Prestressed structure health monitoring and early warning method based on multi-source data fusion

Abstract

The invention provides a multi-source data fusion prestressed structure health monitoring and early warning method, and relates to the technical field of constructional engineering structure health monitoring, and the method comprises the steps: carrying out the processing of a self-adaptive weighted data fusion model based on monitoring data obtained by a sensor, so as to output a prestressed comprehensive loss rate, the monitoring data comprises axial strain field data of the steel strand, and displacement and tensioning force of an anchorage device clamping piece; based on the historical sequence, an LSTM prediction model is improved to output a future trend prediction value; based on the future trend prediction value, the prestress comprehensive loss rate, the anchorage device clamping piece displacement and the tensioning force, performing evaluation through a failure probability model based on a fuzzy Petr network to obtain a real-time failure probability; and comparing and judging the future trend prediction value and the real-time failure probability with a preset threshold value to generate early warning information.

Description

Technical Field

[0001] The present application belongs to the technical field of structural health monitoring of construction engineering, and specifically relates to a prestressed structure health monitoring and early warning method based on multi-source data fusion. Background Art

[0002] Currently, prestressed tension monitoring primarily relies on single-point sensors (such as strain gauges and vibrating wires), fiber Bragg gratings (FBGs), and IoT cloud platforms, enabling the initial digital acquisition of parameters such as tension force and displacement. As building structures grow increasingly complex and large, traditional monitoring methods are no longer able to meet the growing safety demands. The industry is accelerating its transition toward intelligent and lightweight systems, breaking through the limitations of traditional monitoring thresholds through multimodal sensor fusion (strain / acoustic emission / temperature), real-time edge computing analysis, and deep learning predictive models.

[0003] The existing technologies mainly include the following:

[0004] Local monitoring using strain gauges or vibrating wire sensors presents data silos, with different sensor systems operating independently and lacking a unified communication protocol. While highly accurate, these systems are complex to install, require specialized personnel, and are expensive. While they achieve preliminary digital data collection, they suffer from a high false alarm rate (>40%), false alarms caused by environmental vibration, and a lack of AI-based predictive maintenance capabilities. Summary of the Invention

[0005] This application provides a method to solve one of the above technical problems.

[0006] The technical solutions adopted in this application are:

[0007] The present application provides a multi-source data fusion prestressed structure health monitoring and early warning method, including:

[0008] The monitoring data acquired by the sensors is processed through an adaptive weighted data fusion model to output a comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip, and the tension force.

[0009] Based on historical sequences, the LSTM prediction model is improved to output future trend prediction values;

[0010] Based on the future trend prediction value, the prestressed composite loss rate, the anchor clip displacement and the tension force, a failure probability model based on a fuzzy Petri net is used to evaluate and obtain a real-time failure probability;

[0011] The future trend prediction value and the real-time failure probability are compared and judged with a preset threshold value to generate early warning information.

[0012] According to one embodiment of the present application, the monitoring data acquired based on the sensor is processed by an adaptive weighted data fusion model to output the comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip, and the tension force, specifically:

[0013] The distributed optical fiber sensor, laser displacement meter and hydraulic jack are used to obtain the axial strain field data of the steel strand, the displacement of the anchor clip and the tension force respectively.

[0014] Dynamically update weights through the entropy weight method, and optimize the weighting scheme by considering the information entropy of different sensors;

[0015] The output is the comprehensive loss rate of prestressing.

[0016] According to one embodiment of the present application, the output of future trend prediction values ​​by improving the LSTM prediction model based on the historical sequence is specifically as follows:

[0017] Based on the historical sequences in the monitoring data, the temporal attention mechanism is used to calculate the hidden layer weights, and a regularization term is introduced to prevent overfitting in order to output future trend prediction values.

[0018] According to one embodiment of the present application, the future trend prediction value, the prestressed comprehensive loss rate, the anchor clip displacement and the tension force are evaluated by a failure probability model based on a fuzzy Petri net to obtain a real-time failure probability, specifically:

[0019] Receive future trend forecast values ​​of prestressed composite loss rate, anchor clip displacement and tension force;

[0020] A proposition set is defined, wherein the proposition set includes a displacement index, a stress index, and an acoustic emission energy index, wherein the displacement index indicates whether the displacement exceeds a limit, the stress index indicates whether the stress drops suddenly, and the acoustic emission energy index indicates whether the acoustic emission energy increases suddenly;

[0021] The failure probability in different situations is evaluated according to the pre-set fuzzy production rule base to obtain the real-time failure probability.

[0022] According to one embodiment of the present application, the future trend prediction value and the real-time failure probability are compared with a preset threshold value to generate early warning information, specifically:

[0023] comparing the future trend prediction value with a first preset threshold, and generating an early warning signal if the future trend prediction value is greater than the first preset threshold;

[0024] Comparing the real-time failure probability with a second preset threshold, and generating an emergency response mechanism if the real-time failure probability is greater than the second preset threshold;

[0025] The early warning signal and the emergency response mechanism are combined to generate the warning information.

[0026] A multi-source data fusion prestressed structure health monitoring and early warning system, including:

[0027] A weighted fusion module is used to process the monitoring data acquired by sensors through an adaptive weighted data fusion model to output a comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip, and the tension force.

[0028] The prediction module is used to output future trend prediction values ​​based on historical sequences by improving the LSTM prediction model;

[0029] An evaluation calculation module is used to evaluate the future trend prediction value, the prestressed composite loss rate, the anchor clip displacement and the tension force through a fuzzy Petri net-based failure probability model to obtain a real-time failure probability;

[0030] The judgment module is used to compare and judge the future trend prediction value and the real-time failure probability with a preset threshold value to generate early warning information.

[0031] According to one embodiment of the present application, the weighted fusion module is specifically:

[0032] The distributed optical fiber sensor, laser displacement meter and hydraulic jack are used to obtain the axial strain field data of the steel strand, the displacement of the anchor clip and the tension force respectively.

[0033] Dynamically update weights through the entropy weight method, and optimize the weighting scheme by considering the information entropy of different sensors;

[0034] The output is the comprehensive loss rate of prestressing.

[0035] According to one embodiment of the present application, the prediction module is specifically:

[0036] Based on the historical sequences in the monitoring data, the temporal attention mechanism is used to calculate the hidden layer weights, and a regularization term is introduced to prevent overfitting in order to output future trend prediction values.

[0037] A computer-readable storage medium stores a program, which implements the steps of the method when executed by a processor.

[0038] An electronic device comprises a memory, a processor and a program stored in the memory and runnable on the processor, wherein the processor implements the steps in the method when executing the program.

[0039] Due to the adoption of the above technical solution, the beneficial effects achieved by this application are as follows:

[0040] This application uses distributed fiber optic sensors, laser displacement meters and hydraulic jack sensors to obtain the axial strain field data of steel strands, the displacement of anchor clips and the tension force, and uses the entropy weight method to dynamically update the weights and optimize the weighting scheme, thereby overcoming the contradiction between the spatial resolution and measurement accuracy of a single sensor and improving the monitoring accuracy.

[0041] This application utilizes an improved LSTM prediction model combined with a temporal attention mechanism and regularization terms to prevent overfitting. It can accurately capture long-term dependencies and predict future trends in prestressed comprehensive loss rates, providing a forward-looking assessment of future risks and avoiding the limitations of traditional methods that can only reflect the current status.

[0042] This application defines a proposition set and uses a fuzzy production rule library to evaluate the failure probability under different circumstances. Combining the trend prediction value of the LSTM prediction model and real-time monitoring data, the probability of structural failure is calculated in real time, which improves the ability to quickly respond to immediate abnormal situations and reduces the false alarm rate.

[0043] This application generates an early warning signal by comparing the future trend prediction value with a first preset threshold, and generates an emergency response mechanism by comparing the real-time failure probability with a second preset threshold, thereby ensuring timely identification and rapid response to potential risks, improving the reliability and safety of the overall system, and helping the engineering team plan preventive maintenance in advance to avoid major accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0045] Figure 1 A flowchart of a multi-source data fusion prestressed structure health monitoring and early warning method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] In order to more clearly illustrate the overall concept of the present application, a detailed description is given below in an illustrative manner in conjunction with the accompanying drawings.

[0047] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application may also be implemented in other ways than those described herein, and therefore, the scope of protection of the present application is not limited by the specific embodiments disclosed below. It should be noted that the embodiments of the present application and the features of each embodiment may be combined with each other unless there is a conflict.

[0048] In this application, unless otherwise expressly specified and limited, a first feature "above" or "below" a second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments or examples.

[0049] Example 1

[0050] like Figure 1 As shown in FIG, a multi-source data fusion prestressed structure health monitoring and early warning method includes:

[0051] The monitoring data acquired by the sensors is processed through an adaptive weighted data fusion model to output the comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip and the tension force.

[0052] As mentioned above, the monitoring data acquired by the sensors is processed using an adaptive weighted data fusion model to output the comprehensive prestress loss rate. The monitoring data here primarily consists of three components: the axial strain field of the strand (ε(x, t)), the displacement of the anchor clip (Δd(t)), and the tension force (F(t)). This raw data comes from a variety of sensors—distributed fiber optic sensors, laser displacement meters, and hydraulic jack sensors.

[0053] First, distributed fiber optic sensors are used to collect axial strain field data of the steel strands, providing information about the internal deformation of the steel strands; laser displacement meters are used to accurately measure the displacement changes of the anchor clips relative to their initial positions; and hydraulic jack sensors record the magnitude of the tension applied to the steel strands during construction.

[0054] Next, this raw data is fed into an adaptive weighted data fusion model. This model uses an entropy weighting method to dynamically adjust the weights of different sensor data, optimizing the weight assignment w_i(t) based on the information entropy E_i(t) provided by each sensor type. This is done to better reflect the importance of each sensor data to the overall assessment and improve the accuracy of the final calculated prestressing loss rate η(t).

[0055] Finally, after the above steps, the system will output a comprehensive indicator - the comprehensive prestress loss rate η(t), which reflects the health status of the current structure in terms of prestress and can be used to further analyze whether there are potential risks or problems.

[0056] For example, consider a large-scale bridge construction project where engineers need to monitor the condition of prestressed steel strands within the bridge's main beam in real time. To do this, they deploy distributed fiber optic sensors, laser displacement meters, and hydraulic jack sensors to collect data on the strand's axial strain field, anchor clip displacement, and tension, respectively.

[0057] One day, during routine inspection, the system received the following data:

[0058] Distributed fiber optic sensors revealed a slight but increasing strain change in a section of the strand.

[0059] The laser displacement meter reported a small but significant displacement of the anchor clip attached to that section of strand.

[0060] The hydraulic jack sensor registered a slight decrease in the tension applied to this section of strand.

[0061] This data was fed into an adaptive weighted data fusion model for processing. Based on historical data and current conditions, the model dynamically adjusted the weights of each sensor data point, calculating a combined prestress loss rate, η(t), of 8% for this time period. This value exceeded the pre-set safety threshold of 7%, triggering an early warning signal. This enabled the project team to take timely measures, such as recalibrating tensioning equipment and inspecting and reinforcing relevant areas, thereby avoiding potentially more serious structural problems. This demonstrates how multi-source sensor data, combined with intelligent algorithms, can be used to effectively monitor and provide early warnings about the health of prestressed concrete structures.

[0062] Based on historical sequences, the LSTM prediction model is improved to output future trend prediction values.

[0063] For example, based on historical data, an improved LSTM prediction model can be used to output future trend forecasts. The "historical data" here refers to the time series of data collected from multiple sensors, including historical records of parameters such as the prestressing loss rate η(t), anchor clip displacement Δd(t), and tension force F(t). This data reflects the state changes of the structure over time.

[0064] The improved LSTM (Long Short-Term Memory) prediction model is a deep learning algorithm particularly well-suited for processing time series data. It can capture long-term dependencies in data and predict future trends based on past data patterns. In this application, the LSTM model was used to analyze the state changes of prestressed concrete structures, specifically the trend of prestress loss.

[0065] The specific steps are as follows:

[0066] Data preparation: Obtain raw monitoring data from distributed fiber optic sensors, laser displacement meters, and hydraulic jack sensors, and construct a historical sequence {X(t-τ),...,X(t)}, where the data X(t) at each time point t is a vector containing η(t), Δd(t), and F(t).

[0067] Model training: Use historical data to train the LSTM model and adjust the network weights so that the model can accurately simulate the changing patterns of the structural state.

[0068] Prediction output: Use the trained LSTM model to predict η(t) in the future, that is, calculate the future development trend of the prestressed comprehensive loss rate.

[0069] In addition, the model also introduces a temporal attention mechanism to optimize the hidden layer weights and adds a regularization term to prevent overfitting, thereby improving prediction accuracy.

[0070] For example, in a real-world case, suppose a bridge is undergoing regular maintenance inspections. Engineers hope to proactively identify potential risks through an intelligent monitoring system. To this end, they deploy distributed fiber optic sensors, laser displacement meters, and hydraulic jack sensors to continuously monitor the axial strain field ε(x, t) of the steel strands, the displacement Δd(t) of the anchor clips, and the tension force F(t).

[0071] After several months of data accumulation, the system has collected sufficient historical series data. Now, engineers want to understand the prestress loss of the steel strands inside the bridge's main beams over the next three months. Therefore, they input this historical data into the improved LSTM predictive model for analysis.

[0072] After processing this data, the LSTM model predicted that the combined prestressing loss rate (η(t)) in a specific area would gradually increase over the next three months, reaching approximately 10%. While this value has not yet reached a dangerous threshold (e.g., 12%), it is close to the warning line, suggesting the need to increase monitoring frequency and consider preventive measures.

[0073] Based on this prediction, the maintenance team decided to conduct more frequent inspections of the area and scheduled an additional on-site assessment. They also developed a contingency plan to prevent further prestressing losses. In this way, the LSTM predictive model helped the engineering team identify potential risks in advance, ensuring the safe operation of the bridge.

[0074] Based on the future trend prediction value, the prestressed comprehensive loss rate, the anchor clip displacement and the tensioning force, an evaluation is performed through a failure probability model based on a fuzzy Petri net to obtain a real-time failure probability.

[0075] As described above, based on the predicted future trend (provided by the LSTM model), the prestressed concrete structure's comprehensive loss rate η(t), the anchor clip displacement Δd(t), and the tension force F(t), a failure probability model based on fuzzy Petri nets is used to evaluate the real-time failure probability P_fail(t). This process aims to utilize multi-source data fusion and intelligent analysis technology to conduct real-time monitoring and risk assessment of the safety status of prestressed concrete structures.

[0076] Input data preparation: First, the system obtains trend prediction values ​​from the LSTM prediction model and real-time monitoring data (including η(t), Δd(t), and F(t)). These data reflect the current state of the structure and its possible future changes.

[0077] Define a proposition set: In the fuzzy Petri net model, define a proposition set P, which contains several key indicators:

[0078] p_1: Indicates whether the displacement is out of limit (for example, beyond the normal range).

[0079] p_2: Indicates whether the stress drops suddenly (e.g., drops suddenly).

[0080] p_3: Indicates whether the acoustic emission energy suddenly increases (for example, an increase in the acoustic emission signal intensity may indicate crack propagation inside the material).

[0081] Fuzzy production rule base: Evaluate the failure probability under different circumstances based on the preset rule base. The specific rules are as follows:

[0082] IF p_1AND p_2THEN Anchor slip (confidence β=0.92)

[0083] IF p_2OR p_3THEN The strand breaks (confidence β = 0.87)

[0084] Calculating real-time failure probability: Based on the aforementioned rules and the currently monitored data, the system calculates the probability of structural failure, P_fail(t), in real time. This step involves matching the data from each sensor with pre-defined rules and deriving a specific failure probability value through fuzzy inference.

[0085] Output: Ultimately, the system outputs a real-time failure probability, P_fail(t), to determine whether the current structure has a high risk of failure. If P_fail(t) exceeds a preset threshold (e.g., 0.7), an emergency response mechanism is activated to alert relevant personnel to the potential risk and take necessary preventive measures.

[0086] For example, in a real-world case, let's assume a large construction project is using this intelligent monitoring system to monitor the health of its prestressed concrete structure. Recently, the system has collected the following data:

[0087] The distributed optical fiber sensor shows that the axial strain field data ε(x,t) of the steel strand has a slight but continuous increasing trend.

[0088] The laser displacement meter reports that the displacement Δd(t) of the anchor clip relative to the initial position has increased and is close to exceeding the limit.

[0089] The hydraulic jack sensor recorded a slight decrease in the tension force F(t) applied to the steel strand.

[0090] The LSTM prediction model predicts that the comprehensive loss rate of prestressing η(t) may reach about 10% in the next month.

[0091] These data are fed into a failure probability model based on fuzzy Petri nets for processing. Based on the defined proposition set P, the model finds:

[0092] p_1: The displacement Δd(t) is close to exceeding the limit.

[0093] p_2: The tension force F(t) shows signs of a sudden drop.

[0094] p_3: No significant increase in acoustic emission energy was observed.

[0095] According to the rules in the fuzzy production rule base:

[0096] IF p_1AND p_2THEN the anchor slips, since the displacement is close to exceeding the limit and the tension force drops sharply, the model calculates that the possibility of anchor slip is high, with a confidence level of 0.92.

[0097] If the p_2OR p_3THEN strand breaks, although the acoustic emission energy does not increase significantly, the model calculates the possibility of strand breakage due to the sudden drop in tension, with a confidence level of 0.87.

[0098] After fuzzy inference calculations, the system determined the real-time probability of failure (P_fail(t)) of the current structure to be 0.85, exceeding the preset threshold of 0.7. Consequently, the system immediately triggered an emergency response mechanism, alerting on-site management to potential risks and recommending further inspection and reinforcement measures. This timely warning helped the engineering team avoid potential serious structural failure and ensured construction safety.

[0099] The future trend prediction value and the real-time failure probability are compared and judged with a preset threshold value to generate early warning information.

[0100] As mentioned above, the future trend prediction and real-time failure probability are compared with the preset threshold to generate warning information. This process is a crucial step in the entire intelligent monitoring system, as it determines when to issue a warning and the level of the warning. The specific steps are as follows:

[0101] Obtaining predicted values ​​and failure probabilities: First, obtain the trend prediction value of the prestressed comprehensive loss rate η(t) in the future period from the LSTM prediction model, and calculate the real-time failure probability P_fail(t) of the current structure through the failure probability model based on the fuzzy Petri net.

[0102] Setting thresholds: Based on engineering safety standards and historical data, two key thresholds are pre-set:

[0103] The design threshold of the comprehensive loss rate of prestressing force η(t) (e.g. 12%) is used to assess the risk of long-term prestressing force loss.

[0104] The warning threshold value (eg, 0.7) of the real-time failure probability P_fail(t) is used to determine the possibility of failure occurring in the short term.

[0105] Comparative judgment:

[0106] For η(t): If the predicted prestressed comprehensive loss rate η(t) exceeds the design threshold, an early warning signal is triggered, indicating that attention should be paid to potential risks in the long-term trend.

[0107] For P_fail(t): If the real-time failure probability P_fail(t) exceeds the warning threshold, the emergency response mechanism is activated, indicating that there is a high risk of immediate failure and immediate action is required.

[0108] Generate warning information: Based on the comparison results, the system automatically generates corresponding warning information. This information not only includes specific values ​​(such as the specific percentage of η(t) or the specific value of P_fail(t)), but also includes recommended action guidelines to guide relevant personnel on how to deal with the current situation.

[0109] For example, in a real-world case, let’s assume a bridge is using this intelligent monitoring system to monitor the safety of its prestressed concrete structure. Recent data analysis shows:

[0110] The LSTM prediction model predicts that the comprehensive loss rate of prestressing η(t) will reach 13% in the next month, exceeding the preset design threshold of 12%.

[0111] The failure probability model based on fuzzy Petri nets calculates that the real-time failure probability P_fail(t) of the current structure is 0.75, which also exceeds the preset warning threshold of 0.7.

[0112] Comparative judgment process

[0113] For η(t): Because the predicted prestress loss rate η(t) reached 13%, exceeding the design threshold of 12%, the system triggered an early warning signal. This means that although there is no immediate danger, if no measures are taken, serious prestress loss problems may occur in the future.

[0114] For P_fail(t): The real-time failure probability P_fail(t) is 0.75, exceeding the warning threshold of 0.7, indicating a high risk of failure. Therefore, the system activates the emergency response mechanism, prompting on-site management personnel to take immediate emergency measures.

[0115] Early warning information generation

[0116] Based on the above comparison results, the system generates detailed warning information, including:

[0117] Early warning signal: Inform relevant persons in charge that the comprehensive loss rate of prestressing η(t) may reach 13% in the next month, approaching a dangerous level. It is recommended to increase the monitoring frequency and consider adjusting the tensioning force or implementing temporary reinforcement measures.

[0118] Emergency Response Mechanism: On-site personnel are informed that the current real-time probability of failure of the structure is 0.75, exceeding the safety limit. All activities involving the hazardous area are immediately halted, and engineers and technicians are dispatched to the site for a detailed inspection. Clear operational instructions are also provided on how to safely evacuate or take necessary emergency measures.

[0119] In this way, the intelligent monitoring system can effectively identify potential risks and issue early warning information to relevant personnel in a timely manner, ensuring that appropriate preventive and response measures are taken in the shortest possible time, thereby protecting the safety and reliability of the structure.

[0120] According to one embodiment of the present application, the monitoring data acquired based on the sensor is processed by an adaptive weighted data fusion model to output the comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip, and the tension force, specifically:

[0121] The distributed optical fiber sensor, laser displacement meter and hydraulic jack are used to obtain the axial strain field data of the steel strand, the displacement of the anchor clip and the tension force respectively.

[0122] Dynamically update weights through the entropy weight method, and optimize the weighting scheme by considering the information entropy of different sensors;

[0123] The output is the comprehensive loss rate of prestressing.

[0124] As mentioned above, the monitoring data obtained by the sensors is processed by the adaptive weighted data fusion model to output the prestressed composite loss rate. The specific steps are as follows:

[0125] Axial strain field data (ε(x, t)) of the steel strand is obtained using distributed fiber optic sensors. These sensors can measure the strain changes at different locations on the steel strand with high precision, providing information about the internal deformation of the strand.

[0126] Anchor clip displacement (Δd(t)): Measured using a laser displacement meter. This device accurately records minute displacements of the anchor clip relative to its initial position, helping to assess the stability of the structure.

[0127] Tension force (F(t)): Provided by a hydraulic jack sensor. This sensor is used to monitor the tension applied to the strand during construction to ensure that the tensioning process meets design requirements.

[0128] Entropy Weighting Method Dynamically Updates Weights: To optimize the weighting of different sensor data, the entropy weighting method is used. This method uses the concept of entropy from information theory to measure the amount of information provided by each sensor. Lower entropy indicates more reliable and representative sensor data; lower entropy indicates greater noise or uncertainty.

[0129] Calculating information entropy E_i(t): For each type of sensor data (such as strain field data, displacement, and tension), first calculate its information entropy E_i(t). This step involves statistically analyzing the distribution of each type of data in its time series and calculating the entropy value based on this.

[0130] Dynamically adjust weights w_i(t): Based on the information entropy of each sensor data, dynamically adjust its weight w_i(t) in the final result. The weight distribution formula takes the influence of information entropy into account, so that more reliable data has a greater proportion in calculating the prestressing comprehensive loss rate. For example, if a sensor's data exhibits high information entropy (i.e., low reliability), its weight will be reduced accordingly.

[0131] After the above steps, the system combines the data from all sensors and their optimized weights to calculate a comprehensive indicator—the prestressing loss rate η(t). This value reflects the current health of the structure in terms of prestressing and can be used to further analyze potential risks or problems.

[0132] The comprehensive prestress loss rate η(t) not only takes into account the data of a single sensor, but also integrates the advantages of multi-source data through weighted fusion, thereby improving the accuracy and reliability of the structural health assessment.

[0133] Distributed fiber optic sensors, laser displacement meters, and hydraulic jacks are used to obtain data on the axial strain field of the steel strands, displacement of the anchor clips, and tension. The entropy weighting method is then used to dynamically update the weights, optimizing the weighting scheme based on the information entropy provided by the different sensors. This ensures that the quality and reliability of each data type are fully considered during the data fusion process. Ultimately, the system outputs the comprehensive prestress loss rate η(t), which provides a key parameter for structural safety assessment. This method effectively improves the accuracy and robustness of the monitoring system, helping to promptly detect and prevent potential structural failure risks.

[0134] According to one embodiment of the present application, the output of future trend prediction values ​​by improving the LSTM prediction model based on the historical sequence is specifically as follows:

[0135] Based on the historical sequences in the monitoring data, the temporal attention mechanism is used to calculate the hidden layer weights, and a regularization term is introduced to prevent overfitting in order to output future trend prediction values.

[0136] As mentioned above, we first need to extract a historical series from the monitoring data. This data includes historical records of parameters such as the prestressing loss rate η(t), the anchor clip displacement Δd(t), and the tension force F(t). For example, we can construct a time series {X(t-τ),...,X(t)}, where the data X(t) at each time point t is a vector containing η(t) (the prestressing loss rate), Δd(t) (the displacement), and F(t) (the tension force).

[0137] LSTM (Long Short-Term Memory) is a special type of recurrent neural network (RNN) that excels at processing and predicting long-term dependencies in time series data. In this application, the LSTM model is used to analyze the state changes of prestressed concrete structures and predict their future trends.

[0138] The temporal attention mechanism is used to optimize the hidden layer weights in the LSTM model. The temporal attention mechanism allows the model to dynamically adjust its attention based on the importance of different parts of the input sequence, thereby improving the learning effect of key information.

[0139] During training, regularization is introduced to prevent the model from overfitting the training data, which can lead to decreased generalization ability on new data. Regularization typically involves L1 or L2 regularization. Specifically, the inclusion of a regularization term in the loss function (λ = 0.01, where L is the number of network layers) helps control model complexity and reduce the risk of overfitting.

[0140] After the above steps, the improved LSTM model can learn the patterns in the input sequence and use them to predict η(t) over a period of time in the future. Specifically, the model generates a series of predicted values ​​of the prestressing comprehensive loss rate η(t) corresponding to future time points.

[0141] The final output is a forecast of future trends, representing the changing trend of the prestressing loss rate η(t) over a period of time. These forecasts can help engineers identify potential risks in advance, such as potential safety hazards caused by excessive prestressing loss.

[0142] An improved LSTM prediction model is used to analyze historical sequences in monitoring data. This model dynamically adjusts hidden layer weights through a temporal attention mechanism, enabling it to better capture important information in the input sequence. Furthermore, a regularization term is introduced to prevent overfitting. The trained and optimized LSTM model can output a trend forecast of the prestressed composite loss rate η(t) over a period of time. This prediction method not only improves prediction accuracy but also enhances the model's generalization ability, enabling it to more accurately warn of potential structural problems in practical application scenarios.

[0143] According to one embodiment of the present application, the future trend prediction value, the prestressed comprehensive loss rate, the anchor clip displacement and the tension force are evaluated by a failure probability model based on a fuzzy Petri net to obtain a real-time failure probability, specifically:

[0144] Receive future trend forecast values ​​of prestressed composite loss rate, anchor clip displacement and tension force;

[0145] A proposition set is defined, wherein the proposition set includes a displacement index, a stress index, and an acoustic emission energy index, wherein the displacement index indicates whether the displacement exceeds a limit, the stress index indicates whether the stress drops suddenly, and the acoustic emission energy index indicates whether the acoustic emission energy increases suddenly;

[0146] The failure probability in different situations is evaluated according to the pre-set fuzzy production rule base to obtain the real-time failure probability.

[0147] As mentioned above, the trend forecast value of the prestressed comprehensive loss rate η(t) in the future period is provided by the improved LSTM prediction model.

[0148] Current monitoring data:

[0149] Prestress comprehensive loss rate η(t): reflects the health status of the current structure in terms of prestress.

[0150] Anchor clip displacement Δd(t): The displacement change of the anchor clip relative to the initial position measured by the laser displacement meter.

[0151] Tension force F(t): The tension force applied to the steel strand recorded by the hydraulic jack sensor.

[0152] These data together form the input of the system for further analysis and assessment of potential failure risks.

[0153] Proposition Set P: Contains three key indicators to describe different situations that may lead to structural failure:

[0154] p_1 (displacement indicator): Indicates whether the displacement exceeds the limit. For example, if the displacement of the anchor clip Δd(t) exceeds the preset safety range, the indicator is considered true.

[0155] p_2 (stress index): Indicates whether there is a sudden drop in stress. If the prestress loss rate η(t) or the tension force F(t) suddenly drops, this index is considered true.

[0156] p_3 (Acoustic Emission Energy Index): Indicates whether there is a sudden increase in acoustic emission energy. Acoustic emission signals are often associated with the growth of microcracks within the material. If a significant increase in acoustic emission energy is detected, this indicator is considered true.

[0157] Fuzzy production rule base: This base evaluates the probability of failure under different circumstances based on pre-defined rules. These rules are based on engineering experience and historical data and are designed to capture the relationships between various failure modes. The specific rules are as follows:

[0158] IF p_1AND p_2THEN anchor slip (confidence β = 0.92): If it is found that the displacement exceeds the limit and the stress drops suddenly, it is inferred that anchor slip may have occurred. The confidence level of this judgment is 0.92.

[0159] IF p_2OR p_3THEN Steel Strand Broken (Confidence β=0.87): If a sudden drop in stress or a sudden increase in acoustic emission energy is found, it is inferred that the steel strand may be broken. The confidence level of this judgment is 0.87.

[0160] Real-time failure probability P_fail(t): Based on the proposition set and fuzzy production rule base defined above, the system calculates the probability of structural failure in real time. The specific process includes:

[0161] Analyze the current monitoring data and determine whether each proposition (p_1, p_2, p_3) is true.

[0162] According to the rules in the fuzzy production rule base and the status of each proposition, the probability of different failure modes is calculated.

[0163] By combining the probabilities of all failure modes, we can get a total real-time failure probability P_fail(t), which reflects the probability of failure of the current structure.

[0164] Based on future trend predictions, the combined prestress loss rate η(t), the anchor clip displacement Δd(t), and the tension force F(t), a failure probability model based on fuzzy Petri nets is used to evaluate the real-time failure probability. Data from various sensors is first received and processed, followed by a definition of a proposition set containing displacement, stress, and acoustic emission energy indicators. Next, a predefined fuzzy production rule base is used to evaluate the failure probability under different scenarios. Ultimately, the system outputs a real-time failure probability, P_fail(t), which is used to determine whether the current structure has a high risk of failure and to take appropriate preventive measures. This method can effectively improve the monitoring accuracy and early warning capabilities of the health status of prestressed concrete structures.

[0165] According to one embodiment of the present application, the future trend prediction value and the real-time failure probability are compared with a preset threshold value to generate early warning information, specifically:

[0166] comparing the future trend prediction value with a first preset threshold, and generating an early warning signal if the future trend prediction value is greater than the first preset threshold;

[0167] Comparing the real-time failure probability with a second preset threshold, and generating an emergency response mechanism if the real-time failure probability is greater than the second preset threshold;

[0168] The early warning signal and the emergency response mechanism are combined to generate the warning information.

[0169] As mentioned above, the future trend forecast value is the trend forecast value of the prestressed comprehensive loss rate η(t) in the future period provided by the improved LSTM forecast model.

[0170] Real-time failure probability P_fail(t): The probability of failure of the current structure calculated by the failure probability model based on fuzzy Petri net.

[0171] First preset threshold: This is used to compare the predicted future trend value, usually set as a safety margin, such as the design threshold of the prestressing comprehensive loss rate η(t) (e.g., 12%). This threshold helps identify long-term potential risks.

[0172] The second preset threshold is used to compare the real-time failure probability P_fail(t), which is usually set to a higher warning line, such as 0.7. This threshold helps identify high-risk events that occur in the short term.

[0173] Early warning signs:

[0174] Comparing the future trend prediction value with the first preset threshold: comparing the future trend prediction value obtained by the LSTM prediction model (eg, the prestressed comprehensive loss rate η(t) in the next month) with the first preset threshold (eg, 12%).

[0175] Generate an early warning signal: If the future trend prediction value is greater than a first preset threshold (i.e., η(t) > 12%), the system generates an early warning signal. This indicates that although there is no immediate danger, if no action is taken, a serious prestressing loss problem may occur in the future.

[0176] Emergency Response Mechanism:

[0177] Comparing the real-time failure probability with a second preset threshold: comparing the real-time failure probability P_fail(t) calculated based on the fuzzy Petri net with a second preset threshold (eg, 0.7).

[0178] Generate an emergency response mechanism: If the real-time failure probability is greater than the second preset threshold (i.e., P_fail(t)>0.7), the system activates the emergency response mechanism. This means that there is a high risk of immediate failure and immediate action is required.

[0179] Combining early warning signals and emergency response mechanisms: Based on the comparison results, the system generates corresponding warning information. This information not only includes specific values ​​(such as the specific percentage of η(t) or the specific value of P_fail(t)), but also includes recommended operational guidelines to guide relevant personnel on how to deal with the current situation.

[0180] Early warning signal: Notify relevant persons in charge that the comprehensive loss rate of prestressing may reach a certain high level in the future, approaching the danger limit, and recommend increasing the monitoring frequency and considering adjusting the tensioning force or implementing temporary reinforcement measures.

[0181] Emergency Response Mechanism: Inform on-site personnel that the real-time probability of structural failure has exceeded safety limits, require them to immediately cease all activities involving the hazardous area, and dispatch engineers and technicians to the site for a detailed inspection. Clear operational instructions are also provided on how to safely evacuate or take necessary emergency measures.

[0182] The predicted future trend value is compared with a first preset threshold. If the predicted future trend value exceeds the first preset threshold, an early warning signal is generated. The real-time failure probability is compared with a second preset threshold. If the real-time failure probability exceeds the second preset threshold, an emergency response mechanism is initiated. Combining these two signals, the system can generate comprehensive early warning information. This approach ensures timely identification and rapid response to potential risks, thereby ensuring the safety and reliability of the structure. Early warning information not only helps the engineering team plan preventive maintenance in advance, but also enables swift action in emergency situations to avoid major accidents.

[0183] A multi-source data fusion prestressed structure health monitoring and early warning system, including:

[0184] A weighted fusion module is used to process the monitoring data acquired by sensors through an adaptive weighted data fusion model to output a comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip, and the tension force.

[0185] The prediction module is used to output future trend prediction values ​​based on historical sequences by improving the LSTM prediction model;

[0186] An evaluation calculation module is used to evaluate the future trend prediction value, the prestressed composite loss rate, the anchor clip displacement and the tension force through a fuzzy Petri net-based failure probability model to obtain a real-time failure probability;

[0187] The judgment module is used to compare and judge the future trend prediction value and the real-time failure probability with a preset threshold value to generate early warning information.

[0188] According to one embodiment of the present application, the weighted fusion module is specifically:

[0189] The distributed optical fiber sensor, laser displacement meter and hydraulic jack are used to obtain the axial strain field data of the steel strand, the displacement of the anchor clip and the tension force respectively.

[0190] Dynamically update weights through the entropy weight method, and optimize the weighting scheme by considering the information entropy of different sensors;

[0191] The output is the comprehensive loss rate of prestressing.

[0192] According to one embodiment of the present application, the prediction module is specifically:

[0193] Based on the historical sequences in the monitoring data, the temporal attention mechanism is used to calculate the hidden layer weights, and a regularization term is introduced to prevent overfitting in order to output future trend prediction values.

[0194] A computer-readable storage medium stores a program, which implements the steps of the method when executed by a processor.

[0195] An electronic device comprises a memory, a processor and a program stored in the memory and runnable on the processor, wherein the processor implements the steps in the method when executing the program.

[0196] Anything not described in this application can be achieved by adopting or drawing on existing technologies.

[0197] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0198] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A multi-source data fusion prestressed structure health monitoring and early warning method, characterized in that: include: The monitoring data acquired by the sensors is processed through an adaptive weighted data fusion model to output a comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip, and the tension force. Based on historical sequences, the LSTM prediction model is improved to output future trend prediction values; Based on the future trend prediction value, the prestressed composite loss rate, the anchor clip displacement and the tension force, a failure probability model based on a fuzzy Petri net is used to evaluate and obtain a real-time failure probability; The future trend prediction value and the real-time failure probability are compared and judged with a preset threshold value to generate early warning information.

2. The method according to claim 1, characterized in that The monitoring data acquired based on the sensor is processed by an adaptive weighted data fusion model to output the comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip, and the tension force, specifically: The distributed optical fiber sensor, laser displacement meter and hydraulic jack are used to obtain the axial strain field data of the steel strand, the displacement of the anchor clip and the tension force respectively. Dynamically update weights through the entropy weight method, and optimize the weighting scheme by considering the information entropy of different sensors; The output is the comprehensive loss rate of prestressing.

3. The method according to claim 1, characterized in that Based on the historical sequence, the LSTM prediction model is improved to output the future trend prediction value, specifically: Based on the historical sequences in the monitoring data, the temporal attention mechanism is used to calculate the hidden layer weights, and a regularization term is introduced to prevent overfitting in order to output future trend prediction values.

4. The method according to claim 1, wherein The future trend prediction value, the prestressed comprehensive loss rate, the anchor clip displacement and the tension force are evaluated by a failure probability model based on a fuzzy Petri net to obtain a real-time failure probability, specifically: Receive future trend forecast values ​​of prestressed composite loss rate, anchor clip displacement and tension force; A proposition set is defined, wherein the proposition set includes a displacement index, a stress index, and an acoustic emission energy index, wherein the displacement index indicates whether the displacement exceeds a limit, the stress index indicates whether the stress drops suddenly, and the acoustic emission energy index indicates whether the acoustic emission energy increases suddenly; The failure probability in different situations is evaluated according to the pre-set fuzzy production rule base to obtain the real-time failure probability.

5. The method according to claim 1, wherein The future trend prediction value and the real-time failure probability are compared with a preset threshold value to generate early warning information, specifically: comparing the future trend prediction value with a first preset threshold, and generating an early warning signal if the future trend prediction value is greater than the first preset threshold; Comparing the real-time failure probability with a second preset threshold, and generating an emergency response mechanism if the real-time failure probability is greater than the second preset threshold; The early warning signal and the emergency response mechanism are combined to generate the warning information.

6. A multi-source data fusion prestressed structure health monitoring and early warning system, characterized by: include: A weighted fusion module is used to process the monitoring data acquired by sensors through an adaptive weighted data fusion model to output a comprehensive prestress loss rate. The monitoring data includes the axial strain field data of the steel strand, the displacement of the anchor clip, and the tension force. The prediction module is used to output future trend prediction values ​​based on historical sequences by improving the LSTM prediction model; An evaluation calculation module is used to evaluate the future trend prediction value, the prestressed comprehensive loss rate, the anchor clip displacement and the tension force through a failure probability model based on a fuzzy Petri net to obtain a real-time failure probability; The judgment module is used to compare and judge the future trend prediction value and the real-time failure probability with a preset threshold value to generate early warning information.

7. The system according to claim 6, characterized in that The weighted fusion module is specifically: The distributed optical fiber sensor, laser displacement meter and hydraulic jack are used to obtain the axial strain field data of the steel strand, the displacement of the anchor clip and the tension force respectively. Dynamically update weights through the entropy weight method, and optimize the weighting scheme by considering the information entropy of different sensors; The output is the comprehensive loss rate of prestressing.

8. The system according to claim 6, wherein: The prediction module is specifically: Based on the historical sequences in the monitoring data, the temporal attention mechanism is used to calculate the hidden layer weights, and a regularization term is introduced to prevent overfitting in order to output future trend prediction values.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method according to any one of claims 1 to 5 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps in the method according to any one of claims 1 to 5 are implemented.

Images