Solid waste aggregate concrete crack real-time monitoring method and system

Abstract

The invention relates to the field of concrete crack monitoring, in particular to a solid waste aggregate concrete crack real-time monitoring method and system, and the method comprises the steps: collecting sensor data in real time; calculating a deviation value between the current data and the initial reference data; when the deviation value exceeds a threshold value, the deviation value is used for compensating the subsequently collected sensor data; updating the initial reference data by using the mean value of the compensated data in the previous period; processing the compensated data in different channels, extracting strain accumulation trend characteristics in a long-term channel through a large time window, and extracting acoustic emission instantaneous characteristics in a short-term channel through a small time window; setting a long-term feature weight to be higher than a short-term feature weight, calculating a long-term feature change rate, and when the change rate is stable, improving the long-term weight and reducing the short-term weight, and filtering features lower than a noise threshold in a short-term channel; and fusing the weighted long-term features with the effective short-term features, and inputting the fused features into a pre-training model to obtain a fracture state prediction result in a future predetermined period.

Description

Technical Field

[0001] This invention belongs to the field of concrete crack monitoring, specifically relating to a real-time monitoring method and system for cracks in solid waste aggregate concrete. Background Technology

[0002] Existing solid waste aggregate concrete crack monitoring technologies suffer from the problem of "drift accumulation" of long-term data. This is because existing solid waste aggregate concrete crack monitoring technologies rely on historical data trends for prediction. However, in long-term monitoring, the historical data errors caused by sensor drift accumulate over time, causing the model to learn "pseudo-trends" (such as slow shifts in strain data being misjudged as precursors to crack propagation) and ignore the true long-term trends (such as slow strain accumulation), resulting in predictions that deviate from reality. Summary of the Invention

[0003] The purpose of this invention is to provide a method for real-time monitoring of cracks in solid waste aggregate concrete, and at the same time, to provide a system for real-time monitoring of cracks in solid waste aggregate concrete, so as to solve the problems mentioned in the background art.

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The first aspect discloses a method for real-time monitoring of cracks in solid waste aggregate concrete, including the following steps: Import the initial baseline data of the sensor in the crack-free state of solid waste aggregate concrete; Real-time acquisition of sensor data; Calculate the deviation between the current data and the initial baseline data; When the deviation value exceeds the threshold, the deviation value is used to compensate for the subsequent sensor data. The initial baseline data is updated using the mean of the data after compensation in the previous period according to the set period. The compensated data is processed by channel. The long-term channel is used to extract the cumulative strain trend features with a large time window, while the short-term channel is used to extract the instantaneous acoustic emission features with a small time window. Set the weight of long-term features higher than that of short-term features, calculate the rate of change of long-term features, and when the rate of change is stable, increase the weight of long-term features and decrease the weight of short-term features to filter out features in the short-term channel that are below the noise threshold. The weighted long-term features are fused with the effective short-term features and input into the pre-trained model to obtain the crack state prediction results for the future predetermined period. When the result meets the alarm conditions, a real-time alarm is triggered.

[0005] Furthermore, the import of initial reference data from sensors in the crack-free state of solid waste aggregate concrete specifically includes: selecting the crack-free state after the solid waste aggregate concrete has been poured and reached its design strength for initial reference data acquisition; the acquisition objects include strain sensors, acoustic sensors, temperature and humidity sensors, and image sensors; acquiring corresponding stable reference data for each type of sensor in the crack-free state; the acquired initial reference data is first transmitted to an edge computing node for preprocessing; the preprocessed initial reference data is synchronously stored in a cloud server as a reference for subsequent real-time data comparison and deviation calculation.

[0006] Furthermore, the real-time acquisition of sensor data specifically includes: each sensor synchronously acquiring corresponding data at a set frequency; the acquisition range includes sensors embedded in stress concentration areas inside the concrete and sensors attached to easily cracked parts of the concrete surface; the acquired data is transmitted to the edge computing node in real time via wireless communication.

[0007] Furthermore, the calculation of the deviation between the current data and the initial reference data specifically includes: after receiving the data collected in real time by each sensor, the edge computing node calculates the deviation value from the corresponding initial reference data according to the sensor type; the strain sensor subtracts the initial stable strain value under the crack-free state from the real-time collected strain value to obtain the strain deviation value; the acoustic sensor subtracts the initial background noise value from the real-time collected acoustic emission signal intensity to obtain the acoustic signal deviation value; the temperature sensor subtracts the initial stable temperature value from the real-time collected temperature value to obtain the temperature deviation value; the humidity sensor subtracts the initial stable humidity value from the real-time collected humidity value to obtain the humidity deviation value; the image sensor calculates the image feature deviation value by comparing the gray value changes and edge feature differences between the real-time collected image and the initial crack-free image sequence; after all deviation values ​​are calculated, they are temporarily stored in the local cache of the edge computing node for subsequent compensation judgment.

[0008] Furthermore, when the deviation value exceeds the threshold, the subsequent sensor data is compensated using the deviation value. Specifically, this includes: when the deviation value calculated by a certain type of sensor exceeds the corresponding threshold, the subsequent sensor data of that type is compensated using the deviation value; the compensated sensor data is temporarily stored in the local cache of the edge computing node and simultaneously uploaded to the cloud server for backup; the deviation of the subsequent collected data is continuously monitored during the compensation process; if the deviation value collected multiple times returns to the threshold range, the current compensation operation is stopped and the normal data acquisition mode is restored; if the deviation value still exceeds the threshold after compensation, a sensor calibration prompt is triggered to remind staff to check the working status of the sensor.

[0009] Furthermore, the step of updating the initial baseline data with the average of the compensated data from the previous period according to a set cycle specifically includes: setting an update cycle for the initial baseline data, and automatically triggering an update operation after each cycle; the edge computing node acquiring all sensor data that has been compensated in the previous cycle, filtering out abnormal data during acquisition; if the proportion of effective compensated data for a certain type of sensor in the previous cycle is not up to standard, then pausing the current update of the baseline data for that type of sensor, using the original initial baseline data, and issuing a sensor data abnormality prompt; for sensor categories with sufficient effective data, calculating the average of their compensated data from the previous cycle; after calculation, updating the initial baseline data stored locally on the edge computing node with the average values ​​corresponding to each type of sensor, and simultaneously uploading the updated baseline data of each type of sensor to the cloud server to replace the original stored initial baseline data.

[0010] Furthermore, the process of processing the compensated data by channel, with the long-term channel using a large time window to extract strain accumulation trend features and the short-term channel using a small time window to extract acoustic emission instantaneous features, specifically includes: first, determining the time window parameters for the long-term and short-term channels; edge computing nodes acquiring the compensated data from both channels and extracting data segments according to the corresponding time windows; the long-term channel processing the extracted data segments to extract strain accumulation trend features and continuously updating these features using a set sliding window method; the short-term channel processing the extracted data segments to extract acoustic emission instantaneous features and capturing these features in real time using a set sliding window method; and the features extracted from both channels undergoing data normalization processing to eliminate dimensional differences.

[0011] Furthermore, the process of fusing the weighted long-term features with the effective short-term features and inputting them into the pre-trained model to obtain the crack state prediction result for the future predetermined period specifically includes: concatenating the weighted long-term features with the effective short-term features, normalizing the two types of features to form a fused feature vector, and matching the vector dimension with the input layer dimension of the pre-trained model; the pre-trained model is an improved LSTM model with an added attention mechanism, and after inputting the fused feature vector into the model, the model assigns key weights to the key features affecting crack expansion through the attention mechanism; the model outputs the crack state prediction result for the future predetermined period based on the input fused features, and the prediction result includes the predicted crack width value and change curve within the period, the crack risk level at each time period, and the possible development stage of the crack.

[0012] Furthermore, the real-time alarm triggering when the result meets the alarm conditions specifically includes: setting multiple types of alarm triggering conditions; initiating the alarm process when the prediction result of any time period within a predetermined future period output by the pre-trained model meets any condition; the edge computing node synchronizing alarm information including crack monitoring location, predicted crack width and change trend, risk level, and triggering reason to the cloud server; the cloud server initiating multi-channel notifications: for the PC terminal of the remote monitoring layer, a window displaying complete alarm information pops up on the monitoring interface and remains until the staff manually confirms; for the mobile terminal, an alarm push containing crack location and risk level is sent to the bound personnel's device, and an alarm SMS containing crack monitoring location, predicted crack width, risk level, and a link to view detailed information is sent to relevant personnel; after each alarm is triggered, the cloud server automatically records an alarm log containing alarm trigger time, crack monitoring location, prediction parameters that triggered the alarm, a list of alarm notification recipients, and alarm confirmation time.

[0013] The second aspect discloses a real-time monitoring system for cracks in solid waste aggregate concrete, comprising: Sensors are used for initial baseline data acquisition and real-time monitoring data acquisition of crack-free solid waste aggregate concrete. The benchmark management unit is used for initial benchmark data import, periodic updates, and edge-cloud synchronization. The deviation calculation and compensation unit is used to calculate the deviation between the monitoring data and the benchmark in real time, and to perform data compensation when the deviation exceeds the threshold to correct the sensor drift error. The feature-segmentation processing unit is used to extract and filter features from the compensated data, separating features from interference information. The model prediction unit is used to predict crack trends based on fused features and output the crack status in the future period; the alarm management unit is used to monitor the prediction results, trigger alarms and notify through multiple channels.

[0014] Beneficial Effects: This application imports initial reference data from sensors under crack-free conditions, calculates the deviation between the current data and the reference in real time, and uses this deviation to compensate subsequent data when the deviation exceeds a threshold. The reference is then updated periodically with the average of the compensated data. This dynamically corrects the accumulation of historical data errors caused by sensor drift, avoids "pseudo-trend" interference such as slow strain data shifts, and accurately captures true long-term trends such as slow strain accumulation. It ensures the authenticity of monitoring data from the data source and solves the problem in existing technologies where models learn "pseudo-trends" due to data drift, leading to predictions deviating from reality. The compensated data is processed in two channels: long-term and short-term. The long-term channel extracts strain accumulation trend features with a large time window, while the short-term channel extracts instantaneous acoustic emission features with a small time window. The long-term features are initially given higher weights, which are dynamically adjusted based on the rate of change of the long-term features (increasing long-term weights and decreasing short-term weights when the rate of change is stable). Features below the noise threshold in the short-term channel are filtered out, and data normalization eliminates dimensional differences, effectively separating key monitoring features from interference information and providing high-quality feature input for subsequent predictions. Attached Figure Description

[0015] Figure 1 This is a flowchart of a method for real-time monitoring of cracks in solid waste aggregate concrete according to the present invention.

[0016] Figure 2 This is a block diagram of a real-time monitoring system for cracks in solid waste aggregate concrete according to the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] This application discloses a method for real-time monitoring of cracks in solid waste aggregate concrete, such as... Figure 1 The steps include: Import the initial baseline data of the sensor in the crack-free state of solid waste aggregate concrete; Real-time acquisition of sensor data; Calculate the deviation between the current data and the initial baseline data; When the deviation value exceeds the threshold, the deviation value is used to compensate for the subsequent sensor data. The initial baseline data is updated using the mean of the data after compensation in the previous period according to the set period. The compensated data is processed by channel. The long-term channel is used to extract the cumulative strain trend features with a large time window, while the short-term channel is used to extract the instantaneous acoustic emission features with a small time window. Set the weight of long-term features higher than that of short-term features, calculate the rate of change of long-term features, and when the rate of change is stable, increase the weight of long-term features and decrease the weight of short-term features to filter out features in the short-term channel that are below the noise threshold. The weighted long-term features are fused with the effective short-term features and input into the pre-trained model to obtain the crack state prediction results for the future predetermined period. When the result meets the alarm conditions, a real-time alarm is triggered.

[0019] The import of initial reference data from sensors in a crack-free state of solid waste aggregate concrete is specifically implemented by selecting a crack-free state after the solid waste aggregate concrete has been poured and reached its design strength for initial reference data acquisition. The data acquisition covers strain sensors, acoustic sensors, temperature and humidity sensors, and image sensors in the system. For the strain sensors, stable strain values ​​under crack-free conditions are acquired. These sensors have an accuracy of ±2με and a range of -2000~2000με, and data is continuously acquired for 12 hours to obtain stable data.

[0020] For acoustic sensors, background noise values ​​in the 20-200kHz frequency band are collected under crack-free conditions to ensure accurate acoustic signal reference when no cracks are generated.

[0021] For the temperature and humidity sensors, the stable values ​​of ambient temperature and humidity under the crack-free condition are collected. The accuracy of the temperature sensor is ±0.5℃, and the accuracy of the humidity sensor is ±3%RH.

[0022] For the image sensor, a sequence of images of the concrete surface in a crack-free state was acquired. The sensor has a resolution of 1280×720 and a frame rate of 25fps. 100 frames were continuously acquired to cover the entire monitoring area.

[0023] During the data collection process, interference factors such as external vibration and drastic temperature changes should be avoided.

[0024] The initial reference data collected is first transmitted to edge computing nodes for preprocessing, including noise filtering to remove interference signals from the acquisition process. The preprocessed initial reference data is then synchronously stored on a cloud server as a benchmark for calculating deviations from subsequent real-time acquired data. Simultaneously, initial reference data is recorded according to the sensor deployment location, distinguishing between the reference values ​​corresponding to sensors embedded in concrete and those mounted on the concrete surface.

[0025] In practice, the real-time acquisition of sensor data involves each sensor synchronously acquiring data at a set frequency after the system is started. The acquisition range covers all deployed sensors, including sensors embedded in stress concentration areas within the concrete and sensors mounted on easily cracked surfaces.

[0026] The collected data is transmitted in real time to the edge computing node via LoRa wireless communication. This communication method has a transmission distance of no more than 1km and a power consumption of no more than 50mW. When critical data such as abnormal acoustic emission signals are detected, a priority transmission mechanism is activated to ensure that emergency information is delivered first. The sensor's operating status is continuously monitored during the data acquisition process, and a sensor fault warning is immediately issued if data interruption or abnormality occurs.

[0027] The calculation of the deviation between the current data and the initial reference data is specifically implemented as follows: after receiving the data collected in real time by each sensor, the edge computing node calculates the deviation from the corresponding initial reference data according to the sensor type. For strain sensors, the strain deviation value is obtained by subtracting the initial stable strain value under crack-free conditions from the real-time collected strain value, with the calculation accuracy retained to ±2με.

[0028] For acoustic sensors, in the 20~200kHz frequency band, the acoustic signal deviation value is obtained by subtracting the initial background noise value from the real-time acquired acoustic emission signal intensity.

[0029] For temperature sensors, the temperature deviation is obtained by subtracting the initial stable temperature value from the real-time acquired temperature value, and the calculation result is retained to an accuracy of ±0.5℃.

[0030] For humidity sensors, the humidity deviation value is obtained by subtracting the initial stable humidity value from the real-time collected humidity value, and the calculation result is retained to an accuracy of ±3%RH.

[0031] For image sensors, image feature deviation values ​​are calculated by comparing the grayscale changes and edge feature differences between real-time acquired images and initial crack-free image sequences. These deviation values ​​are expressed as pixel differences at a resolution of 1280×720. After all deviation values ​​are calculated, they are temporarily stored in the local cache of the edge computing node for subsequent compensation decisions.

[0032] When the deviation value exceeds the threshold, the deviation value is used to compensate for the subsequent sensor data. In practice, this means that a deviation threshold is first set for each type of sensor in the system.

[0033] The deviation threshold of the strain sensor is set to ±5με. When the calculated strain deviation value exceeds ±5με, the excess strain deviation value is subtracted from each subsequent strain data acquisition to keep the strain data within a reasonable reference range.

[0034] The deviation threshold of the acoustic sensor is set to 1.5 times the initial background noise value under crack-free conditions. When the acoustic signal deviation exceeds this threshold, the intensity of each subsequent acoustic emission signal is reduced by the excess acoustic signal deviation value to eliminate the interference of noise drift on the data.

[0035] The temperature sensor's deviation threshold is set to ±2℃. When the temperature deviation exceeds ±2℃, the excess temperature deviation value is subtracted from every subsequent temperature data acquisition to correct for errors caused by ambient temperature drift.

[0036] The humidity sensor's deviation threshold is set to ±8%RH. When the humidity deviation exceeds ±8%RH, this excess deviation value is subtracted from every subsequent humidity data acquisition to ensure the humidity data reflects the true environmental conditions. The image sensor's deviation threshold is set to twice the average pixel difference of the initial image sequence under crack-free conditions. When the image feature deviation exceeds this threshold, this excess image feature deviation value is subtracted from the pixel difference of every subsequent image acquisition to restore the true image features of the concrete surface.

[0037] The compensated sensor data is temporarily stored in the local cache of the edge computing node and simultaneously uploaded to the cloud server for backup. During the compensation process, the deviation of subsequent data acquisition is continuously monitored. If the deviation values ​​of three consecutive acquisitions return to the threshold range, the current compensation operation is stopped, and normal data acquisition mode is restored. If the deviation value still exceeds the threshold after compensation, a sensor calibration prompt will be triggered, reminding staff to check the sensor's working status.

[0038] The process of updating the initial baseline data using the average of the compensated data from the previous period, according to a set period, is implemented by first setting the update period for the initial baseline data to 24 hours. The update period begins timing after the system starts up and completes the first initial baseline data acquisition, and the update operation is automatically triggered at the end of each period.

[0039] The edge computing node first acquires all the sensor data that has been compensated in the previous period. When acquiring data, abnormal data is first filtered out. If the effective compensation data volume of a certain type of sensor is less than 80% in the previous period, the update of the baseline data of that type of sensor is paused, the original initial baseline data is used, and a sensor data abnormality prompt is issued.

[0040] For sensor categories with sufficient effective data volume, the mean value of the data after the previous compensation cycle is calculated. For strain sensors, the arithmetic mean of the strain data after the previous compensation cycle is calculated, with an accuracy of ±2με. For acoustic sensors, the mean value of the acoustic emission signal intensity in the 20–200kHz frequency band after the previous compensation cycle is calculated. For temperature sensors, the mean value of the temperature data after the previous compensation cycle is calculated, with an accuracy of ±0.5℃. For humidity sensors, the mean value of the humidity data after the previous compensation cycle is calculated, with an accuracy of ±3%RH. For image sensors, the mean value of the image feature deviation value after the previous compensation cycle is calculated; this mean value is derived from pixel difference data at a resolution of 1280×720.

[0041] After calculation, the initial baseline data stored locally on the edge computing nodes is updated using the average values ​​corresponding to each type of sensor. Simultaneously, the updated baseline data for each sensor is uploaded to the cloud server, replacing the original initial baseline data stored in the cloud, ensuring consistency between the baseline data at the edge and in the cloud. After each update operation, the edge computing node records an update log, which includes the update time, the amount of valid data from each sensor in the previous period, the updated baseline average value for each sensor, and the change in baseline data before and after the update. The updated initial baseline data will serve as the new baseline for calculating the deviation values ​​of the real-time sensor data in the next period.

[0042] The existing method updates the initial baseline data stored locally on the edge computing node with the mean values ​​corresponding to various sensors. This method uses simple arithmetic mean values ​​and does not consider short-term fluctuations in the data within the period (such as occasional instantaneous interference). This can easily cause the baseline value to "follow the fluctuations" and fail to reflect the true long-term stable state. It also ignores the temporal continuity of the data (such as frequent short-term fluctuations that can lead to mean distortion) and the reliability of the sensors themselves (such as if the historical data of a certain sensor has drifted for a long time, even if the effective proportion meets the standard, the mean value is still unreliable). Therefore, this application further proposes a benchmark update method that integrates temporal continuity testing and sensor confidence, including: Input data preparation, input data list, for a single type of sensor: Time series data after compensation in the previous cycle: ,in This refers to the total number of data collections within the period (e.g., 24 hours × 60 minutes / collection = 1440 collections). For the first Data after compensation ); Sensor historical parameters: ① Historical deviation stability coefficient (The coefficient of variation of the benchmark deviation over the last 3 periods; the closer it is to 0, the more stable it is); ② Acquisition frequency compliance rate (Actual number of data collections within the period / Theoretical number of data collections); Time series stationarity parameter: a preset threshold for the coefficient of variation of the difference between adjacent data points. (such as strain sensors) Acoustic sensors ).

[0043] Perform a time series continuity test (removing "pseudo-valid data"). By testing the time series stationarity of the data, "continuously fluctuating data" caused by instantaneous sensor failures are removed, ensuring that the data used in the calculation has reliability in the time dimension. Calculate the difference sequence between adjacent data: ; Reflects the data fluctuation range between adjacent acquisition times; Calculate the coefficient of variation of fluctuations (a measure of time series stationarity): ;in: , for standard deviation for The mean; Data filtering rules: If Time series stable, retain all compensated data. ;like Excessive fluctuations will result in the removal of items with fluctuations exceeding a certain threshold. of corresponding (That is, removing "abnormal fluctuation points"), to obtain the filtered data. ( , (The number of valid counts after filtering).

[0044] Sensor confidence level calculation (quantifying sensor reliability) is performed based on the sensor's historical performance and current data acquisition status. ( A higher value indicates more reliable sensor data, which serves as the basis for subsequent weighted calculations. ; : This represents the historical deviation stability weight (taken as 0.6, because historical stability has a greater impact on data reliability). : This is the weight for the frequency acquisition compliance rate (taken as 0.4, to complementarily reflect the sensor's working status). : This is the historical deviation stability coefficient (such as the coefficient of variation of the benchmark deviation over the last 3 periods). , (Based on the first four cycles). : The current period's sampling frequency compliance rate ( If the theoretical number of trials is 1440, but the actual number of trials is 1390, then... ).

[0045] Intermediate output: Sensor confidence level (such as strain sensors) Image sensor ).

[0046] The dynamic effective data threshold determination (replacing the fixed 80% threshold) no longer uses a fixed threshold, but instead dynamically adjusts the effective data percentage threshold T based on the sensor confidence level C (the higher the confidence level, the higher the tolerance for data volume, and the lower the threshold can be; conversely, a higher data volume is required to ensure reliability): First, calculate the percentage of valid data: (M is the number of valid samples after the second step of filtering, and N is the total number of samples collected); If The sensor data is eligible for updating; proceed to the next step of mean calculation. : Pause updates, continue using the original baseline values, and issue a "Insufficient reliability of sensor data" warning (instead of a simple "data anomaly", which is more accurate).

[0047] Then, a time-smoothed weighted mean is calculated (instead of the arithmetic mean). The benchmark value is calculated using a combination of "confidence weighting + time-smoothing", which reflects the reliability of the sensor and suppresses the impact of short-term fluctuations on the benchmark. ; : This is the baseline value for the final update; : This is the time series smoothing coefficient (set to 0.6 to balance the continuity between the current period data and the historical benchmark, and avoid abrupt changes in the benchmark). : This refers to the i-th valid data after the second step of filtering; : This refers to the original baseline value before the update; : The calculated sensor confidence level (since the sensor confidence level is fixed within the same period, it can be simplified to) The confidence weight is reflected in "whether it is included in the calculation" and the subsequent system's confidence level in the benchmark. In actual calculation, it is simplified to "mean of filtered data × smoothing coefficient + original benchmark × smoothing remainder"). For different sensor adaptation adjustments: for example, acoustic sensors: adjustments are required... The data is limited to the 20~200kHz frequency band, and the formula remains unchanged; for example, image sensors: The formula remains unchanged for the filtered image feature deviation value (based on 1280×720 pixel difference); for example, for a temperature and humidity sensor: only adjustment is needed. (Due to more frequent fluctuations in temperature and humidity, a higher historical smoothing weight is required).

[0048] Then output the results: the updated baseline value μ for each sensor. new It can also output a confidence report of the sensor (such as C, R, and T values).

[0049] The process involves processing the compensated data across multiple channels. The long-term channel uses a large time window to extract cumulative strain trend features, while the short-term channel uses a small time window to extract instantaneous acoustic emission features. Specifically, the time window parameters for the long-term and short-term channels are first determined. The long-term channel sets a time window of 6 hours for the compensated strain data. The short-term channel sets a time window of 1 minute for the compensated acoustic emission data.

[0050] Edge computing nodes acquire compensated data from two channels and extract data segments according to the corresponding time windows. The long-term channel processes the extracted 6-hour strain data segments, calculating the mean strain, cumulative strain change, and strain change slope within the data segments to extract strain cumulative trend characteristics.

[0051] The short-term channel processes the captured 1-minute acoustic emission data segment, calculates the peak acoustic emission signal, acoustic emission signal frequency, and acoustic emission signal energy value within the data segment, thereby extracting the instantaneous characteristics of acoustic emission.

[0052] The long-term channel uses a sliding window every hour to continuously update the cumulative strain trend features. The short-term channel uses a sliding window every 10 seconds to capture the instantaneous acoustic emission features in real time. The features extracted from both channels are normalized to the range of 0 to 1 to eliminate dimensional differences and prepare for subsequent feature fusion.

[0053] The method involves setting the weight of long-term features higher than that of short-term features, calculating the rate of change of long-term features, and when the rate of change stabilizes, increasing the weight of long-term features and decreasing the weight of short-term features to filter out features in the short-term channel that are below the noise threshold. Specifically, this is done by first setting initial weights for long-term and short-term features. The initial weight of long-term features is set to 0.7, and the initial weight of short-term features is set to 0.3, ensuring that the weight of long-term features is higher than that of short-term features.

[0054] Next, the long-term characteristic change rate is calculated. Based on the strain accumulation trend characteristics updated hourly in the long-term channel, the strain change slope is selected as the calculation basis. The difference in strain change slope between two adjacent sliding windows is calculated, and then divided by the time interval of 1 hour between the two windows to obtain the long-term characteristic change rate. Then, it is determined whether the long-term characteristic change rate is stable.

[0055] The threshold for determining the stability of the rate of change is set to 0.05. When the absolute value of the rate of change of the long-term feature corresponding to three consecutive sliding windows is less than 0.05, the rate of change of the long-term feature is determined to be stable.

[0056] When the rate of change of long-term features stabilizes, adjust the feature weights. For example, increase the weight of long-term features to 0.85 while decreasing the weight of short-term features to 0.15. Then filter the short-term channel features. First, set noise thresholds for the acoustic emission features of the short-term channel: the noise threshold for the peak acoustic emission signal is set to 1.2 times the peak initial background noise under crack-free conditions; the noise threshold for the frequency of the acoustic emission signal is set to once every 10 seconds; and the noise threshold for the energy value of the acoustic emission signal is set to 1.2 times the energy value of the initial background noise under crack-free conditions. Filter out the instantaneous acoustic emission features in the short-term channel that are below the corresponding noise thresholds, retaining only the valid short-term features that are above the noise thresholds.

[0057] The process of fusing weighted long-term features with effective short-term features and inputting them into a pre-trained model to obtain a prediction result of the crack state in a predetermined future period involves, in practice, first concatenating the weighted long-term features with the effective short-term features. Both types of features have been normalized, and the concatenated feature vector is then formed.

[0058] The dimension of the fused feature vector must match the dimension of the input layer of the pre-trained model to ensure that the data can be correctly input into the model. The pre-trained model is an improved LSTM model with an added attention mechanism. After the fused feature vector is input into the improved LSTM model, the model will use the attention mechanism to assign weights to key features affecting crack propagation in the fused features. These key features include the cumulative strain change in the long-term channel and the frequency of acoustic emission signals in the short-term channel. The future predetermined period is set to 24 hours. Based on the input fused features, the model will output the crack state prediction results for the next 24 hours.

[0059] The prediction results include hourly crack width forecasts for the next 24 hours, which will form a complete crack width variation curve. The model will also output the corresponding crack risk level for each hour within the next 24 hours. Furthermore, the model will predict the potential development stage of the crack within the next 24 hours.

[0060] During the prediction process, the model incorporates environmental factor correction terms. By combining real-time collected temperature and humidity data, error correction is applied to the prediction results. This correction keeps the prediction error below 8%, ensuring the accuracy of the predictions.

[0061] In practice, an improved LSTM model is used as a pre-trained model to adapt to the overall process of monitoring cracks in solid waste aggregate concrete. The model is based on the standard LSTM framework, retaining the core gating mechanisms of input gate, forget gate, and output gate. This mechanism avoids the gradient vanishing or exploding problems of traditional RNNs, adapting to the processing requirements of time-series monitoring data. The input layer dimension of the model must be consistent with the dimension of the fused feature vector. The fused feature vector is formed by concatenating weighted long-term features with effective short-term features. Long-term features include the mean strain, cumulative strain change, and strain change slope. Short-term features include the peak value of acoustic emission signals, the frequency of acoustic emission signals, and the energy value of acoustic emission signals. Therefore, the input layer dimension is set to 6 dimensions.

[0062] The model incorporates an attention mechanism to assign weights to key features influencing crack propagation. These key features include the cumulative strain change in the long-term channel and the frequency of acoustic emission signals in the short-term channel. These two types of features are most strongly correlated with the initiation and propagation of concrete cracks. The attention mechanism strengthens their influence on the prediction results while weakening irrelevant features.

[0063] The model has two hidden layers, each containing 64 LSTM neurons. This setup balances the need for in-depth mining of temporal features with the computational limitations of edge computing nodes. It avoids computational delays caused by overly complex models and adapts to the real-time data processing requirements of monitoring systems.

[0064] The model's training data comes from measured monitoring data of solid waste aggregate concrete. The data covers the entire life cycle of concrete, including no cracks, micro-cracks, medium cracks, and wide cracks. The training data must be labeled with corresponding actual crack state parameters. The labeling includes actual crack width, crack risk level, and crack development stage. Crack risk levels are categorized by width as low risk, medium risk, and high risk. Crack development stages are categorized by width growth rate as initial stage, propagation stage, and critical stage.

[0065] The model was trained using the Adam optimizer. The initial learning rate was set to 0.001. After every 100 training epochs, the learning rate was reduced to 1 / 10 of its original value to balance training speed and convergence accuracy.

[0066] When the result meets the alarm conditions, a real-time alarm is triggered. Specifically, three types of alarm trigger conditions are first set. The first type is when the predicted crack width exceeds 0.3mm. The second type is when the predicted crack risk level reaches high risk. The third type is when the predicted crack development stage enters stage 3.

[0067] An alarm process is initiated when the prediction result of any hour within the next 24 hours output by the improved LSTM model meets any of the above conditions.

[0068] Edge computing nodes first synchronize alarm information to the cloud server. The alarm information includes the location of the crack monitoring, corresponding to the specific area where the sensors are deployed, including the locations of pre-embedded sensors in areas of stress concentration within the concrete and the locations of surface-mounted sensors in areas prone to cracking. The alarm information also includes the currently predicted crack width value, the crack width trend over the next 24 hours, the current corresponding crack risk level, and the specific reason for triggering the alarm, such as the crack width exceeding 0.3mm or the risk level reaching high risk.

[0069] After receiving the alarm information, the cloud server initiates multi-channel notifications. For the PC client in the remote monitoring layer, an alarm window pops up on the monitoring interface. The alarm window displays the complete alarm information and remains open until staff manually confirms it. For mobile devices, an alarm push notification is sent to the devices of the linked project manager and maintenance personnel. The push notification includes the crack location and current risk level. Simultaneously, an alarm SMS is sent to the mobile phones of the project manager and maintenance personnel. The SMS contains the crack monitoring location, the current predicted crack width, the risk level, and a link to view detailed information.

[0070] If audible and visual alarm devices are deployed on-site, the cloud server will send control commands to the edge computing nodes. Upon receiving the commands, the edge computing nodes will trigger the on-site audible and visual alarm devices to activate. The audible and visual alarm devices will continuously emit signals until personnel arrive on-site to confirm and disable the alarm through the system.

[0071] Each time an alarm is triggered, the cloud server automatically records an alarm log. The alarm log includes the alarm trigger time, the location of the crack being monitored, the predicted parameters that triggered the alarm (such as the predicted crack width and risk level), the list of personnel who received the alarm notification, and the alarm confirmation time. The alarm log is stored permanently on the cloud server for easy retrieval and analysis.

[0072] This application also discloses a real-time monitoring system for cracks in solid waste aggregate concrete, such as... Figure 2 ,include: Sensors are used for initial baseline data acquisition and real-time monitoring data acquisition of crack-free solid waste aggregate concrete. The benchmark management unit is used to complete the initial benchmark data import, periodic updates and edge-cloud synchronization, providing an accurate benchmark for deviation calculation; The deviation calculation and compensation unit is used to calculate the deviation between the monitoring data and the benchmark in real time, and to perform data compensation when the deviation exceeds the threshold to correct the sensor drift error. The feature-channel processing unit is used to extract and filter features from the compensated data, separate key features from interference information, and provide high-quality input for the prediction model. The model prediction unit is used to predict crack trends based on fused features and output the crack status in the future period; the alarm management unit is used to monitor the prediction results, trigger alarms and notify through multiple channels to ensure timely risk response; the data storage and backtracking unit is used to store full-process data, support historical data query and system optimization analysis, and improve system maintainability.

[0073] This application also provides an embodiment of an electronic device. The electronic device is manifested in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: one or more processors or processing units, memory, and buses connecting different components (including memory and processing units).

[0074] A bus refers to one or more of several bus architectures, including memory buses or memory controllers, peripheral buses, graphics acceleration ports, processors, or local buses using any of the various bus architectures. Examples of these architectures include, but are not limited to, Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MCA) buses, Enhanced ISA buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0075] Electronic devices typically include a variety of computer-readable media. These media can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, and removable and non-removable media.

[0076] The memory may include computer-readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. Electronic devices may further include other removable / non-removable, volatile / non-volatile computer device storage media. By way of example only, the storage system may be used to read and write non-removable, non-volatile magnetic media.

[0077] The electronic device can also communicate with one or more external devices (e.g., keyboard, pointing device, camera, etc.), may include a display, and may communicate with one or more devices that enable a user to interact with the electronic device, and / or with any device that enables the electronic device to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via an input / output (I / O) interface. Furthermore, the electronic device can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN)) and / or public networks, such as the Internet) via a network adapter. The network adapter communicates with other modules of the electronic device via a bus. The processor executes various functional applications and data processing by running programs stored in memory, such as implementing the real-time monitoring method for cracks in solid waste aggregate concrete provided in the above embodiments of the present invention.

[0078] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a real-time monitoring method for cracks in solid waste aggregate concrete as provided in the embodiments of the present invention.

[0079] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, system, or device.

[0080] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, system, or device.

[0081] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time monitoring of cracks in solid waste aggregate concrete, characterized in that, Including the following steps: Import the initial baseline data of the sensor in the crack-free state of solid waste aggregate concrete; Real-time acquisition of sensor data; Calculate the deviation between the current data and the initial baseline data; When the deviation value exceeds the threshold, the deviation value is used to compensate for the subsequent sensor data. The initial baseline data is updated using the mean of the data after compensation in the previous period according to the set period. The compensated data is processed by channel. The long-term channel is used to extract the cumulative strain trend features with a large time window, while the short-term channel is used to extract the instantaneous acoustic emission features with a small time window. Set the weight of long-term features higher than that of short-term features, calculate the rate of change of long-term features, and when the rate of change is stable, increase the weight of long-term features and decrease the weight of short-term features to filter out features in the short-term channel that are below the noise threshold. The weighted long-term features are fused with the effective short-term features and input into the pre-trained model to obtain the crack state prediction results for the future predetermined period. When the result meets the alarm conditions, a real-time alarm is triggered.

2. The method for real-time monitoring of cracks in solid waste aggregate concrete according to claim 1, characterized in that, The process of importing initial reference data from sensors in a crack-free state of solid waste aggregate concrete specifically includes: selecting a crack-free state after the solid waste aggregate concrete has been poured and reached its design strength for initial reference data acquisition; the acquisition objects include strain sensors, acoustic sensors, temperature and humidity sensors, and image sensors; acquiring corresponding stable reference data for each type of sensor in a crack-free state; the acquired initial reference data is first transmitted to an edge computing node for preprocessing; the preprocessed initial reference data is synchronously stored in a cloud server as a reference for subsequent real-time data comparison and deviation calculation.

3. The method for real-time monitoring of cracks in solid waste aggregate concrete according to claim 1, characterized in that, The real-time acquisition of sensor data specifically includes: each sensor synchronously acquiring corresponding data at a set frequency; the acquisition range includes sensors embedded in stress concentration areas inside the concrete and sensors attached to easily cracked parts of the concrete surface; the acquired data is transmitted to the edge computing node in real time via wireless communication.

4. The method for real-time monitoring of cracks in solid waste aggregate concrete according to claim 1, characterized in that, The calculation of the deviation between the current data and the initial reference data specifically includes: after receiving the data collected in real time by each sensor, the edge computing node calculates the deviation value from the corresponding initial reference data according to the sensor type; the strain sensor subtracts the initial stable strain value under the crack-free state from the real-time collected strain value to obtain the strain deviation value; the acoustic sensor subtracts the initial background noise value from the real-time collected acoustic emission signal intensity to obtain the acoustic signal deviation value; the temperature sensor subtracts the initial stable temperature value from the real-time collected temperature value to obtain the temperature deviation value; the humidity sensor subtracts the initial stable humidity value from the real-time collected humidity value to obtain the humidity deviation value; the image sensor calculates the image feature deviation value by comparing the gray value changes and edge feature differences between the real-time collected image and the initial crack-free image sequence; after all deviation values ​​are calculated, they are temporarily stored in the local cache of the edge computing node for subsequent compensation judgment.

5. The method for real-time monitoring of cracks in solid waste aggregate concrete according to claim 1, characterized in that, When the deviation value exceeds the threshold, the deviation value is used to compensate for the subsequent sensor data. Specifically, this includes: when the deviation value calculated by a certain type of sensor exceeds the corresponding threshold, the deviation value is used to compensate for the subsequent sensor data of that type; the compensated sensor data is temporarily stored in the local cache of the edge computing node and simultaneously uploaded to the cloud server for backup; the deviation of the subsequently collected data is continuously monitored during the compensation process; if the deviation value collected multiple times returns to the threshold range, the current compensation operation is stopped and the normal data acquisition mode is restored; if the deviation value still exceeds the threshold after compensation, a sensor calibration prompt is triggered to remind the staff to check the working status of the sensor.

6. The method for real-time monitoring of cracks in solid waste aggregate concrete according to claim 1, characterized in that, The step of updating the initial baseline data with the average of the compensated data from the previous period according to a set period specifically includes: setting an update period for the initial baseline data, and automatically triggering the update operation after each period ends; the edge computing node obtains all sensor data that has been compensated in the previous period, and filters out abnormal data during the acquisition. If the proportion of effective compensated data for a certain type of sensor in the previous period does not meet the standard, the update of the baseline data for that type of sensor is paused, the original initial baseline data is used, and a sensor data abnormality prompt is issued at the same time; for sensor categories with a sufficient amount of effective data, the average of the compensated data from the previous period is calculated for each category; after the calculation is completed, the initial baseline data stored locally on the edge computing node is updated with the average values ​​corresponding to each type of sensor, and the updated baseline data for each type of sensor is simultaneously uploaded to the cloud server to replace the original stored initial baseline data.

7. The method for real-time monitoring of cracks in solid waste aggregate concrete according to claim 1, characterized in that, The process of processing the compensated data by channel is as follows: the long-term channel extracts strain accumulation trend features using a large time window, while the short-term channel extracts acoustic emission instantaneous features using a small time window. Specifically, this includes: first, determining the time window parameters for the long-term and short-term channels; edge computing nodes acquiring the compensated data from both channels and extracting data segments according to the corresponding time windows; the long-term channel processing the extracted data segments to extract strain accumulation trend features and continuously updating these features using a sliding window method; the short-term channel processing the extracted data segments to extract acoustic emission instantaneous features and capturing these features in real time using a sliding window method; and normalizing the features extracted from both channels to eliminate dimensional differences.

8. The method for real-time monitoring of cracks in solid waste aggregate concrete according to claim 1, characterized in that, The process of fusing weighted long-term features with effective short-term features and inputting them into a pre-trained model to obtain a crack state prediction result for a predetermined future period includes: concatenating the weighted long-term features with effective short-term features; normalizing the two types of features to form a fused feature vector, with the vector dimension matching the dimension of the input layer of the pre-trained model; the pre-trained model is an improved LSTM model with an added attention mechanism; after inputting the fused feature vector into the model, the model assigns key weights to the critical features affecting crack expansion through the attention mechanism; the model outputs a crack state prediction result for a predetermined future period based on the input fused features, including the predicted crack width and its change curve within the period, the crack risk level at each time period, and the possible development stage of the crack.

9. The method for real-time monitoring of cracks in solid waste aggregate concrete according to claim 1, characterized in that, When the result meets the alarm conditions, a real-time alarm is triggered, specifically including: setting multiple types of alarm trigger conditions; initiating the alarm process when the prediction result of any time period within a predetermined future period output by the pre-trained model meets any condition; the edge computing node synchronizes alarm information including crack monitoring location, predicted crack width and change trend, risk level, and triggering reason to the cloud server; the cloud server initiates multi-channel notifications: for the PC terminal of the remote monitoring layer, a window displaying complete alarm information pops up on the monitoring interface and remains until the staff manually confirms; for the mobile terminal, an alarm push containing crack location and risk level is sent to the bound personnel's device, and an alarm SMS containing crack monitoring location, predicted crack width, risk level, and a link to view detailed information is sent to relevant personnel; after each alarm is triggered, the cloud server automatically records an alarm log containing alarm trigger time, crack monitoring location, prediction parameters that triggered the alarm, a list of alarm notification recipients, and alarm confirmation time.

10. A real-time monitoring system for cracks in solid waste aggregate concrete, characterized in that, include: Sensors are used for initial baseline data acquisition and real-time monitoring data acquisition of crack-free solid waste aggregate concrete. The benchmark management unit is used for initial benchmark data import, periodic updates, and edge-cloud synchronization. The deviation calculation and compensation unit is used to calculate the deviation between the monitoring data and the benchmark in real time, and to perform data compensation when the deviation exceeds the threshold to correct the sensor drift error. The feature-segmentation processing unit is used to extract and filter features from the compensated data, separating features from interference information. The model prediction unit is used to predict crack trends based on fused features and output the crack status in future cycles. The alarm management unit is used to monitor prediction results, trigger alarms, and notify users through multiple channels.

Images