A method, device, equipment and storage medium for intelligent monitoring of the stress on concrete beams

Abstract

This application provides a method, device, equipment, and storage medium for intelligent monitoring of concrete beam stress. The method includes: real-time acquisition of concrete beam monitoring data via sensors; analysis of the data using a deep learning model to output probability values ​​for overload and crack states; conversion of these two values ​​into basic probability assignments, followed by fusion using D-S evidence theory to detect evidence conflict values ​​and generate a comprehensive probability distribution through conflict resolution rules; and triggering graded early warnings accordingly. This application integrates sensors, deep learning, and D-S evidence theory to achieve real-time and accurate monitoring and intelligent early warning of concrete beam stress. Combined with visual decision support, it effectively improves the efficiency and reliability of structural safety management and reduces the risk of safety accidents.

Description

Technical Field

[0001] This application belongs to the field of beam stress monitoring, and particularly relates to a method, device, equipment and storage medium for intelligent monitoring of concrete beam stress. Background Technology

[0002] In the daily operation of pier-type wharves, prestressed concrete beams, as key load-bearing structures, directly affect the operational efficiency and safety of the entire wharf. However, with the continuous growth of waterway traffic and the increasing service life of the wharves, these concrete beams are facing increasingly severe challenges.

[0003] Traditionally, monitoring of prestressed concrete beams relies primarily on various types of sensors, such as stress sensors, strain sensors, and displacement sensors, which can collect diverse data on the beam under different stress states. However, the data formats, frequencies, and accuracies acquired by these sensors vary significantly, leading to difficulties in data integration and utilization. More seriously, historical monitoring data is often scattered across different databases, lacking a systematic integration mechanism, resulting in low data utilization. When processing this data, staff not only need to perform extensive manual processing of raw data but also frequently encounter problems such as data loss and mismatches, severely impacting the efficiency and accuracy of monitoring. Furthermore, the lack of unified processing standards for multi-source heterogeneous data prevents its direct use in structural health analysis, further limiting the intelligence level of the monitoring system.

[0004] In structural damage identification, existing methods largely rely on manual inspections and experience-based judgment, making it difficult to detect early cracks, stress concentrations, and other subtle damage in real time and with high accuracy. This easily leads to missed detections and misjudgments, preventing the timely discovery and handling of structural hazards and increasing the risk of safety accidents. Furthermore, developing reasonable maintenance plans for identified damage is also a challenge. Traditional methods require manual review of numerous standards and case studies, a cumbersome process that struggles to guarantee the suitability and effectiveness of solutions, hindering scientific and efficient maintenance management. In addition, monitoring data and maintenance information are primarily transmitted through manual recording and reporting, lacking intuitive and visual presentation methods. This makes it difficult for managers to comprehensively and in real-time grasp the structural health status and maintenance progress, resulting in data-driven decision-making that severely impacts the safety and economy of port operations. Summary of the Invention

[0005] The purpose of this application is to overcome the defects in the prior art and provide a method, device, equipment and storage medium for intelligent monitoring of the stress on concrete beams.

[0006] This application provides a method for intelligent monitoring of the stress on concrete beams, including:

[0007] Real-time monitoring data of concrete beams is collected through sensors;

[0008] Based on the analysis of the real-time monitoring data using a deep learning model, the probability values ​​of overload state and crack state are output.

[0009] The overload state probability value is converted into a first basic probability allocation, and the crack state probability value is converted into a second basic probability allocation;

[0010] Performing a DS evidence theory fusion operation on the first basic probability allocation and the second basic probability allocation includes: detecting evidence conflict values ​​between the first basic probability allocation and the second basic probability allocation; based on the evidence conflict values, fusing the first basic probability allocation and the second basic probability allocation through conflict resolution rules to generate a comprehensive probability distribution; and based on the comprehensive probability distribution, triggering a graded early warning signal according to a preset numerical range.

[0011] Optionally, the real-time monitoring data acquisition of the concrete beam via sensors includes:

[0012] Strain sensors were deployed at the mid-span of the concrete beam.

[0013] Deploy crack width sensors at the support locations of concrete beams;

[0014] The strain sensor is used to collect strain data of the beam.

[0015] Crack propagation data is collected using the crack width sensor;

[0016] The beam strain data and the crack propagation data are used as real-time monitoring data.

[0017] Optionally, the analysis of the real-time monitoring data based on the deep learning model includes:

[0018] Construct a CNN-LSTM hybrid neural network;

[0019] The real-time monitoring data is input into the CNN-LSTM hybrid neural network;

[0020] Spatial feature distribution is extracted using convolutional layers;

[0021] Capture temporal feature changes through long short-term memory layers;

[0022] Output the probability values ​​for overload state and crack state.

[0023] Optionally, converting the overload state probability value into a first basic probability assignment includes:

[0024] Map the probability value of the overload state to the basic probability assignment of the overload event;

[0025] The step of converting the crack state probability value into a second basic probability assignment includes: mapping the crack state probability value to a basic probability assignment of crack events.

[0026] Optionally, detecting the evidence conflict value between the first basic probability allocation and the second basic probability allocation includes:

[0027] Calculate the event pairs where the intersection of the first basic probability assignment and the second basic probability assignment is empty;

[0028] For event pairs with empty intersections, the sum of their probability distribution products is calculated as the evidence conflict value.

[0029] Optionally, the step of fusing the first basic probability allocation and the second basic probability allocation based on the evidence conflict value using conflict resolution rules includes:

[0030] Calculate the sum of the products of the probability distributions of non-conflicting events;

[0031] The sum of the products of the probability distributions of the non-conflicting events is divided by a normalization factor, which is 1 minus the evidence conflict value, to generate a comprehensive probability distribution.

[0032] Optionally, the step of triggering a graded early warning signal based on the comprehensive probability distribution and a preset numerical range includes:

[0033] A yellow warning signal is triggered when the comprehensive probability distribution value is in the range [0.4, 0.6].

[0034] An orange warning signal is triggered when the comprehensive probability distribution value is in the range (0.6, 0.8).

[0035] A red warning signal is triggered when the comprehensive probability distribution value is greater than 0.8.

[0036] This application also provides an intelligent monitoring device for the stress of concrete beams, including:

[0037] The monitoring module collects real-time monitoring data of the concrete beam through sensors;

[0038] The analysis module analyzes the real-time monitoring data based on a deep learning model and outputs the probability values ​​of overload state and crack state.

[0039] The conversion module converts the overload state probability value into a first basic probability allocation and the crack state probability value into a second basic probability allocation.

[0040] The fusion module performs a DS evidence theory fusion operation on the first basic probability allocation and the second basic probability allocation, including: detecting the evidence conflict value between the first basic probability allocation and the second basic probability allocation; based on the evidence conflict value, fusing the first basic probability allocation and the second basic probability allocation through conflict resolution rules to generate a comprehensive probability distribution; and based on the comprehensive probability distribution, triggering a graded early warning signal according to a preset numerical range.

[0041] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0042] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the above-described method.

[0043] The beneficial effects of this application are:

[0044] This application provides an intelligent monitoring method for the stress of concrete beams, comprising: collecting real-time monitoring data of the concrete beams through sensors; analyzing the real-time monitoring data based on a deep learning model to output overload state probability values ​​and crack state probability values; converting the overload state probability values ​​into a first basic probability allocation and the crack state probability values ​​into a second basic probability allocation; performing a DS evidence theory fusion operation on the first and second basic probability allocations, including: detecting evidence conflict values ​​between the first and second basic probability allocations; based on the evidence conflict values, fusing the first and second basic probability allocations through conflict resolution rules to generate a comprehensive probability distribution; and triggering graded early warning signals according to a preset numerical range based on the comprehensive probability distribution. This application, by integrating sensors, deep learning, and DS evidence theory, achieves real-time and accurate monitoring, intelligent early warning, and visualized decision support for the stress of concrete beams, effectively improving the efficiency and reliability of structural safety management and reducing the risk of safety accidents. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the intelligent monitoring process for the stress on concrete beams in this application. Detailed Implementation

[0046] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is to be understood that various forms of implementation of the present disclosure are intended and should not be limited to the embodiments set forth herein. Rather, the embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0047] Please refer to Figure 1 As shown, this application provides an intelligent monitoring method for the stress of concrete beams, which, based on deep learning technology, enables intelligent monitoring of the stress state of prestressed concrete beams in pier-type wharves. The method includes steps such as real-time data acquisition, deep learning analysis, probability allocation transformation, DS evidence theory fusion, and hierarchical early warning triggering.

[0048] S101. Real-time monitoring data of concrete beams is collected through sensors;

[0049] The process of acquiring data is monitored in real time to ensure that the data fully reflects the stress state of the beam.

[0050] Data is acquired in real time at a frequency of 10Hz using sensors deployed at key locations on the beam. The sensors include: a strain sensor deployed at the mid-span of the concrete beam to capture beam strain data; and crack width sensors deployed at the supports of the concrete beam to monitor crack propagation data. The crack width sensor is a crack width sensor (model: PD-2021) manufactured by Tokyo Gyokko Co., Ltd., Japan, with a resolution of 0.01mm and a measurement accuracy of ±0.05mm. The strain sensor is a fiber optic surface strain gauge with a range of ±2500με, an accuracy class of 0.1, and a sampling frequency set to 10Hz. Both types of sensors are deployed on the surface of the concrete beam after being treated with IP68 waterproof sealing. The strain sensor at the mid-span location must be attached parallel to the main reinforcement bars to ensure coordinated deformation with the beam.

[0051] Strain data is acquired using fiber optic surface strain gauges, which are attached to the beam surface and internal main reinforcement positions. These gauges are resistant to electromagnetic interference and offer high precision, enabling stable monitoring of strain distribution. Crack width data is acquired in real time using crack width sensors, allowing for precise measurement of crack size changes.

[0052] The collected data also included vertical and horizontal loads, but the real-time monitoring data focused on beam strain and crack propagation data, providing a basis for subsequent analysis.

[0053] The collected data needs to be preprocessed: First, outlier detection is performed, and outlier data is handled using the 3σ principle. Specifically, if the stress data at a certain moment exceeds three times the standard deviation of its mean, it is determined to be an outlier, and the weighted average of adjacent time points is used to replace it.

[0054] Subsequently, the data was standardized using a normalization formula:

[0055]

[0056] in, Here, x represents the normalized data value, and x represents the original data value. This represents the minimum value in the training set data. This indicates the maximum value of the training set data. This step ensures that the data format is consistent with that used during deep learning model training, eliminating the influence of dimensional differences. The preprocessed beam strain data and crack propagation data are then used as input for the next stage of real-time monitoring.

[0057] S102. Analyze the real-time monitoring data based on a deep learning model, and output the probability values ​​of overload state and crack state.

[0058] Real-time monitoring data is processed using deep learning models, and probabilistic analysis results are output.

[0059] The deep learning model employs a CNN-LSTM hybrid neural network architecture to adapt to the complex characteristics of multi-source data. The model construction includes: a Convolutional Neural Network (CNN) part with three convolutional layers, each with a 3x3 kernel and a stride of 1, used to extract the spatial feature distribution between different monitoring points of the prestressed concrete beam, such as the distribution pattern of strain data within the beam structure; a Long Short-Term Memory (LSTM) part with 128 memory units, using a gating mechanism to capture changes in time-series features, such as the periodic fluctuations in load data; and a global average pooling layer to reduce the dimensionality of the features, followed by fully connected layers, and finally, a Softmax activation function to output probability values.

[0060] The input layer of the CNN-LSTM hybrid neural network has a time step of 60 sampling points and contains three convolutional layers (each with 32, 64, and 128 kernels respectively), with a kernel size of 3×1. The LSTM layer has 128 hidden units and a dropout rate of 0.25 to avoid overfitting. The network is trained using the Adam optimizer with an initial learning rate of 0.001, a batch size of 32, and 200 training epochs. The max-pooling layers following the convolutional layers are set to a size of 2×1 to compress the feature dimension.

[0061] The analysis process is as follows: Preprocessed real-time monitoring data is input into a CNN-LSTM hybrid neural network. Spatial feature distribution is extracted through convolutional layers to uncover local patterns in the data; temporal feature changes, such as the rate trend of crack propagation, are captured through LSTM layers. The model output includes overload state probability values ​​and crack state probability values. The overload state probability value is calculated using the Softmax function based on the classification task output.

[0062]

[0063] Where K represents the total number of categories, including states such as "normal", "minor damage", "serious damage", and "overload"; This represents the probability output value for the j-th category; This represents the inactive raw output value corresponding to the j-th category in the model's output layer; This represents the inactive original output value of the k-th category in the output layer; e is a natural constant used for exponential transformation mapping to the positive number range.

[0064] The probability value for crack condition is also calculated using this formula, but it applies to crack-related categories. The output directly reflects the likelihood of abnormal conditions in the beam.

[0065] S103. The overload state probability value is converted into a first basic probability allocation, and the crack state probability value is converted into a second basic probability allocation.

[0066] The transformation from probability values ​​to probability assignments prepares for evidence fusion. The transformation process is based on the direct mapping principle: the probability value of the overload state is transformed into the first basic probability assignment, representing the degree of confidence in the overload event; the probability value of the crack state is transformed into the second basic probability assignment, representing the degree of confidence in the crack event.

[0067] The specific operation is as follows: if the model predicts an overload state probability of 0.7, then the first basic probability allocation is m({overload}) = 0.7, which represents the degree of support of Evidence 1 for the "overload" proposition; similarly, if the crack state probability is 0.6, then the second basic probability allocation is m({crack}) = 0.6, representing the degree of support of Evidence 2 for the "crack" proposition. This transformation ensures that the probability results can be directly used in DS evidence theory, preserving uncertainty information.

[0068] S104. Perform a DS evidence theory fusion operation on the first basic probability allocation and the second basic probability allocation, including: detecting the evidence conflict value between the first basic probability allocation and the second basic probability allocation; and based on the evidence conflict value, fusing the first basic probability allocation and the second basic probability allocation through conflict resolution rules to generate a comprehensive probability distribution.

[0069] By fusing multiple pieces of evidence using the DS evidence theory, a robust comprehensive probability is generated.

[0070] The fusion operation consists of two steps:

[0071] First, detect the conflict value of evidence. Calculate the event pairs where the intersection of the first and second basic probability assignments is empty, i.e., when the propositions of the two pieces of evidence do not overlap. For example, if evidence 1 supports "overload" and evidence 2 supports "crack," and their intersection is empty, then calculate the sum of the products of the probability assignments of these event pairs as the conflict value of evidence. The formula is:

[0072]

[0073] Here, A and B are different evidentiary propositions; for example, A may be "overload" and B may be "crack". Let be the first basic probability assignment value, representing the degree to which evidence 1 supports proposition A; The second basic probability assignment value represents the degree to which evidence 2 supports proposition B; This represents a pair of propositions where the intersection of A and B is empty. The above formula is used for summation operations to calculate the conflict coefficient.

[0074] Secondly, based on the evidence conflict value, probability assignments are fused using conflict resolution rules. The sum of the products of probability assignments for non-conflicting events is calculated, i.e., when the intersection of the propositions of the two pieces of evidence is not empty. The formula is:

[0075]

[0076] Where C represents the synthesized proposition after fusion, such as "critical state of the beam"; A∩B=C represents a proposition pair whose intersection of A and B equals proposition C. The above formula is used for summation. Then, the non-conflicting sum is divided by the normalization factor (1 minus the conflict value) to generate the synthesized probability distribution:

[0077]

[0078] Here, m(c) represents the comprehensive probability distribution value for the final state of the beam; the normalization factor in the denominator is used to eliminate the influence of conflicting evidence. This fusion enhances the reliability of decision-making; for example, when there is conflicting evidence of overload and cracks, the system can still output a consistent comprehensive probability.

[0079] S105. Based on the comprehensive probability distribution, trigger a graded early warning signal according to a preset numerical range.

[0080] Multi-level early warnings are triggered based on comprehensive probability values ​​to achieve risk-based response.

[0081] The early warning mechanism is divided into three levels: a yellow warning signal is triggered when the comprehensive probability distribution value is in the range of [0.4, 0.6], indicating a low potential risk, and the system recommends increasing the monitoring frequency; an orange warning signal is triggered when the comprehensive probability distribution value is in the range of (0.6, 0.8), indicating a medium risk, requiring professional personnel to conduct a detailed investigation; and a red warning signal is triggered when the comprehensive probability distribution value is greater than 0.8, indicating a high risk, requiring an immediate comprehensive inspection and emergency repairs. Warning signals are sent through multiple channels, including SMS, email, and audible and visual alarms, to ensure timely response by staff. Simultaneously, the system records the warning time, comprehensive probability value, and processing results for subsequent model optimization.

[0082] This application also provides an intelligent monitoring device for the stress of concrete beams, including:

[0083] The monitoring module collects real-time monitoring data of the concrete beam through sensors;

[0084] The analysis module analyzes the real-time monitoring data based on a deep learning model and outputs the probability values ​​of overload state and crack state.

[0085] The conversion module converts the overload state probability value into a first basic probability allocation and the crack state probability value into a second basic probability allocation.

[0086] The fusion module performs a DS evidence theory fusion operation on the first basic probability allocation and the second basic probability allocation, including: detecting the evidence conflict value between the first basic probability allocation and the second basic probability allocation; based on the evidence conflict value, fusing the first basic probability allocation and the second basic probability allocation through conflict resolution rules to generate a comprehensive probability distribution; and based on the comprehensive probability distribution, triggering a graded early warning signal according to a preset numerical range.

[0087] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0088] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the above-described method.

[0089] The above description of the embodiments is provided to enable those skilled in the art to understand and apply this application. Those skilled in the art will readily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without inventive effort. Therefore, this application is not limited to the above embodiments, and any improvements and modifications made to this application based on the disclosure thereof should be within the scope of protection of this application.

Claims

1. A method for intelligent monitoring of the stress on concrete beams, characterized in that, include: Real-time monitoring data of concrete beams is collected through sensors; Based on the analysis of the real-time monitoring data using a deep learning model, the probability values ​​of overload state and crack state are output. The overload state probability value is converted into a first basic probability allocation, and the crack state probability value is converted into a second basic probability allocation; Performing a DS evidence theory fusion operation on the first basic probability allocation and the second basic probability allocation includes: detecting evidence conflict values ​​between the first basic probability allocation and the second basic probability allocation; based on the evidence conflict values, fusing the first basic probability allocation and the second basic probability allocation through conflict resolution rules to generate a comprehensive probability distribution; and based on the comprehensive probability distribution, triggering a graded early warning signal according to a preset numerical range. The overload state probability value is calculated using the Softmax function based on the classification task output: ； K is the total number of categories. This represents the output value indicating the probability of the j-th category. This represents the inactive raw output value corresponding to the j-th category in the model's output layer. This represents the inactive raw output value of the k-th category in the output layer, where e is the natural constant; the crack state probability value is also calculated using this formula, but only for crack-related categories; ； A and B are different evidentiary propositions; A is "overload" and B is "crack". Let be the first basic probability assignment value, representing the degree to which evidence 1 supports proposition A; The second basic probability assignment value represents the degree to which evidence 2 supports proposition B; This represents a pair of propositions where the intersection of A and B is empty; The first basic probability assignment and the second basic probability assignment are fused through conflict resolution rules, including: calculating the sum of the products of the probability assignments of non-conflict events. ； C represents the integrated proposition after fusion. For proposition pairs where the intersection of A and B equals proposition C, divide the non-conflicting sum by (1 minus the conflicting value) to generate the comprehensive probability distribution: ； m(c) represents the comprehensive probability distribution value of the final state of the beam.

2. The method according to claim 1, characterized in that, The real-time monitoring data of the concrete beam collected by sensors includes: Strain sensors were deployed at the mid-span of the concrete beam. Deploy crack width sensors at the support locations of concrete beams; The strain sensor is used to collect strain data of the beam. Crack propagation data is collected using the crack width sensor; The beam strain data and the crack propagation data are used as real-time monitoring data.

3. The method according to claim 1, characterized in that, The analysis of the real-time monitoring data based on the deep learning model includes: Construct a CNN-LSTM hybrid neural network; The real-time monitoring data is input into the CNN-LSTM hybrid neural network; Spatial feature distribution is extracted using convolutional layers; Capture temporal feature changes through long short-term memory layers; Output the probability values ​​for overload state and crack state.

4. The method according to claim 1, characterized in that, The step of converting the overload state probability value into a first basic probability assignment includes: Map the probability value of the overload state to the basic probability assignment of the overload event; The step of converting the crack state probability value into a second basic probability assignment includes: mapping the crack state probability value to a basic probability assignment of crack events.

5. The method according to claim 1, characterized in that, The detection of the evidence conflict value between the first basic probability allocation and the second basic probability allocation includes: Calculate the event pairs where the intersection of the first basic probability assignment and the second basic probability assignment is empty; For event pairs with empty intersections, the sum of their probability distribution products is calculated as the evidence conflict value.

6. The method according to claim 1, characterized in that, The step of fusing the first basic probability allocation and the second basic probability allocation based on the evidence conflict value through conflict resolution rules includes: Calculate the sum of the products of the probability distributions of non-conflicting events; The sum of the products of the probability distributions of the non-conflicting events is divided by a normalization factor, which is 1 minus the evidence conflict value, to generate a comprehensive probability distribution.

7. The method according to claim 1, characterized in that, The step of triggering graded early warning signals based on the comprehensive probability distribution and a preset numerical range includes: A yellow warning signal is triggered when the comprehensive probability distribution value is in the range [0.4, 0.6]. An orange warning signal is triggered when the comprehensive probability distribution value is in the range (0.6, 0.8). A red warning signal is triggered when the comprehensive probability distribution value is greater than 0.

8.

8. A smart monitoring device for the stress on a concrete beam, characterized in that, include: The monitoring module collects real-time monitoring data of the concrete beam through sensors; The analysis module analyzes the real-time monitoring data based on a deep learning model and outputs the probability values ​​of overload state and crack state. The conversion module converts the overload state probability value into a first basic probability allocation and the crack state probability value into a second basic probability allocation. The fusion module performs a DS evidence theory fusion operation on the first basic probability allocation and the second basic probability allocation, including: detecting the evidence conflict value between the first basic probability allocation and the second basic probability allocation; based on the evidence conflict value, fusing the first basic probability allocation and the second basic probability allocation through conflict resolution rules to generate a comprehensive probability distribution; and based on the comprehensive probability distribution, triggering a graded early warning signal according to a preset numerical range. The overload state probability value is calculated using the Softmax function based on the classification task output: ； K is the total number of categories. This represents the output value indicating the probability of the j-th category. This represents the inactive raw output value corresponding to the j-th category in the model's output layer. This represents the inactive raw output value of the k-th category in the output layer, where e is the natural constant; the crack state probability value is also calculated using this formula, but only for crack-related categories; ； A and B are different evidentiary propositions; A is "overload" and B is "crack". Let be the first basic probability assignment value, representing the degree to which evidence 1 supports proposition A; The second basic probability assignment value represents the degree to which evidence 2 supports proposition B; This represents a pair of propositions where the intersection of A and B is empty; The first basic probability assignment and the second basic probability assignment are fused through conflict resolution rules, including: calculating the sum of the products of the probability assignments of non-conflict events. ； C represents the integrated proposition after fusion. For proposition pairs where the intersection of A and B equals proposition C, divide the non-conflicting sum by (1 minus the conflicting value) to generate the comprehensive probability distribution: ； m(c) represents the comprehensive probability distribution value of the final state of the beam.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1-7.

Images