loads, environmental factors (temperature changes, precipitation, ultraviolet radiation, etc.), and material aging. Cracks are the most common and earliest form of road surface damage. If cracks are not detected and treated in time, they will continue to develop over time, gradually evolving from small cracks into moderate to severe cracks, and even leading to road surface damage and collapse. This not only increases maintenance costs but also causes traffic accidents, threatening the lives and property of the public.

[0003] Currently, road maintenance monitoring technologies are mainly divided into two categories: traditional manual monitoring and intelligent monitoring technologies. Among them, traditional manual monitoring remains the most widely used method. Traditional manual monitoring relies on maintenance personnel carrying simple tools (such as measuring tapes and crack width gauges) to conduct regular inspections of the road surface, recording information such as the location, length, and width of cracks. Based on experience, they then assess the severity of the cracks and formulate maintenance plans. The advantages of this method are low equipment investment and simple operation. However, manual inspection has several drawbacks: First, it is inefficient, requiring significant manpower, resources, and time, especially for critical road conditions with large spans and wide distributions, making comprehensive and high-frequency monitoring difficult. Second, it lacks accuracy, as manual observation is heavily influenced by subjective factors, easily leading to missed or false detections of tiny cracks less than 0.5mm wide, and it cannot accurately measure characteristic parameters such as crack length and width. Third, it lacks timely warnings, as the long inspection cycle (usually 1-3 months) makes it impossible to monitor crack development in real time. When cracks rapidly develop or reach a point requiring repair, timely warnings are not issued, causing maintenance delays. Fourth, it cannot predict trends, as manual inspections only record the current state of cracks and cannot combine historical data and environmental factors to predict crack development trends, hindering preventative maintenance. For example, tiny cracks in critical locations such as bridge expansion joints and intersections are easily overlooked during manual inspections. These cracks, under the influence of vehicle loads and environmental factors, may rapidly develop in a short period, leading to road surface damage, increased maintenance costs, and safety hazards.

[0004] With the rapid development of technologies such as artificial intelligence, the Internet of Things, and big data, the application of intelligent monitoring technology in road maintenance is gradually increasing. Intelligent monitoring technologies mainly include image recognition-based monitoring technology, sensor-based monitoring technology, and drone-based monitoring technology. These technologies collect road surface data through automated equipment, combine it with algorithms for analysis, and achieve automatic identification and monitoring of cracks.

[0005] Image recognition-based monitoring technology is currently a research hotspot in the field of intelligent monitoring. Its principle involves acquiring road surface images using high-definition cameras and then employing image processing and machine learning algorithms to identify and extract features from cracks within the images. Existing technologies employ traditional image processing algorithms (such as edge detection and threshold segmentation) to identify cracks. However, these algorithms are poorly adapted to complex road surface backgrounds (such as road markings, debris, and shadows), resulting in low recognition accuracy, especially for small cracks. Some systems use deep learning algorithms (such as CNN and U-Net) for crack identification, improving accuracy. However, most of these systems only achieve crack identification and classification, lacking the ability to predict crack development trends, detect rapidly developing cracks in a timely manner, and cannot scientifically classify cracks according to industry standards, thus failing to meet the needs of "monitoring small cracks and providing early warning of rapid development." For example, the invention patent CN120668661A, entitled "Method, System, Device, UAV, Medium and Product for Monitoring Road Cracks", discloses a method to collect road images by camera and identify cracks using deep learning algorithms to achieve automatic crack monitoring. However, the system does not consider the prediction of crack development trends, nor does it combine environmental parameters and load data for comprehensive analysis, resulting in insufficient timeliness and pertinence of early warning.

[0006] Sensor-based monitoring technology primarily involves embedding sensors (such as strain sensors, pressure sensors, and temperature sensors) into the road surface to collect data on road stress, strain, and temperature, indirectly determining the presence and damage of cracks. The advantages of this technology are real-time data acquisition and high monitoring accuracy. However, its disadvantages include the difficulty and high cost of sensor installation, susceptibility to damage from vehicle loads, and short lifespan. Furthermore, it can only indirectly assess crack conditions, failing to directly reflect the morphology and specific location of cracks, thus hindering comprehensive monitoring of critical road conditions. For example, the invention patent CN113203690A, "A Crack Monitoring System and Method for Continuously Reinforced Concrete Pavement Based on OFDR+EMI," discloses a method of embedding fiber optic sensors into the road surface to collect road strain data and determine the existence and development of cracks. However, this system is complex to install, costly, and cannot accurately identify the specific characteristics of cracks, limiting its practicality.

[0007] Drone-based monitoring technology primarily uses drones equipped with high-definition cameras to conduct aerial inspections of road surfaces, collecting images and then using algorithms to identify cracks. The advantages of this technology are its wide inspection range and high efficiency, making it suitable for monitoring large areas of road. The disadvantages are that it is greatly affected by weather (e.g., it cannot fly in rainy or windy weather), and for complex road conditions such as intersections and bridges, the accuracy of drone inspections is insufficient, making high-frequency, routine monitoring impossible. Furthermore, it cannot acquire real-time environmental parameters and load data for the road surface, making it difficult to accurately predict crack development trends.

[0008] Furthermore, most existing intelligent monitoring systems suffer from low data processing efficiency, poor system compatibility, and incomplete maintenance management functions, and have not yet formed a mature and complete application system. For example, data collected by some systems needs to be transmitted to the cloud for centralized processing. When the data volume is large, transmission delays and lags are prone to occur, affecting the real-time performance of monitoring. Some systems cannot be integrated with existing maintenance management platforms, requiring maintenance personnel to switch between multiple systems, which is cumbersome. Although some systems can issue warnings, they cannot provide specific maintenance suggestions and emergency response plans, resulting in a lack of targeted maintenance work and an inability to achieve the goal of timely repair. Summary of the Invention

[0009] Therefore, the purpose of this invention is to propose an AI-driven long-term monitoring and analysis system for road maintenance technology. Installed at key road condition locations such as intersections and bridges, it automatically monitors and analyzes the development trend of road cracks. It monitors only minor cracks, and provides timely warnings when cracks develop rapidly or become severe enough to require repair, enabling timely repair. Integrating multi-source sensing, edge computing, and deep learning technologies, it achieves full-cycle, automated, and intelligent monitoring and analysis of road cracks, providing scientific and precise support for road maintenance work. This solves problems such as low efficiency in existing road crack monitoring, insufficient identification accuracy, untimely warnings, inability to predict crack development trends, and lack of targeted maintenance recommendations. It achieves comprehensive, automated, and routine monitoring of road cracks at key road conditions such as intersections and bridges without manual intervention, improving monitoring efficiency and reducing labor costs. It accurately identifies road cracks, especially minor cracks with a width of not less than 0.1 mm, reducing missed and false detection rates, with an identification accuracy of not less than 98%. The system enables precise monitoring of minute cracks; combining historical data, environmental parameters, and load data, it accurately predicts crack development trends and calculates crack growth rates. When the growth rate exceeds a preset threshold or the crack reaches a level requiring repair, it issues timely warnings, achieving early detection, early warning, and early repair. Strictly adhering to industry standard DB41 / T1469-2017, it scientifically classifies cracks and, based on crack level and development trends, pushes targeted maintenance suggestions and emergency response plans, improving the pertinence and effectiveness of maintenance work and reducing maintenance costs. It achieves real-time data acquisition, preprocessing, analysis, and storage, improving data processing efficiency and system stability, ensuring normal operation even in severe weather. Encrypted transmission and access control mechanisms ensure the security of monitoring data and the system, preventing data leakage and tampering, and strengthening system security. It improves maintenance management functions, achieving closed-loop management of monitoring-early warning-maintenance-feedback, enhancing the overall coordination and efficiency of maintenance work, and ensuring road traffic safety.

[0010] This invention provides an AI-driven long-term monitoring and analysis system for road maintenance technology, comprising: a multi-source sensing monitoring terminal, an edge computing processing module, an AI intelligent analysis platform, an early warning push module, and a maintenance management terminal. The modules interact with each other through a wireless communication network, forming a closed-loop process system of data acquisition, preprocessing, analysis, early warning, and maintenance. The multi-source sensing and monitoring terminal is deployed at key road condition locations such as intersections and bridges to collect real-time images of road surface cracks, environmental parameters, and road load data, providing basic data support for subsequent analysis. The edge computing processing module is connected to the multi-source sensing and monitoring terminal and is used to preprocess, filter noise and extract features from the collected data, screen valid data and make a preliminary judgment on whether cracks exist, reduce the amount of data transmitted to the AI ​​intelligent analysis platform and improve data processing efficiency. The AI ​​intelligent analysis platform is connected to the edge computing processing module and is used to perform in-depth analysis on the pre-processed effective data, including accurate crack identification, crack level classification, and crack development trend prediction. It is the key analysis unit of the system. The early warning push module is connected to the AI ​​intelligent analysis platform and is used to push early warning information and corresponding maintenance suggestions to the maintenance management terminal according to the analysis results of the AI ​​intelligent analysis platform and the preset early warning level, so as to ensure that the early warning information is delivered in a timely manner. The maintenance management terminal is used to receive early warning information, view monitoring data and trend analysis reports, realize the initiation, tracking and closed-loop management of maintenance tasks, and provide operational support for maintenance personnel.

[0011] The multi-source sensing and monitoring terminal adopts an integrated design, is compact in size, and is easy to install. It can be fixed to traffic light poles at intersections, bridge railings, etc. with expansion bolts, or it can be embedded in the road surface to meet the installation requirements of different critical road conditions. The terminal is waterproof, dustproof, impact-resistant, and low-temperature resistant, with a protection level of not less than IP67. It can operate stably in environments ranging from -40℃ to 85℃, adapting to the application needs of different climate regions.

[0012] Most existing road surface monitoring systems use a single sensor to collect data, resulting in limited monitoring dimensions and an inability to comprehensively reflect the actual condition of road surface cracks. The multi-source sensing monitoring terminal of this invention includes a high-definition camera unit, an environmental sensing unit, a load monitoring unit, and a data transmission unit. These units work collaboratively to achieve real-time acquisition of multi-dimensional data. The high-definition camera unit uses an industrial-grade high-definition camera with the lens facing the road surface. The lens focal length is adjustable (range 8-12mm) and supports autofocus for capturing real-time images of the road surface. The camera resolution is no less than 1080P, and the frame rate is adjustable from 1 to 30 frames per second. Under normal operating conditions, it uses a frame rate of 10 frames per second, which is automatically increased to 20 frames per second at night or in inclement weather to ensure image clarity. It supports low-light compensation technology and has a built-in infrared fill light, which can achieve clear imaging within a range of 5-10 meters in the absence of light at night, avoiding missed detections at night. The camera uses a wide-angle lens with a shooting angle of 120°, which can cover the entire road surface area of ​​a single monitoring point, reducing monitoring blind spots.

[0013] The environmental sensing unit includes a temperature sensor, a humidity sensor, a precipitation sensor, and a wind speed sensor, used to collect environmental parameters of the monitoring area and provide environmental data support for predicting crack development trends. Specifically, the temperature sensor uses a PT1000 platinum resistance thermometer with a measurement range of -40℃ to 85℃ and an accuracy of ±0.5℃, used to collect temperature changes in the road surface and surrounding environment and analyze the impact of temperature changes on crack development. The humidity sensor uses a capacitive humidity sensor with a measurement range of 0 to 100%RH and an accuracy of ±3%RH, used to collect air humidity and analyze the impact of humidity on road surface materials. The precipitation sensor uses a tipping bucket rain gauge with a measurement range of 0 to 4 mm / min and an accuracy of ±0.1 mm, used to collect precipitation and analyze the erosive effect of rainwater on cracks. The wind speed sensor uses a cup-type anemometer with a measurement range of 0 to 60 m / s and an accuracy of ±0.3 m / s, used to collect wind speed and analyze the impact of strong winds on the road surface.

[0014] The load monitoring unit employs a piezoelectric pressure sensor embedded 5-10 cm below the road surface, closely adhering to the road material. It collects load data from passing vehicles to analyze the impact of vehicle load on crack development. The sensor has a measurement range of 0-200 kN, an accuracy of ±1 kN, and a response time of no more than 10 ms, enabling real-time capture of dynamic changes in vehicle load. The sensor possesses excellent anti-interference capabilities, effectively preventing the influence of road vibration and temperature changes on the measurement results. Two to four load sensors are deployed at each monitoring point, evenly distributed across the road surface, ensuring the representativeness of the collected load data.

[0015] The data transmission unit adopts a 5G / 4G+LoRa dual-mode communication method, supports encrypted data transmission, and ensures data security and real-time performance. 5G / 4G communication is used for real-time data transmission under normal operating conditions, with a transmission rate of no less than 10Mbps and a latency of no more than 50ms. LoRa communication is used for emergency transmission when 5G / 4G signals are interrupted, supports data caching, and has a cache capacity of no less than 16GB. When the network is restored, the cached data is automatically synchronized to the edge computing processing module. The data transmission unit has a built-in SIM card slot, supports full network compatibility, can be adapted to the networks of different operators, and also supports WiFi backup communication to further improve the reliability of data transmission.

[0016] In addition, the multi-source sensing and monitoring terminal also includes a power supply unit, which adopts a dual power supply method of solar power and lithium battery backup; the solar panel power is not less than 20W, which can power the terminal and charge the lithium battery under sunlight conditions; the lithium battery capacity is not less than 10000mAh, which can ensure that the terminal can work continuously for more than 72 hours under no sunlight conditions, avoiding monitoring interruption due to power failure.

[0017] The edge computing processing module adopts an embedded design and is integrated into the multi-source sensing and monitoring terminal or deployed nearby at the monitoring point. It interacts with the multi-source sensing and monitoring terminal via a wired connection (USB, Ethernet) to exchange data. It is mainly used to perform real-time preprocessing of the collected data, reduce invalid data transmission, improve the overall processing efficiency of the system, and avoid excessive processing pressure on the cloud.

[0018] The edge computing processing module includes a data preprocessing unit, a preliminary identification unit, and a data caching unit. The functions of each unit are as follows: The data preprocessing unit is used to perform targeted preprocessing on various types of data collected by the multi-source sensing and monitoring terminal to ensure the accuracy and validity of the data: ① Image data preprocessing: The road surface images acquired by the high-definition camera unit are subjected to denoising, enhancement, and size normalization. The denoising process uses a Gaussian filtering algorithm to remove noise points in the image (such as road debris, dust, shadows, etc.) and improve image clarity. The enhancement process uses a histogram equalization algorithm to enhance the image contrast and highlight the features of crack areas for easier subsequent identification. The size normalization process adjusts the image size to a uniform 1024×768 pixels to ensure consistency in subsequent algorithm processing. At the same time, the image is converted to grayscale to reduce the amount of data and improve processing speed.

[0019] ② Environmental parameter preprocessing: Outlier removal, unit unification, and standardization are performed on the temperature, humidity, precipitation, and wind speed data collected by the environmental sensing unit. Outlier removal adopts the 3σ criterion to remove data that exceeds the normal range (such as temperature exceeding 85℃ or below -40℃, precipitation exceeding 4mm / min, etc.) to avoid abnormal data affecting the analysis results. Unit unification converts all environmental parameters to international standard units (temperature: ℃, humidity: %RH, precipitation: mm, wind speed: m / s). Standardization uses the min-max standardization algorithm to map the data to the [0,1] interval for easy subsequent algorithm analysis.

[0020] ③ Load data preprocessing: The vehicle load data collected by the load monitoring unit is filtered, peak value extracted, and statistically processed. The filtering process uses the Kalman filter algorithm to remove vibration noise from the load data and extract the true load peak value. Peak value extraction is used to extract the maximum load value of each vehicle. Statistical processing is used to calculate the average load, maximum load, and load frequency per unit time (such as 1 hour or 1 day), providing data support for crack development trend prediction.

[0021] The preliminary identification unit employs a lightweight CNN algorithm (MobileNetV3) to perform preliminary identification on the preprocessed grayscale road image to determine whether cracks exist. This algorithm is small in size and fast in operation, and can run in real time on the edge computing module, with a recognition speed of no less than 10 frames per second. The preliminary identification unit sets a crack identification threshold (e.g., pixel percentage ≥ 0.01%). When the pixel percentage of the crack area in the image exceeds the threshold, it is determined that a crack exists, and the image and corresponding environmental and load data are marked as valid data and transmitted to the AI ​​intelligent analysis platform. When the pixel percentage is lower than the threshold, it is determined that there is no crack, and the image and data are marked as invalid data, stored only in the local cache, and not transmitted, thereby reducing the amount of data transmission and reducing network pressure.

[0022] The data caching unit adopts a dual caching mechanism of local SD card + cloud backup. The SD card has a storage capacity of no less than 128GB and is used to cache the raw data (images, environment, payload data) and preprocessed data of the past 7 days, which facilitates subsequent data traceability and anomaly investigation. The cloud backup synchronizes key data (valid data, system operation logs, etc.) to the database of the AI ​​intelligent analysis platform through a wireless communication network to achieve dual data backup and prevent data loss. The data caching unit supports an automatic cleanup function. When the SD card storage reaches 90%, the oldest invalid data is automatically deleted to ensure sufficient cache space.

[0023] The AI ​​intelligent analysis platform is a key analysis unit of the system. Deployed on a cloud server, it adopts a distributed architecture and supports multi-node parallel processing. It can process effective data from multiple monitoring points simultaneously, improving analysis efficiency. The platform has powerful computing and data storage capabilities, supporting the processing of no less than 100 frames of image data per second. The data storage capacity can be flexibly expanded according to the number of monitoring points (supporting TB-level storage).

[0024] The AI ​​intelligent analysis platform includes a crack precision identification unit, a crack classification unit, a trend prediction unit, and a historical data management unit. These units work together to achieve accurate crack identification, classification, and development trend prediction. While some existing monitoring systems can automatically identify cracks, they lack accurate prediction of crack development trends. They can only issue warnings after cracks have progressed to a certain extent, failing to meet the needs for monitoring minute cracks, providing early warnings of rapid development, and timely repair. Furthermore, existing monitoring systems have weak data analysis capabilities, unable to scientifically classify cracks according to industry standards or provide precise maintenance recommendations, resulting in a lack of targeted maintenance and high maintenance costs. The crack precision identification unit of this invention uses an improved U-Net deep learning algorithm to accurately identify cracks in valid images transmitted from the edge computing processing module, extracting feature parameters such as crack length, width, area, and direction, thus achieving precise monitoring of minute cracks.

[0025] The traditional U-Net algorithm is easily affected by road background interference when identifying tiny cracks, resulting in insufficient recognition accuracy. This invention improves upon it by adding an attention mechanism module and a residual connection module. The specific improvements are as follows: ① Attention Mechanism Module: The module employs a channel attention mechanism (SE-Net) to assign channel weights to the feature maps extracted by the network, highlighting the feature channels in the crack region and suppressing interfering channels in the background region, thereby improving the recognition accuracy of small cracks. This module obtains channel feature vectors by performing global average pooling on each channel of the feature map, and then generates channel weight coefficients through a fully connected layer and a sigmoid activation function. The weight coefficients are multiplied with the original feature map to obtain an enhanced feature map, making the network pay more attention to the crack region.

[0026] ② Residual Connection Module: A residual connection is added between the encoder and decoder of the U-Net network to fuse the shallow features extracted by the encoder with the deep features of the decoder, solving the gradient vanishing problem in the training process of deep networks, accelerating the model training speed, and improving the model's generalization ability; at the same time, the residual connection can retain more detailed features, which helps to accurately identify tiny cracks.

[0027] The training process of the improved U-Net algorithm of this invention is as follows: ① Dataset Construction: Collect road surface images of different road conditions (intersections, bridges), different weather conditions (sunny, rainy, cloudy, night), and different crack levels, collecting no less than 100,000 images in total. Label the images (label the location, length, width, and other information of the cracks) to construct a training dataset. Divide the dataset into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0028] ② Model training: The Adam optimizer was used with a learning rate of 0.001, a batch size of 32, and 100 training epochs. The cross-entropy loss function was used in combination with the Dice loss function to address the data imbalance problem (few samples with small gaps). An early stopping strategy was adopted during training. Training was stopped when the loss function of the validation set did not decrease for 10 consecutive epochs to avoid overfitting of the model.

[0029] ③ Model Validation and Optimization: The trained model is validated using a test set. Model parameters (such as attention mechanism weights, residual connection coefficients, etc.) are adjusted to ensure that the model’s recognition accuracy is not less than 98%, that it can identify tiny cracks with a width of not less than 0.1 mm, and that the false negative rate is not higher than 1% and the false positive rate is not higher than 1%.

[0030] The workflow of the crack precision identification unit is as follows: it receives valid images transmitted by the edge computing processing module, processes the images using the improved U-Net algorithm to identify crack regions, then separates the cracks from the background using image segmentation technology, and finally extracts feature parameters such as crack length (unit: mm), width (unit: mm), area (unit: mm²), and orientation (e.g., longitudinal, transverse, diagonal) through pixel calculation and geometric analysis. The feature parameters are then associated and stored with the corresponding environmental parameters and load data to support subsequent crack classification and trend prediction.

[0031] According to the industry standard (DB41 / T1469-2017 "Technical Specification for Preventive Maintenance of Asphalt Pavement of Ordinary Highways"), the severity of pavement cracks is divided into different levels, and different levels of cracks require different maintenance measures. Most existing monitoring systems do not strictly adhere to this standard for crack classification, resulting in a lack of scientific basis for formulating maintenance measures and failing to achieve the goal of preventive maintenance. The crack classification unit of this invention, based on crack feature parameters extracted by the crack precision identification unit and combined with the industry standard DB41 / T1469-2017 "Technical Specification for Preventive Maintenance of Asphalt Pavement of Ordinary Highways," classifies pavement cracks into four levels. Different levels of cracks correspond to different monitoring frequencies and maintenance measures. The specific classification standards are as follows: ① Minor cracks (Level 1 cracks): Crack width ≤ 2mm, crack length ≤ 500mm, crack area ≤ 1000mm², no loose material, no deformation, clear crack direction, no branch cracks or only a few branch cracks; This level of crack has little impact on the pavement structure, only requires routine monitoring, and does not require immediate repair.

[0032] ② Minor cracks (secondary cracks): Crack width 2-5mm, crack length 500-1000mm, crack area 1000-5000mm², with slight loosening or a few branch cracks, and no obvious deformation; cracks of this level have a slow development trend, and it is necessary to increase the monitoring frequency, pay close attention to the development of cracks, and take simple preventive maintenance measures when necessary.

[0033] ③ Moderate cracks (Level III cracks): Crack width 5-10mm, crack length 1000-2000mm, crack area 5000-10000mm², with obvious scattering or deformation, and many branch cracks; this level of cracks develops rapidly and has a certain impact on the pavement structure. It is necessary to issue early warnings in a timely manner and take targeted maintenance measures for repair.

[0034] ④ Severe cracks (Level IV cracks): Crack width ≥ 10mm, crack length ≥ 2000mm, crack area ≥ 10000mm², severe scattering, obvious deformation, dense branch joints, and potential hazards such as road surface damage and collapse; This level of cracks has a significant impact on road structure and traffic safety, and an emergency warning should be issued immediately and emergency maintenance measures should be taken to prevent the cracks from deteriorating further.

[0035] The crack classification unit also has an adaptive adjustment function, which can fine-tune the parameters of the classification standard according to the climate conditions and road conditions of different regions (such as the difference between bridge pavement and ordinary intersection pavement) to ensure the scientific nature and pertinence of the classification. At the same time, the crack level is associated with the corresponding monitoring frequency: micro cracks are monitored once every 24 hours, light cracks are monitored once every 12 hours, moderate cracks are monitored once every 6 hours, and severe cracks are monitored once every 1 hour, so as to realize differentiated monitoring of the classification.

[0036] (3) Trend prediction unit: Based on the LSTM neural network algorithm, combined with historical crack data, real-time environmental parameters and load data, a crack development trend prediction model is constructed to achieve accurate prediction of crack development speed and future state, and to realize early warning of rapid development.

[0037] The LSTM neural network algorithm has powerful time series data processing capabilities, which can capture the time series characteristics of crack development and accurately predict the future trend of crack changes by combining the influence of environmental parameters and load data. Based on the LSTM algorithm, this invention introduces an attention mechanism to focus on factors that have a significant impact on crack development (such as temperature changes, precipitation, vehicle load, etc.) to improve prediction accuracy.

[0038] The process of constructing a trend prediction model is as follows: ① Input data selection: Select historical characteristic parameters of cracks (width, length, area), historical environmental parameters (temperature, humidity, precipitation, wind speed), and historical load data (average load, maximum load, load frequency) as input data. The time span is the past 3-6 months, and the data sampling interval is 1 hour.

[0039] ② Data preprocessing: Standardize the input data and map it to the [0,1] interval to eliminate the influence of dimensions; use linear interpolation to fill in missing data to ensure data integrity; remove outlier data to avoid affecting the model's prediction accuracy.

[0040] ③ Model Construction: Construct an LSTM neural network model, including an input layer, hidden layers, attention layers, and an output layer. The number of neurons in the input layer is equal to the dimension of the input data (e.g., 7 dimensions such as crack width, length, temperature, humidity, precipitation, wind speed, and average load, with 7 neurons). There are 3 hidden layers, each with 64 neurons, using the tanh activation function. The attention layer uses a multi-head attention mechanism to assign weights to the feature vectors output by the hidden layers, highlighting key influencing factors. The number of neurons in the output layer is equal to the dimension of the prediction target (e.g., predicting crack width and length for the next month, with 2 neurons), using a linear activation function.

[0041] ④ Model Training: The Adam optimizer was used with a learning rate of 0.0001, a batch size of 64, and 200 training epochs. The mean squared error (MSE) loss function was used to measure the deviation between the predicted and actual values. Regularization (L2 regularization) and Dropout (dropout rate = 0.2) were used during training to prevent overfitting. The model parameters were adjusted using the validation set to ensure that the prediction error of the model did not exceed 10%.

[0042] The trend prediction unit works as follows: it receives current crack characteristic parameters, environmental parameters, and load data transmitted by the crack precision identification unit in real time, combines them with historical data stored in the historical data management unit, and inputs them into the trained trend prediction model to predict the crack width, length, and area change trends for the next 1-6 months, and calculates the crack development rate (unit: mm / month); it compares the prediction results with the preset development rate threshold (0.5 mm / month), and triggers an early warning mechanism when the crack development rate exceeds the threshold; at the same time, it judges the future grade changes of the crack based on the prediction results and issues an early warning to buy time for maintenance work.

[0043] The historical data management unit is used to store all monitoring data, analysis results, and maintenance records, and to build a pavement maintenance database to provide data support for predicting crack development trends and formulating maintenance plans. The database adopts a hybrid storage architecture of MySQL and MongoDB. MySQL is used to store structured data (such as crack level, environmental parameters, load data, early warning information, maintenance records, etc.), and MongoDB is used to store unstructured data (such as pavement images, videos, etc.), taking into account both the efficiency and flexibility of data storage.

[0044] The functions of the historical data management unit include: ① Data storage: Stores raw data collected by multi-source sensing and monitoring terminals, data preprocessed by edge computing processing modules, analysis results from AI intelligent analysis platforms (crack characteristic parameters, crack levels, trend prediction results, etc.), early warning information from early warning push modules, and maintenance records from maintenance management terminals, etc. The data storage period is no less than 5 years, and permanent data backup is supported.

[0045] ② Data Query: Supports multi-condition queries. Maintenance personnel can query corresponding monitoring data and analysis results based on monitoring point location, time range, crack level, warning level, etc.; supports fuzzy query to improve query efficiency; query results can be exported to Excel, PDF and other formats for easy data statistics and analysis.

[0046] ③ Data statistics: Automatically perform statistical analysis on monitoring data and generate various statistical reports, such as reports on the distribution of crack levels at monitoring points, statistical reports on crack development rates, statistical reports on environmental parameters, statistical reports on load data, statistical reports on maintenance records, etc., to provide decision support for managers.

[0047] ④ Data Update: Receive data transmitted from each module in real time and automatically update the data in the database to ensure data timeliness and accuracy; at the same time, regularly optimize the database (such as index optimization, data compression, etc.) to improve data query and processing efficiency.

[0048] The early warning push module is connected to the AI ​​intelligent analysis platform and is used to push early warning information and corresponding maintenance suggestions to the maintenance management terminal according to the analysis results (crack level, development rate, trend prediction results) of the AI ​​intelligent analysis platform and according to the preset early warning level, so as to ensure that the early warning information is delivered in a timely manner and achieve early warning and early repair.

[0049] The early warning push module includes an early warning level determination unit, an information generation unit, and a multi-channel push unit. The functions of each unit are as follows: The early warning level determination unit classifies the early warning level into four levels based on the crack level and crack development rate, combined with trend prediction results. Different early warning levels correspond to different treatment priorities and maintenance requirements. The specific early warning level classification is as follows: ① Level 1 Early Warning (Monitoring Level): Corresponds to micro-cracks (Level 1 cracks), with a crack development rate ≤ 0.2 mm / month, and trend predictions indicate that the crack level will not be upgraded within the next 3 months; this level only conducts routine monitoring, does not push early warning information, only records monitoring data in the background of the maintenance management terminal, and generates monitoring reports periodically.

[0050] ② Level 2 Warning (Reminder Level): This level corresponds to two situations: one is a minor crack (Level 1 crack), with a crack development rate of 0.2-0.5 mm / month; the other is a mild crack (Level 2 crack), with a crack development rate ≤0.3 mm / month. Trend predictions indicate that the crack level may upgrade to mild or moderate cracks within the next 3 months. This level sends reminder-type warning messages to remind maintenance personnel to pay close attention to the crack development and increase the monitoring frequency.

[0051] ③ Level 3 Warning (Emergency Reminder): This level corresponds to two situations: one is a minor crack (Level 2 crack), with a crack development rate of 0.3-0.5 mm / month; the other is a moderate crack (Level 3 crack), with a crack development rate of ≤0.4 mm / month. Trend predictions indicate that the crack level may escalate to moderate or severe cracks within the next 1-2 months. This level sends an emergency reminder warning and preliminary maintenance recommendations, requiring maintenance personnel to conduct on-site verification of the cracks as soon as possible and formulate a maintenance plan.

[0052] ④ Level 4 Warning (Emergency Warning): This level corresponds to two situations: one is moderate cracks (Level 3 cracks), with a crack development rate > 0.4 mm / month; the other is severe cracks (Level 4 cracks), regardless of the development rate. Trend forecasts indicate that the cracks may further deteriorate within the next month, posing safety hazards such as road surface damage and collapse. This level sends out emergency warning information, detailed maintenance plans, and emergency response suggestions, requiring maintenance personnel to immediately take emergency maintenance measures to prevent safety accidents.

[0053] The early warning level determination unit has a dynamic adjustment function, which can fine-tune the determination criteria of the early warning level based on the feedback of maintenance personnel and the actual maintenance effect, so as to ensure the accuracy and pertinence of the early warning; at the same time, when the crack is repaired, the corresponding early warning will be automatically lifted and normal monitoring will be restored.

[0054] The information generation unit generates standardized early warning information based on the early warning level and the analysis results of the AI ​​intelligent analysis platform. The early warning information includes the following: ①Basic information: Location of monitoring points (accurate to specific intersections and bridge sections, with latitude and longitude coordinates), warning time, warning level, and crack level; ② Crack information: length, width, area, direction, and development rate of the crack, accompanied by crack images (with crack location and characteristic parameters labeled); ③ Environmental and load information: Environmental parameters (temperature, humidity, precipitation, wind speed) and load data (average load, maximum load) during the monitoring period are analyzed to assess the impact of environment and load on crack development. ④ Trend prediction information: Predicted crack development trends for the next 1-6 months, and forecasted changes in crack grade; ⑤ Maintenance recommendations: Based on the crack level and warning level, provide targeted maintenance recommendations, including maintenance methods, maintenance materials, maintenance procedures, and precautions; for level four warnings, emergency response plans should also be provided, including temporary traffic diversion recommendations and a list of emergency maintenance materials and equipment.

[0055] The information generation unit supports a custom template function, allowing managers to customize the template and content of early warning information according to the maintenance needs of different regions and road conditions, thereby improving the practicality of the early warning information.

[0056] The multi-channel push unit supports four push methods: SMS, APP push, WeChat official account push, and background message reminder. Push priorities can be set according to the maintenance personnel's job positions to ensure timely delivery of early warning information. The four push methods are as follows: ① SMS push: Send early warning SMS to the mobile phones of maintenance personnel, including the early warning level, monitoring point location, brief information on cracks and maintenance suggestions. It is applicable to emergency early warning (level 3 and level 4 early warning) to ensure that maintenance personnel receive the early warning information as soon as possible.

[0057] ②APP push: Push early warning information to the mobile APP of the maintenance management terminal, including complete early warning content, crack images, trend prediction results and maintenance suggestions. Maintenance personnel can directly view detailed information and report maintenance progress through the APP.

[0058] ③ WeChat Official Account Push: Push early warning information to the maintenance management team's WeChat official account to facilitate team members to view and coordinate responses. This method is applicable to non-emergency early warnings (Level 2 early warning).

[0059] ④ Background message reminders: In the PC management backend of the maintenance management terminal, a warning message reminder will pop up, and unprocessed warnings will be marked to facilitate managers to view and assign handling tasks. For warnings that are not processed in time (such as level 3 warnings that have not been processed for more than 24 hours or level 4 warnings that have not been processed for more than 6 hours), a second push will be automatically triggered to ensure that the warnings are taken seriously.

[0060] In addition, the early warning push module also has an early warning record query function, which can record the push time, push method, recipient, processing status, and processing result of all early warning information, which is convenient for subsequent traceability and assessment. At the same time, it supports early warning statistical analysis, automatically counts the number of early warnings at different monitoring points, different early warning levels, and different time periods, providing data support for optimizing maintenance work.

[0061] The maintenance management terminal serves as the system's operating terminal, providing a convenient operating interface and comprehensive maintenance management functions for frontline maintenance personnel and managers. It enables closed-loop management of "monitoring-early warning-maintenance-feedback," thereby improving the efficiency and standardization of maintenance work.

[0062] The maintenance management terminal includes a mobile app and a PC-based management backend, with data synchronized in real time between the two to adapt to different usage scenarios. Specific functions are as follows: (1) Mobile App: Supports Android and iOS systems, for use by frontline maintenance personnel, and has the following functions: ① Warning Reception and Viewing: Receives various warning information pushed by the warning push module in real time, and can view complete warning content, crack images, trend prediction results and maintenance suggestions; supports warning information marking (such as "viewed", "handled", "pending verification"), which makes it easy for maintenance personnel to track the progress of handling.

[0063] ② Monitoring data viewing: You can view real-time monitoring data (road surface images, environmental parameters, load data), historical monitoring data and analysis results of the monitoring points you are responsible for. It supports filtering and viewing by time range, crack level and other conditions, which is convenient for comparison and analysis during on-site verification.

[0064] ③ Maintenance Task Management: Receive maintenance tasks (including routine maintenance tasks and emergency maintenance tasks) assigned by management personnel, view task details (such as maintenance location, maintenance requirements, completion deadline, and maintenance materials); support maintenance progress reporting (such as "not started", "in progress", "completed"), and upload maintenance site photos, videos, and text descriptions to provide feedback on maintenance results.

[0065] ④ On-site verification and reporting: In response to the early warning information, maintenance personnel can initiate on-site verification through the APP, fill in the verification results (such as the difference between the actual situation of the cracks and the system monitoring results, the development and changes of the cracks), and upload on-site photos to provide support for system algorithm optimization and early warning level adjustment.

[0066] ⑤ Personal Center: Supports functions such as account and password login, password modification, and permission viewing; allows users to view the monitoring points they are responsible for, unprocessed warnings, and maintenance tasks, facilitating personal work coordination.

[0067] (2) PC-based management backend: Deployed on a server in the maintenance management center for use by management personnel, and has the following functions: ① Data visualization: Using charts (line charts, bar charts, heat maps, pie charts, etc.), it intuitively displays information such as the distribution of crack levels, crack development rate, changes in environmental parameters, and load data statistics at each monitoring point; it supports comparison and viewing of data from multiple monitoring points, and the location of monitoring points can be located through a map to view the real-time status of the monitoring points, making it easy for managers to quickly grasp the overall road conditions.

[0068] ② Early Warning Management: Users can view early warning information for all monitoring points, including early warning level, early warning time, crack condition, and handling status; support early warning information filtering (by monitoring point, early warning level, and time range) and export; manually adjust early warning levels and cancel early warnings; and manually remind relevant maintenance personnel of early warnings that have not been handled in a timely manner.

[0069] ③Maintenance task allocation: Based on early warning information and actual road conditions, initiate maintenance tasks and assign them to the corresponding front-line maintenance personnel, clarifying task requirements, completion deadlines, and responsible persons; track the progress of maintenance tasks, view maintenance feedback results, and supervise tasks that are not completed on time.

[0070] ④ Maintenance Record Management: Stores all relevant records for maintenance tasks, including maintenance time, maintenance location, maintenance personnel, maintenance measures, maintenance materials, and maintenance results; supports multi-condition querying and exporting of maintenance records, and can generate maintenance statistical reports to provide a basis for maintenance work evaluation and optimization.

[0071] ⑤ System Management: It has multi-user permission management function, which can assign different operation permissions according to the position (management personnel, front-line maintenance personnel, system administrator). For example, management personnel have the right to operate all functions, while front-line maintenance personnel only have the right to view warnings, receive tasks, and report progress. It supports adding, deleting, and modifying user accounts, and can view user operation logs to ensure that system operations are traceable. It supports system parameter settings (such as warning level threshold, data collection frequency, cache time, etc.), which can be adjusted according to actual needs.

[0072] ⑥ Report generation: Generate various statistical reports, including monitoring point operation status reports, crack level distribution reports, early warning statistics reports, maintenance task completion reports, load data statistics reports, etc.; reports support custom settings (such as report period, statistical dimensions), and can be exported to Excel, PDF and other formats for easy reporting and archiving by management personnel.

[0073] The mobile app and PC-based management backend work together to adapt to frontline maintenance and management scenarios, improving maintenance efficiency.

[0074] To ensure the accuracy of data collected by the multi-source sensing and monitoring terminal and to avoid measurement errors caused by sensor or camera aging or environmental influences, the system also includes a calibration module. This module is used to periodically calibrate various sensors and high-definition cameras of the multi-source sensing and monitoring terminal to ensure the system's monitoring accuracy.

[0075] The calibration module includes a standard reference and a calibration algorithm, and is connected to a multi-source sensing and monitoring terminal. The specific calibration process and functions are as follows: (1) Calibration cycle: The calibration cycle can be set to 1 month. The system will automatically remind the administrator to perform calibration operations. After severe environments (such as heavy rain, strong sandstorms, and high temperature exposure), calibration can be initiated manually to ensure the accuracy of data collection.

[0076] (2) Calibration content and methods: ① High-definition camera unit calibration: A standard grayscale plate and a standard-sized reference object (such as a crack simulation plate with a known width) are used as standard reference objects. The reference object is placed in a standard position in the monitoring area. The high-definition camera unit acquires images of the reference object. The calibration algorithm compares the acquired images with the parameters of the standard reference object and automatically adjusts the camera's focal length, exposure, white balance and other parameters to eliminate image distortion and sharpness deviation, ensuring the accuracy of crack image acquisition.

[0077] ② Environmental sensing unit calibration: For temperature sensors, humidity sensors, precipitation sensors, and wind speed sensors, corresponding standard testing equipment (such as standard thermometers, standard hygrometers, standard rain gauges, and standard anemometers) are used as standard references. The sensors and standard testing equipment are placed in the same environment, and data is collected for the same time period. The calibration algorithm calculates the deviation between the two and automatically adjusts the measurement parameters of the sensors to ensure the accuracy of environmental parameter acquisition.

[0078] ③ Load monitoring unit calibration: A standard weight block (with known weight and corresponding load value) is used as a standard reference. The weight block is placed in the monitoring area of ​​the load sensor. The sensor's measured value is collected and compared with the load value corresponding to the standard weight block. The calibration algorithm calculates the deviation and automatically adjusts the sensor's sensitivity to ensure the accuracy of load data acquisition.

[0079] (3) Calibration record and feedback: After calibration, the calibration module automatically records the calibration time, calibration object, parameters before and after calibration, calibration deviation and other information, and stores them in the historical data management unit for easy traceability and verification. If the sensor or camera deviation is found to be too large during the calibration process (exceeding the allowable range, such as temperature sensor deviation > ±1℃), the system will automatically issue a calibration abnormality warning to remind the management personnel to check the equipment and repair or replace it if necessary.

[0080] To address emergencies such as severe cracks and rapidly developing cracks, the system also includes an emergency response module connected to the early warning push module. When a level four early warning is triggered, the emergency response process is automatically initiated to ensure that cracks are repaired in a timely manner and to reduce safety hazards.

[0081] The main functions of the emergency response module are as follows: (1) Emergency task generation: When a level 4 warning is triggered, the emergency response module generates an emergency maintenance task based on the warning information (crack location, crack level, development trend), clarifies the emergency response requirements, completion time limit, emergency maintenance materials and equipment list, and associates the nearest maintenance personnel (based on the geographical location and on-duty status of the maintenance personnel) to assign the task to the corresponding personnel.

[0082] (2) Emergency response plan push: Based on the crack level and actual road conditions, generate targeted emergency response plans, including temporary traffic diversion suggestions (such as setting up warning signs, guiding vehicles to detour, and closing some lanes), emergency maintenance methods (such as temporary filling, crack sealing, etc.), emergency maintenance material list (such as sealant, filling material, warning signs, etc.), and emergency equipment list (such as road roller, cutting machine, transport vehicle, etc.), and push them to the maintenance personnel's mobile APP and PC management backend.

[0083] (3) Real-time tracking of emergency progress: After receiving an emergency task, maintenance personnel can report the emergency response progress in real time through the mobile APP (such as "departed", "arrived at the scene", "in progress", "completed") and upload photos and videos of the on-site response; management personnel can view the emergency response progress in real time through the PC management backend, supervise the response process, and manually remind maintenance personnel to speed up the progress if the response progress is lagging behind.

[0084] (4) Feedback and evaluation of emergency response results: After the emergency response is completed, the maintenance personnel report the response results, including the crack repair status, the materials and equipment used, and the response effect. The emergency response module combines the system monitoring data to conduct a preliminary evaluation of the response effect (such as whether the crack stopped developing after repair and whether the road surface smoothness met the standards). The evaluation results are stored in the historical data management unit to provide support for the optimization of subsequent emergency response plans.

[0085] (5) Emergency resource dispatch: If a single maintenance worker is unable to complete the emergency response task, the emergency response module supports sending support requests to surrounding maintenance workers and dispatching nearby emergency resources (such as maintenance equipment and materials) to ensure the efficient progress of emergency response work; at the same time, it can link with traffic management departments to push temporary traffic diversion information in a synchronized manner to jointly ensure road traffic safety.

[0086] The various modules of the system interact with each other via a wireless communication network. To ensure the real-time performance, security, and reliability of data transmission, an encrypted transmission protocol and multiple security protection mechanisms are adopted, as detailed below: (1) Wireless communication network architecture: The communication architecture of 5G / 4G+LoRa+WiFi is adopted. 5G / 4G communication is used first to realize real-time data transmission. LoRa communication is used for emergency buffer transmission when 5G / 4G signal is interrupted. WiFi communication is used as a supplementary backup to ensure uninterrupted data transmission. The communication network supports multi-node concurrent transmission and can transmit images, environment, load and other data from multiple monitoring points at the same time. The transmission delay does not exceed 50ms, which meets the real-time monitoring requirements.

[0087] (2) Encrypted data transmission: The AES-256 encryption algorithm is used to encrypt all transmitted data (images, environmental parameters, load data, early warning information, maintenance records, etc.) throughout the process to prevent data from being stolen or tampered with during transmission; at the same time, the HTTPS protocol is used for data transmission to establish a secure transmission channel and ensure the integrity and security of data transmission.

[0088] (3) Identity authentication and access control: The system adopts a dual identity authentication method of account password + dynamic verification code. When logging in, in addition to entering the account password, users also need to receive a dynamic verification code (via SMS, APP push, etc.). Only after successful verification can they log in to the system to prevent account theft. At the same time, different operation permissions are divided according to the user's position to restrict the user's access to system data and functions. For example, front-line maintenance personnel can only view the data of the monitoring points they are responsible for and receive related tasks. They cannot modify system parameters or view sensitive data of other monitoring points to ensure system security.

[0089] (4) Secure data storage: The historical data management unit adopts a hybrid storage architecture of MySQL+MongoDB, which supports both local storage and cloud backup. Data storage adopts encrypted storage method, and sensitive data (such as monitoring point location and maintenance personnel information) is encrypted. The database is backed up regularly (daily incremental backup and monthly full backup). The backup data is stored on different servers to prevent data loss. At the same time, a data security audit mechanism is established to record all data access, modification and deletion operations, which facilitates subsequent tracing and investigation of security risks.

[0090] (5) System anti-interference capability: The multi-source sensing and monitoring terminal adopts a shielded design, which can effectively resist electromagnetic interference and radio frequency interference, and ensure that data can still be collected normally in complex environments (such as traffic lights at intersections and near high-voltage lines); the edge computing processing module and AI intelligent analysis platform adopt a redundant design, and when some nodes fail, they can automatically switch to backup nodes to ensure the normal operation of the system and improve the stability and anti-interference capability of the system.

[0091] Preferably, the multi-source sensing monitoring terminal of the present invention adopts an integrated design, which is convenient to install and can be flexibly deployed in various key road conditions such as intersections and bridges; it has good environmental adaptability, high protection level, and can operate stably in different climate regions; the system parameters can be finely adjusted according to the road conditions and climate conditions of different regions to adapt to the application needs of different scenarios.

[0092] The AI-driven long-term monitoring and analysis system for road maintenance technology of this invention realizes a closed-loop workflow of data collection, preprocessing, analysis, early warning, maintenance, and feedback, including the following steps: Step 1: Data Acquisition: Multi-source sensing and monitoring terminals deployed at key road condition locations such as intersections and bridges collect real-time images of road surface cracks, environmental parameters (temperature, humidity, precipitation, wind speed), and road load data through high-definition camera units, environmental sensing units, and load monitoring units. The data transmission unit adopts 5G / 4G+LoRa dual-mode communication to transmit the collected raw data to the edge computing processing module. The power supply unit adopts dual power supply of solar energy and lithium batteries to ensure uninterrupted data acquisition.

[0093] Step 2: Data Preprocessing and Preliminary Identification: After receiving the raw data, the edge computing processing module performs targeted preprocessing (denoising, enhancement, outlier removal, standardization, etc.) on the image data, environmental parameters, and load data through the data preprocessing unit to ensure data accuracy; the preliminary identification unit uses a lightweight CNN algorithm (MobileNetV3) to perform preliminary identification on the preprocessed road surface image, determine whether cracks exist, filter valid data, and remove invalid data without cracks; the data caching unit performs dual caching of the raw data and preprocessed data to ensure that no data is lost.

[0094] Step 3: In-depth analysis and trend prediction: The edge computing processing module transmits the filtered valid data to the AI ​​intelligent analysis platform; the crack precision identification unit uses the improved U-Net algorithm to accurately identify cracks in valid images and extract crack feature parameters; the crack classification unit classifies cracks according to industry standards; the trend prediction unit uses the LSTM neural network algorithm, combined with historical data, real-time environmental parameters, and load data, to build a trend prediction model, predict crack development trends, and calculate crack development rates; the historical data management unit stores and manages all data and analysis results.

[0095] Step 4: Early Warning Judgment and Push: The early warning push module receives the analysis results from the AI ​​intelligent analysis platform. The early warning level judgment unit determines the early warning level based on the crack level and development rate. The information generation unit generates early warning information that includes crack information, trend prediction, and maintenance suggestions. The multi-channel push unit pushes the early warning information to the maintenance management terminal through SMS, APP, WeChat official account, background messages, etc., to ensure timely delivery.

[0096] Step 5: Maintenance Management and Emergency Response: Maintenance personnel receive early warning information and maintenance tasks through the maintenance management terminal (mobile APP, PC backend). Frontline maintenance personnel go to the site to verify and carry out maintenance work, and report the maintenance progress and results. Management personnel coordinate and allocate maintenance tasks, track progress, and view maintenance records through the PC backend. When a level 4 early warning is triggered, the emergency response module automatically starts the emergency process, generates emergency tasks and response plans, dispatches emergency resources, and tracks the response progress in real time to ensure timely repair of cracks.

[0097] Step 6: Calibration and Optimization: The calibration module regularly calibrates the sensors and cameras of the multi-source sensing monitoring terminal every month to ensure data acquisition accuracy. Based on maintenance feedback results, calibration data, and historical monitoring data, the system continuously optimizes the AI ​​algorithm (improved U-Net algorithm, LSTM algorithm) and early warning level judgment criteria to improve the system's monitoring accuracy and early warning accuracy. At the same time, through historical data statistical analysis, it provides decision support for maintenance work optimization and realizes continuous system optimization.

[0098] This workflow is applicable to long-term monitoring, trend analysis, and accurate early warning of pavement cracks in key road conditions such as intersections and bridges, providing intelligent support for pavement maintenance.

[0099] Traditional manual inspection methods struggle to cover all critical road conditions and have long inspection cycles, making it impossible to monitor the dynamic changes in pavement cracks in real time. Existing intelligent monitoring systems are either limited by environmental factors (such as drones) or have limited monitoring ranges (such as single sensors), failing to achieve comprehensive, high-frequency, and routine monitoring of critical road conditions such as intersections and bridges. The AI-driven long-term monitoring and analysis system for road maintenance technology provided by this invention enables comprehensive and high-frequency monitoring, improving monitoring efficiency.

[0100] Existing image recognition-based systems have low accuracy in identifying minute cracks (≤2mm in width), and are prone to missed detections and false detections. Sensor-based systems cannot directly identify the shape and specific location of cracks, and can only make indirect judgments, which have limited accuracy. This invention improves crack identification accuracy, reduces the missed detection and false detection rates, and meets the need for precise monitoring of minute cracks.

[0101] Most existing systems can only identify the current state of cracks and cannot combine historical data, environmental parameters (temperature, precipitation, etc.), and load data to predict the development trend of cracks, resulting in delays in maintenance. This invention can accurately predict the development trend of cracks and issue timely warnings when cracks develop rapidly or reach a point where repair is needed, enabling early warning and early repair.

[0102] Existing systems do not strictly adhere to industry standards for crack classification, or the classification standards are unclear, making it impossible to formulate corresponding maintenance measures based on crack levels. This invention, by scientifically and systematically classifying crack levels, can provide targeted maintenance suggestions and accurately push these suggestions to maintenance personnel. Maintenance personnel can quickly grasp appropriate maintenance methods, carry out targeted maintenance work, and reduce maintenance costs.

[0103] Existing systems often use centralized cloud-based data processing, which can lead to transmission delays and buffering issues when dealing with large datasets, impacting real-time monitoring. This invention partially employs edge computing to preprocess the collected data in real time, reducing invalid data, lowering transmission pressure and processing costs, and improving data processing efficiency. Furthermore, this invention enhances its anti-interference capabilities, preventing data acquisition anomalies and system crashes in adverse weather conditions (such as heavy rain, blizzards, and strong winds), demonstrating good system stability.

[0104] Most existing systems only have monitoring and early warning functions, lacking maintenance management functions such as maintenance task allocation, progress tracking, and maintenance record query. This invention realizes closed-loop management of monitoring-early warning-maintenance-feedback, which is conducive to the overall planning of maintenance work and improves maintenance management efficiency.

[0105] Existing systems lack effective encryption measures for data transmission and storage, making them susceptible to data theft and tampering. This invention addresses this by implementing a robust access control mechanism, strictly defining user permissions to prevent data leakage and improve the authenticity and reliability of monitoring data.

[0106] Compared with the prior art, the beneficial effects of the present invention are as follows: The AI-driven long-term monitoring and analysis system for road maintenance technology provided by this invention employs a multi-source sensing monitoring terminal to achieve real-time acquisition of road crack images, environmental parameters, and load data without manual intervention. This enables automated and intelligent monitoring of road cracks throughout their entire lifecycle, solving the problems of low efficiency and high rates of missed or false detections associated with traditional manual inspections, significantly improving monitoring efficiency. The system combines edge computing with AI intelligent analysis to achieve real-time data preprocessing and deep analysis, improving data processing efficiency and enabling comprehensive, high-frequency, routine monitoring of key road conditions such as intersections and bridges, reducing labor costs. It also utilizes an improved U-Net deep... The algorithm incorporates an attention mechanism and residual connection module to effectively suppress road surface background interference and improve the accuracy of micro-crack identification. It achieves high accuracy in identifying micro-cracks with a width of at least 0.1 mm, with an accuracy rate of at least 98% and a false positive and false negative rate of no more than 1%, meeting the needs of routine micro-crack monitoring and achieving the goal of "early detection" and precise monitoring. Based on the LSTM neural network algorithm, combined with historical data, environmental parameters, and load data, a crack development trend prediction model is constructed. This model can accurately predict crack changes over the next 1-6 months, calculate crack development rates, and accurately predict crack propagation. Crack development trends are monitored; a four-level early warning system is implemented based on crack grade and development rate, providing targeted maintenance recommendations and emergency response plans. The timely and targeted warnings enable early detection and repair, preventing further crack deterioration and reducing maintenance costs. Strictly adhering to industry standard DB41 / T1469-2017, cracks are scientifically classified into four levels, and differentiated monitoring frequencies and maintenance measures are developed for different levels based on crack development trends. The early warning information includes detailed and accurate maintenance recommendations, material lists, and procedures, ensuring targeted maintenance, achieving preventative maintenance, and reducing pavement damage and safety hazards. It adopts an edge computing + cloud distributed architecture to improve data processing efficiency and avoid excessive pressure on the cloud; multi-source communication backup, dual power supply, and redundant design ensure that the system can still operate normally in harsh weather and complex environments; it adopts security mechanisms such as AES-256 encryption algorithm, dual authentication, and access control to ensure the security of data transmission and storage and prevent data leakage and tampering; it integrates functions such as maintenance task allocation, progress tracking, maintenance record query, and report generation to realize a closed-loop process management of monitoring-early warning-maintenance-feedback, which improves the overall coordination and standardization of maintenance work and has broad prospects for promotion and application. Attached Figure Description

[0107] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0108] In the attached diagram: Figure 1 This is a block diagram showing the module composition of the AI-driven long-term monitoring and analysis system for road maintenance technology of this invention. Detailed Implementation

[0109] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of systems and products consistent with some aspects of this disclosure as detailed in the appended claims.

[0110] The terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. The singular forms “a,” “the,” and “the” as used in this disclosure and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0111] It should be understood that although the terms first, second, third, etc., may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0112] The embodiments of the present invention will be described in further detail below.

[0113] Example

[0114] This invention provides an AI-driven long-term monitoring and analysis system for road maintenance technology, such as... Figure 1The system includes: a multi-source sensing monitoring terminal, an edge computing processing module, an AI intelligent analysis platform, an early warning push module, and a maintenance management terminal. These modules interact via a wireless communication network. The multi-source sensing monitoring terminal is deployed at key road condition locations such as intersections and bridges to collect real-time images of pavement cracks, environmental parameters, and road load data. The edge computing processing module connects to the multi-source sensing monitoring terminal and performs preprocessing, noise filtering, and feature extraction on the collected data to screen valid data and preliminarily determine the existence of cracks. The AI ​​intelligent analysis platform communicates with the edge computing processing module and performs in-depth analysis of the preprocessed valid data, including crack identification, crack level classification, and crack development trend prediction. The early warning push module connects to the AI ​​intelligent analysis platform and pushes early warning information and corresponding maintenance suggestions to the maintenance management terminal according to preset early warning levels based on the analysis results of the AI ​​intelligent analysis platform. The maintenance management terminal receives early warning information, views monitoring data and trend analysis reports, and enables the initiation, tracking, and closed-loop management of maintenance tasks.

[0115] The multi-source sensing monitoring terminal includes a high-definition camera unit, an environmental sensing unit, a load monitoring unit, and a data transmission unit. The high-definition camera unit uses an industrial-grade high-definition camera with the lens facing the road surface to acquire real-time images of the road surface. The resolution is no less than 1080P, the frame rate is adjustable from 1 to 30 frames per second, and it supports low-light compensation, making it suitable for nighttime and inclement weather monitoring. The environmental sensing unit includes a temperature sensor, a humidity sensor, a precipitation sensor, and a wind speed sensor to collect data on ambient temperature, air humidity, precipitation, and wind speed in the monitored area. The temperature sensor has a measurement range of [missing information - likely related to temperature measurement]. The temperature range is -40℃ to 85℃, with an accuracy of ±0.5℃. The humidity sensor has a measurement range of 0 to 100%RH and an accuracy of ±3%RH. The load monitoring unit uses a piezoelectric pressure sensor embedded 5-10cm below the road surface to collect load data from passing vehicles, with a measurement range of 0 to 200kN and an accuracy of ±1kN. The data transmission unit uses 5G / 4G+LoRa dual-mode communication, supports encrypted data transmission, and ensures data security and real-time performance. When the 5G / 4G signal is interrupted, it switches to LoRa mode for data caching and delayed transmission.

[0116] The edge computing processing module includes a data preprocessing unit, a preliminary identification unit, and a data caching unit. The data preprocessing unit is used to denoise, enhance, and normalize the size of image data collected by multi-source sensing and monitoring terminals, and to remove outliers, unify units, and standardize environmental parameters and load data. The preliminary identification unit uses a lightweight CNN algorithm to perform preliminary identification on the preprocessed road surface images, determine whether cracks exist, and remove invalid images without cracks to reduce the amount of data transmitted to the AI ​​intelligent analysis platform. The data caching unit adopts a dual caching mechanism of local SD card + cloud backup. The SD card storage (capacity 128GB) is used to cache the raw data and preprocessed data of the past 7 days. When the network is restored, the cached data is automatically synchronized to the AI ​​intelligent analysis platform.

[0117] The AI ​​intelligent analysis platform includes a crack precision identification unit, a crack classification unit, a trend prediction unit, and a historical data management unit. The crack precision identification unit uses an improved U-Net deep learning algorithm to identify cracks in valid images transmitted from the edge computing processing module, extracting feature parameters such as crack length, width, area, and direction. The accuracy rate is no less than 98%, and it can identify micro-cracks with a width of no less than 0.1 mm. The crack classification unit, based on crack feature parameters and industry standard DB41 / T1469-2017, classifies pavement cracks into micro-cracks (width ≤ 2 mm, no loose material, no deformation) and minor cracks (width 2-5 mm, with slight loose material or a small amount of material). The road surface is classified into four levels: severe cracks (5-10mm wide, with obvious loosening or deformation), moderate cracks (width ≥10mm, with severe loosening and obvious deformation), and severe cracks (width ≥10mm, with severe loosening and obvious deformation). The trend prediction unit is based on the LSTM neural network algorithm, combined with historical crack data, real-time environmental parameters and load data, to build a crack development trend prediction model, predict the crack width and length change trend in the next 1-6 months, calculate the crack development rate, and trigger an early warning mechanism when the development rate exceeds a preset threshold (0.5mm / month). The historical data management unit is used to store all monitoring data, analysis results and maintenance records, build a road maintenance database, support data query, statistics and export, and the data storage period is not less than 5 years.

[0118] The early warning push module includes an early warning level determination unit, an information generation unit, and a multi-channel push unit. The early warning level determination unit classifies the early warning level into four levels based on the crack's severity and development rate: Level 1 (minor cracks, development rate ≤0.2mm / month), only routine monitoring is performed, and no early warning information is pushed; Level 2 (minor cracks, development rate 0.2-0.5mm / month or mild cracks, development rate ≤0.3mm / month), reminder-type early warning information is pushed; Level 3 (mild cracks, development rate 0.3-0.5mm / month or moderate cracks, development rate ≤0.4mm / month)... The system can push emergency alerts and preliminary maintenance suggestions; for Level 4 alerts (moderate cracks with a development rate > 0.4 mm / month or severe cracks, regardless of the development rate), it pushes emergency alerts, detailed maintenance plans, and emergency response suggestions; the information generation unit generates alert information including crack location, crack level, development trend, alert level, and maintenance suggestions based on the alert level and analysis results; the multi-channel push unit supports four push methods: SMS, APP push, WeChat official account push, and background message reminders, and the push priority can be set according to the maintenance personnel's positions to ensure that the alert information is delivered in a timely manner.

[0119] The maintenance management terminal includes a mobile APP and a PC management backend. The mobile APP supports Android and iOS systems and is used by front-line maintenance personnel. It can receive early warning information, view pavement crack images and monitoring data, report maintenance progress, and provide feedback on maintenance results. The PC management backend is used by management personnel and has functions such as data visualization, trend analysis, early warning management, maintenance task allocation, maintenance record query, and report generation. It supports multi-user permission management and can assign different operation permissions (viewing permission, operation permission, and management permission) according to the job position.

[0120] The improved U-Net deep learning algorithm adds an attention mechanism module and a residual connection module to the traditional U-Net algorithm. The attention mechanism module is used to highlight the features of the crack area, suppress background interference, and improve the recognition accuracy of small cracks. The residual connection module is used to solve the gradient vanishing problem in the training process of deep networks, accelerate the model training speed, and improve the model's generalization ability. The LSTM neural network algorithm, by introducing an attention mechanism, focuses on environmental parameters (such as temperature and precipitation) and load data that have a great impact on crack development, improving the accuracy of trend prediction with a prediction error of no more than 10%.

[0121] The calibration module is used to periodically calibrate the sensors and high-definition cameras of the multi-source sensing monitoring terminal. The calibration cycle is one month to ensure the accuracy of the collected data. The calibration module includes a standard reference and a calibration algorithm. By comparing the data collected by the monitoring terminal with the parameters of the standard reference, the parameter settings of the monitoring terminal are adjusted to eliminate measurement errors.

[0122] The wireless communication network adopts an encrypted transmission protocol, which includes three layers: data encryption, identity authentication, and access control. Data encryption uses the AES-256 encryption algorithm to encrypt the transmitted data and prevent it from being stolen or tampered with. Identity authentication uses a dual authentication method of account password + dynamic verification code to ensure that only authorized users can access the system. Access control restricts the scope of user access to system data and functions based on user permissions to ensure system security.

[0123] The emergency response module is connected to the early warning push module. When a level 4 early warning is triggered, the emergency response process is automatically started, an emergency maintenance task is generated and assigned to the nearest maintenance personnel, and an emergency response plan is pushed, including temporary traffic diversion suggestions, a list of emergency maintenance materials and equipment, and the progress of emergency maintenance is tracked in real time to ensure that cracks are repaired in a timely manner and reduce safety hazards.

[0124] Application examples

[0125] In practical applications such as monitoring at intersections of main urban roads and approach bridges of cross-river bridges, the system of this invention is deployed to achieve long-term monitoring, trend analysis, and accurate early warning of road surface cracks. Specific implementation details are as follows: This embodiment deploys a total of 12 multi-source sensing and monitoring terminals, including 6 at intersections of main urban roads and 6 on the approach bridge sections of cross-river bridges. The specific deployment locations and requirements are as follows: (1) Deployment location: One of the four traffic light poles at the intersection is deployed, and two are deployed in the central median strip of the intersection to ensure coverage of the entire intersection's motor vehicle lanes, non-motor vehicle lanes and pedestrian crossings; one is deployed every 50 meters on the approach road section of the cross-river bridge, installed on the outside of the bridge railing, with the lens facing the bridge surface, covering both lanes, and focusing on monitoring key locations that are prone to cracks, such as bridge expansion joints and bridge surface joints.

[0126] (2) Installation requirements: The monitoring terminal is fixed with expansion bolts and the installation height is 3.5-4.5 meters to ensure that the high-definition camera unit can clearly capture the road surface without any blind spots; the load monitoring unit is embedded 8cm below the surface of the road surface and is closely attached to the asphalt layer of the road surface. Three load sensors are evenly deployed at each monitoring point, located on the left, middle and right sides of the lane respectively, to ensure that the collected load data is representative; the environmental sensing unit is exposed on the terminal housing to avoid obstruction and ensure the accuracy of environmental parameter collection.

[0127] Equipment parameter selection includes: ① High-definition camera unit: It adopts an industrial-grade high-definition camera with a resolution of 1080P and a frame rate of 1-30 frames / second. The lens focal length is 10mm, the wide angle is 120°, the built-in infrared fill light is provided, the low light compensation range is 0-50 lux, the protection level is IP67, and it is suitable for harsh outdoor environments.

[0128] ② Environmental sensing unit: The temperature sensor is a PT1000 platinum resistance sensor with a measurement range of -40℃ to 85℃ and an accuracy of ±0.5℃; the humidity sensor is a capacitive sensor with a measurement range of 0 to 100%RH and an accuracy of ±3%RH; the precipitation sensor is a tipping bucket rain gauge with a measurement range of 0 to 4mm / min and an accuracy of ±0.1mm; the wind speed sensor is a cup sensor with a measurement range of 0 to 60m / s and an accuracy of ±0.3m / s.

[0129] ③ Load monitoring unit: Piezoelectric pressure sensor is selected, with a measurement range of 0~200kN, accuracy of ±1kN, response time ≤10ms, protection level of IP68, which can withstand road surface immersion and vehicle running.

[0130] ④ Data transmission unit: Uses a 5G / 4G+LoRa dual-mode communication module, supports full network compatibility, 5G / 4G transmission rate ≥10Mbps, latency ≤50ms, LoRa communication cache capacity 16GB, supports AES-256 encrypted transmission; Power supply unit uses a 20W solar panel + 10000mAh lithium battery to ensure continuous operation for more than 72 hours under no light conditions.

[0131] The other modules are deployed as follows: (1) Edge computing processing module: an embedded edge computing gateway is adopted, which is integrated into each multi-source sensing and monitoring terminal and connected to each unit of the terminal via a USB interface. The processor is ARM Cortex-A9 with a main frequency of 1.2GHz, 2GB of memory, and 128GB of storage capacity. It supports real-time data preprocessing and preliminary identification.

[0132] (2) AI intelligent analysis platform: Deployed on cloud server, adopting distributed architecture, configured with 4 server nodes, each server CPU is Intel Xeon E5-2690, memory 32GB, hard disk 1TB, supports multi-node parallel processing, can process effective data from 12 monitoring points at the same time; the database adopts MySQL8.0+MongoDB5.0 hybrid storage architecture, MySQL is used to store structured data, and MongoDB is used to store unstructured data such as road images, and the data storage capacity supports TB-level expansion.

[0133] (3) Maintenance management terminal: The mobile APP is deployed on 20 mobile phones of front-line maintenance personnel (Android 8.0 and above, iOS 12.0 and above systems), and the PC management backend is deployed on 3 management computers in the maintenance management center. It is connected to the cloud AI intelligent analysis platform through the local area network to realize real-time data synchronization.

[0134] (4) Auxiliary support module: The calibration module is linked with the multi-source sensing and monitoring terminal and is equipped with calibration references such as standard gray scale plate, standard crack simulation plate, standard thermometer, and standard weight block; the emergency response module is linked with the maintenance management terminal and the traffic management department system to realize the synchronization of emergency resource scheduling and traffic guidance information.

[0135] Based on the road condition characteristics of the monitoring scenario in this embodiment (high traffic volume, bridge sections susceptible to rainfall and temperature effects), the parameters of each module of the system are set specifically to ensure monitoring accuracy and early warning accuracy: Data acquisition parameters: (1) High-definition camera unit: The frame rate is set to 10 frames / second under normal working conditions (sunny daytime) and increased to 20 frames / second at night and in bad weather (rainy day, cloudy day). The infrared fill light is turned on from 18:00 to 6:00 the next day. The image size is uniformly set to 1024×768 pixels.

[0136] (2) Environmental sensing unit: The data collection interval is 10 minutes. The threshold for judging abnormal values ​​is set as follows: temperature > 85℃ or < -40℃, humidity > 100%RH or < 0%RH, precipitation > 4mm / min, wind speed > 60m / s. Data exceeding the threshold is judged as abnormal data and automatically removed.

[0137] (3) Load monitoring unit: The data acquisition interval is 1 second. The average load, maximum load and load frequency are calculated once per hour. The load peak extraction threshold is set to 5kN. Data below this threshold is judged as invalid load data.

[0138] Edge computing processing parameters: (1) Data preprocessing parameters: Gaussian filtering algorithm is used for image denoising, and the filter kernel size is 3×3; histogram equalization algorithm is used for image enhancement, and the contrast is increased by 30%; environmental parameters and load data are normalized using min-max algorithm and mapped to the [0,1] interval.

[0139] (2) Preliminary identification parameters: The crack identification threshold of the lightweight CNN algorithm (MobileNetV3) is set to a pixel ratio of ≥0.01%. Images below this threshold are judged to be without cracks and are stored only on the local SD card, not transmitted to the AI ​​intelligent analysis platform. The automatic cleaning threshold of the data cache unit is set to ≥90% of the SD card storage volume, and the oldest invalid data is automatically deleted.

[0140] AI intelligent analysis parameters: (1) Crack identification parameters: The identification threshold of the improved U-Net algorithm is set to 0.85, the identification accuracy is ≥98%, the false negative rate is ≤1%, the false positive rate is ≤1%, and it can identify micro cracks with a width of ≥0.1mm; Crack feature parameter extraction accuracy: length error ≤5mm, width error ≤0.1mm.

[0141] (2) Crack classification parameters: Strictly in accordance with industry standard DB41 / T1469-2017, and combined with the road conditions of this embodiment, the classification parameters are finely adjusted as follows: micro cracks (width ≤ 2mm, length ≤ 500mm), light cracks (width 2-5mm, length 500-1000mm), moderate cracks (width 5-10mm, length 1000-2000mm), and severe cracks (width ≥ 10mm, length ≥ 2000mm); monitoring frequency: micro cracks are monitored once every 24 hours, light cracks are monitored once every 12 hours, moderate cracks are monitored once every 6 hours, and severe cracks are monitored once every 1 hour.

[0142] (3) Trend prediction parameters: The input data time span of the LSTM neural network model is the past 6 months, the data sampling interval is 1 hour, the prediction period is the next 1-6 months, and the prediction error is ≤10%; the crack development rate warning threshold is set to 0.5 mm / month, and the warning mechanism is triggered when the threshold is exceeded.

[0143] Warning push parameters: Warning level determination parameters: Level 1 warning (minor cracks, development rate ≤0.2mm / month), Level 2 warning (minor cracks, 0.2-0.5mm / month; mild cracks, ≤0.3mm / month), Level 3 warning (mild cracks, 0.3-0.5mm / month; moderate cracks, ≤0.4mm / month), Level 4 warning (moderate cracks, >0.4mm / month; severe cracks, any development rate); Push priority: Level 4 warning > Level 3 warning > Level 2 warning. Level 4 warnings trigger push notifications via SMS, APP, WeChat official account, and background messages simultaneously. Level 3 warnings trigger push notifications via APP and SMS. Level 2 warnings trigger push notifications via APP and WeChat official account; Time for secondary push notifications for unprocessed warnings: Level 3 warnings exceeding 24 hours, Level 4 warnings exceeding 6 hours.

[0144] Calibration parameters: The calibration cycle is set to 1 month, with unified calibration performed by maintenance personnel on the 10th of each month; after severe environmental conditions (heavy rain, strong sandstorm, high temperature exposure), calibration is initiated manually; allowable calibration deviation range: temperature sensor ≤ ±1℃, humidity sensor ≤ ±3%RH, load sensor ≤ ±2kN, high-definition camera image distortion ≤ 5%, if the allowable range is exceeded, a calibration abnormality warning will be issued.

[0145] In this application example, the system operates according to the workflow of data acquisition, preprocessing, analysis, early warning, maintenance, feedback, and optimization. The specific working process is as follows: 1. Data Acquisition Phase: Twelve multi-source sensing and monitoring terminals deployed at intersections and bridge approach ramps collect road surface data in real time: high-definition camera units capture road surface images every 10-20 frames per second, capturing road surface cracks and conditions; environmental sensing units collect temperature, humidity, precipitation, and wind speed data every 10 minutes, recording environmental changes; load monitoring units collect vehicle load data every second, capturing the load dynamics of passing vehicles; the data transmission unit uses 5G communication to transmit the collected raw data to the edge computing processing module in real time; the power supply unit uses both solar power and lithium batteries to ensure uninterrupted data collection; when the 5G signal is interrupted at intersections or bridge sections, it automatically switches to LoRa mode, caches the data to the local SD card, and automatically synchronizes it to the edge computing processing module after the network is restored.

[0146] 2. Data preprocessing and preliminary identification stage: After receiving the raw data, the edge computing processing module performs targeted processing on various types of data: Gaussian filtering for noise reduction, histogram equalization for enhancement, size normalization, and grayscale processing are applied to the road surface image to remove road debris, shadows, and other interference, highlighting crack features; outlier removal, unit unification, and standardization are applied to environmental parameters to remove abnormal data exceeding the threshold; and Kalman filtering for noise reduction, peak extraction, and statistical processing are applied to the load data to extract the effective load data.

[0147] The preliminary identification unit uses the MobileNetV3 lightweight CNN algorithm to perform preliminary identification on the preprocessed road surface grayscale image. When the pixel proportion of the crack area in the image is ≥0.01%, it is determined that a crack exists, and the image and the corresponding environmental and load data are marked as valid data and transmitted to the AI ​​intelligent analysis platform. When the pixel proportion is <0.01%, it is determined that there is no crack, and the data is only stored on the local SD card without being transmitted. The data caching unit performs dual caching of raw data and preprocessed data. The SD card stores nearly 7 days' worth of data, while key valid data is synchronized to the cloud for backup to ensure that no data is lost.

[0148] 3. Intelligent Analysis and Trend Prediction Stage: After receiving valid data from the edge computing processing module, the AI ​​intelligent analysis platform's various units work collaboratively: the crack precision identification unit uses an improved U-Net algorithm to accurately identify cracks in valid images, extracting feature parameters such as crack length, width, area, and direction, and identifying micro cracks with a width of 0.1mm or more and moderate to severe cracks, achieving an accuracy rate of 98.5%; the crack classification unit, based on the extracted crack feature parameters and industry standards, classifies cracks into four levels: micro, light, moderate, and severe. Among them, 3 micro cracks and 1 light crack were detected at the intersection, and 4 micro cracks, 2 light cracks, and 1 moderate crack were detected on the bridge approach road section.

[0149] The trend prediction unit, based on the LSTM neural network algorithm and combining six months of historical monitoring data, real-time environmental parameters, and load data, constructs a crack development trend prediction model to predict crack changes over the next six months: At the intersection, the development rates of three microcracks are 0.15 mm / month, 0.18 mm / month, and 0.22 mm / month, respectively, while the development rate of one minor crack is 0.28 mm / month. At the bridge approach road section, the development rates of four microcracks are 0.16-0.23 mm / month, two minor cracks are 0.29-0.32 mm / month, and one moderate crack is 0.38 mm / month. The historical data management unit stores all monitoring data, identification results, grading results, and trend prediction results in a database, supporting subsequent queries and statistical analysis.

[0150] 4. Early Warning Judgment and Push Phase: After receiving the analysis results from the AI ​​intelligent analysis platform, the early warning push module determines the early warning level of each crack based on its crack grade and development rate: At the intersection, of the three minor cracks, two have a development rate ≤0.2mm / month (Level 1 warning, routine monitoring only), and one has a development rate of 0.22mm / month (Level 2 warning, alert-type warning); one minor crack has a development rate of 0.28mm / month (Level 2 warning, alert-type warning); on the bridge approach road section, of the four minor cracks, three have a development rate ≤0.2mm / month (Level 1 warning), and one has a development rate of 0.23mm / month (Level 2 warning); of the two minor cracks, one has a development rate of 0.29mm / month (Level 2 warning), and one has a development rate of 0.32mm / month (Level 3 warning, emergency alert-type warning); one moderate crack has a development rate of 0.38mm / month (Level 3 warning, emergency alert-type warning).

[0151] The information generation unit generates standardized early warning information based on the early warning level, including the location of the monitoring point (accurate to the specific intersection traffic light pole number and bridge pile number), early warning time, early warning level, crack characteristic parameters, development trend, maintenance suggestions, etc.; the multi-channel push unit pushes early warning information to the maintenance management terminal according to the push priority: Level 2 early warnings are pushed to front-line maintenance personnel in the corresponding area through the APP and WeChat official account, Level 3 early warnings are pushed to front-line maintenance personnel and management personnel through the APP and SMS, and Level 1 early warnings are not pushed, but are only recorded in the background.

[0152] 5. Maintenance and Emergency Response Phase: After receiving early warning information via a mobile app, maintenance personnel carry out corresponding maintenance work: For cracks with a level-two warning, maintenance personnel will go to the site for verification within 24 hours. After confirming that the actual situation of the crack is consistent with the system monitoring results, they will closely monitor the development of the crack, increase the monitoring frequency, and take simple preventive maintenance measures (such as cleaning the crack surface and dust prevention treatment). They will also report the verification results and maintenance progress through the app. For cracks with a level-three warning, maintenance personnel will go to the site for verification within 12 hours. Based on the maintenance suggestions pushed by the system, they will formulate a targeted maintenance plan, use crack sealant for sealing treatment, and upload photos and videos of the maintenance site through the app to provide feedback on the maintenance effect.

[0153] Managers can view early warning information and maintenance task progress of all monitoring points through the PC-based management backend, coordinate and allocate maintenance resources, and generate maintenance statistical reports. If severe cracks or crack development rates > 0.4 mm / month (Level 4 warning) are detected, the emergency response module automatically initiates the emergency process, generates emergency maintenance tasks, assigns them to the nearest maintenance personnel, pushes emergency response plans (including temporary traffic diversion suggestions and emergency maintenance material lists), coordinates with traffic management departments to set up warning signs, guides vehicles to detour, and tracks the progress of emergency maintenance in real time to ensure timely crack repair and reduce safety hazards.

[0154] 6. Calibration and System Optimization Phase: On the 10th of each month, maintenance personnel use standard references to calibrate the sensors and high-definition cameras of 12 multi-source sensing and monitoring terminals: a standard crack simulation board is placed in the monitoring area, and the camera parameters are adjusted to ensure accurate crack image acquisition; the temperature and humidity sensors are compared with standard testing equipment, and the measurement parameters are adjusted; a standard weight block is placed on top of the load sensor to calibrate the sensor sensitivity. After calibration, the system automatically records the calibration data. If the temperature sensor of a terminal is found to have a deviation of 1.2℃, exceeding the allowable range, the system issues a calibration anomaly warning, and maintenance personnel promptly repair or replace the sensor.

[0155] The system continuously optimizes its AI algorithm based on maintenance feedback, calibration data, and historical monitoring data: by adding new crack monitoring data after maintenance, the attention mechanism weights of the improved U-Net algorithm are adjusted to improve the recognition accuracy of micro-cracks; the parameters of the LSTM trend prediction model are fine-tuned according to the actual development of cracks to reduce prediction errors; and the early warning level judgment criteria are optimized in combination with maintenance effects to improve the pertinence and accuracy of early warnings, thus achieving continuous system optimization.

[0156] In this application example, the implementation effect was verified after the system was deployed and running for 6 months. The verification results are as follows: (1) Monitoring efficiency: The system has achieved automated and routine monitoring of road surface cracks at intersections and bridge approach sections, eliminating the need for manual inspections and saving approximately RMB 8,000 in labor costs per month. The monitoring coverage rate reaches 100%, with no blind spots. The accuracy rate of identifying micro-cracks reaches 98.5%, with a missed detection rate of 0.8% and a false detection rate of 0.7%, which is far superior to traditional manual inspections (with a missed detection rate of approximately 15% and a false detection rate of approximately 10%).

[0157] (2) Accuracy of early warning: A total of 8 Level II early warnings and 2 Level III early warnings were issued, with an accuracy rate of 95%. There were no false or missed early warnings. All early warning cracks were promptly checked and dealt with, and no cracks were further deteriorated into severe cracks. The goal of early warning and early repair was achieved.

[0158] (3) Maintenance effect: Through systematic and precise monitoring and targeted maintenance, the road surface crack repair rate of intersections and bridge approach sections reached 100%, and the maintenance cost was reduced by 30% compared with traditional manual inspection and maintenance; after timely repair of moderate cracks in bridge approach sections, no road surface damage or collapse occurred, ensuring road traffic safety.

[0159] (4) System stability: During system operation, there were no major failures, and the average failure rate was ≤0.5%; under severe weather conditions such as rainstorms, high temperatures, and strong winds, it can still collect and analyze data normally, with a data transmission success rate of ≥99.5%, and the caching mechanism effectively avoids data loss; the calibration module ensures the accuracy of data collection, and the data measurement error is controlled within the allowable range.

[0160] In summary, the AI-driven long-term monitoring and analysis system for road maintenance technology in this embodiment solves the problems of low efficiency, untimely early warning, and inability to predict crack development trends in traditional manual inspections. It improves the level of intelligence in road maintenance, realizes systematic monitoring, trend analysis, accurate early warning, and timely repair of road cracks, reduces maintenance costs, and ensures road traffic safety. It has good practicality and promotional value.

[0161] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0162] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI-driven long-term monitoring and analysis system for road maintenance technology, characterized in that, The system includes a multi-source sensing monitoring terminal, an edge computing processing module, an AI intelligent analysis platform, an early warning push module, and a maintenance management terminal. These modules interact via a wireless communication network. The multi-source sensing monitoring terminal is deployed at key road condition locations such as intersections and bridges to collect real-time images of pavement cracks, environmental parameters, and road load data. The edge computing processing module, connected to the multi-source sensing monitoring terminal, preprocesses the collected data, filters noise, and extracts features to select valid data and preliminarily determine the existence of cracks. The AI ​​intelligent analysis platform, communicatively connected to the edge computing processing module, performs in-depth analysis of the preprocessed valid data, including crack identification, crack level classification, and crack development trend prediction. The early warning push module, connected to the AI ​​intelligent analysis platform, pushes early warning information and corresponding maintenance suggestions to the maintenance management terminal according to preset early warning levels based on the analysis results of the AI ​​intelligent analysis platform. The maintenance management terminal is used to receive early warning information, view monitoring data and trend analysis reports, and realize the initiation, tracking and closed-loop management of maintenance tasks.

2. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 1, characterized in that, The multi-source sensing and monitoring terminal includes a high-definition camera unit, an environmental sensing unit, a load monitoring unit, and a data transmission unit. The high-definition camera unit uses an industrial-grade high-definition camera with the lens facing the road surface to collect real-time images of the road surface. The environmental sensing unit includes a temperature sensor, a humidity sensor, a precipitation sensor, and a wind speed sensor to collect data on ambient temperature, air humidity, precipitation, and wind speed in the monitored area. The load monitoring unit uses a piezoelectric pressure sensor embedded 5-10 cm below the road surface to collect load data from passing vehicles. The data transmission unit uses a 5G / 4G+LoRa dual-mode communication method, supports encrypted data transmission, ensures data security and real-time performance, and switches to LoRa mode for data caching and delayed transmission when the 5G / 4G signal is interrupted.

3. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 1, characterized in that, The edge computing processing module includes a data preprocessing unit, a preliminary identification unit, and a data caching unit. The data preprocessing unit is used to perform noise reduction, enhancement, and size normalization on the image data collected by the multi-source sensing and monitoring terminal, and to perform outlier removal, unit unification, and standardization on the environmental parameters and load data. The preliminary identification unit uses a lightweight CNN algorithm to perform preliminary identification on the preprocessed road surface image, determine whether there are cracks, and remove invalid images without cracks to reduce the amount of data transmitted to the AI ​​intelligent analysis platform. The data caching unit adopts a dual caching mechanism of local SD card + cloud backup. The SD card stores raw data and preprocessed data for the past 7 days. When the network is restored, the cached data is automatically synchronized to the AI ​​intelligent analysis platform.

4. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 1, characterized in that, The AI ​​intelligent analysis platform includes a crack precision identification unit, a crack level classification unit, a trend prediction unit, and a historical data management unit. The crack precision identification unit uses an improved U-Net deep learning algorithm to identify cracks in valid images transmitted from the edge computing processing module, extracting feature parameters such as crack length, width, area, and orientation. The crack level classification unit classifies pavement cracks into four levels: micro-cracks, light cracks, moderate cracks, and severe cracks, based on crack feature parameters and industry standards. The trend prediction unit, based on the LSTM neural network algorithm and combined with historical crack data, real-time environmental parameters, and load data, constructs a crack development trend prediction model to predict the crack width and length changes over the next 1-6 months, calculates the crack development rate, and triggers an early warning mechanism when the development rate exceeds a preset threshold. The historical data management unit stores all monitoring data, analysis results, and maintenance records, constructing a pavement maintenance database that supports data querying, statistics, and export.

5. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 1, characterized in that, The early warning push module includes an early warning level determination unit, an information generation unit, and a multi-channel push unit; The warning level determination unit divides the warning level into four levels based on the crack level and development rate: Level 1 warning, which only performs routine monitoring and does not push warning information; Level 2 warning, sending push notifications and alerts; Level 3 early warning: pushes emergency reminders and preliminary maintenance suggestions; Level 4 early warning will push emergency warning information, detailed maintenance plans and emergency response suggestions; The information generation unit is used to generate early warning information, including crack location, crack level, development trend, early warning level, and maintenance recommendations, based on the early warning level and analysis results. The multi-channel push unit supports four push methods: SMS, APP push, WeChat official account push, and background message reminder. Push priorities can be set according to the maintenance personnel's positions to ensure that early warning information is delivered in a timely manner.

6. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 1, characterized in that, The maintenance management terminal includes a mobile APP and a PC management backend. The mobile APP supports Android and iOS systems and is used by front-line maintenance personnel. It can receive early warning information, view pavement crack images and monitoring data, report maintenance progress, and provide feedback on maintenance results. The PC management backend is used by management personnel and has functions such as data visualization, trend analysis, early warning management, maintenance task allocation, maintenance record query, and report generation. It supports multi-user permission management and can assign different operation permissions according to job positions.

7. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 4, characterized in that, The improved U-Net deep learning algorithm adds an attention mechanism module and a residual connection module to the traditional U-Net algorithm. The attention mechanism module is used to highlight the features of the crack area, suppress background interference, and improve the recognition accuracy of small cracks. The residual connection module is used to solve the gradient vanishing problem in the deep network training process, accelerate the model training speed, and improve the model's generalization ability; the LSTM neural network algorithm introduces an attention mechanism to focus on environmental parameters and load data that have a significant impact on crack development.

8. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 1, characterized in that, It also includes a calibration module, which is used to periodically calibrate the sensors and high-definition cameras of the multi-source sensing monitoring terminal. The calibration cycle is one month to ensure the accuracy of the collected data. The calibration module includes a standard reference and a calibration algorithm. By comparing the data collected by the monitoring terminal with the parameters of the standard reference, the parameter settings of the monitoring terminal are adjusted to eliminate measurement errors.

9. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 1, characterized in that, The wireless communication network adopts an encrypted transmission protocol, which includes three layers: data encryption, identity authentication, and access control. The data encryption uses the AES-256 encryption algorithm to encrypt the transmitted data. Identity authentication uses a dual authentication method of account password + dynamic verification code to ensure that only authorized users can access the system; access control restricts the scope of user access to system data and functions based on user permissions.

10. The AI-driven long-term monitoring and analysis system for road maintenance technology according to claim 1, characterized in that, It also includes an emergency response module, which is connected to the early warning push module. When a level four early warning is triggered, the emergency response process is automatically started, an emergency maintenance task is generated, assigned to the nearest maintenance personnel, and an emergency response plan is pushed, including temporary traffic diversion suggestions, a list of emergency maintenance materials and equipment, and the progress of emergency maintenance is tracked in real time to ensure that the cracks are repaired in a timely manner.

Images