A system and method for predicting the life of a steel structure bridge after fatigue crack repair

Abstract

The application relates to the technical field of bridge engineering, in particular to a steel structure bridge fatigue crack repair life prediction system and method; the method comprises the following steps: collecting load data and stress data in the operation process of a bridge structure after fatigue cracks are repaired; dynamic response characteristic parameters are extracted in combination with stress distribution characteristics of a fatigue crack repair area; multi-dimensional data fusion and preprocessing are performed on the dynamic response characteristic parameters, and a data set for training a deep learning algorithm is constructed; a crack propagation rate prediction model and a residual life prediction model are constructed in combination with a deep learning algorithm and a fracture mechanics theory; the residual life of the structure after repair is predicted according to the bridge operation state and the performance degradation law of the repair area, and multi-level early warning is performed. The method provided by the application can realize bridge structure performance evaluation and life prediction and potential risk early warning, and ensures the safety and reliability of bridge operation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of bridge engineering, in particular to a steel structure bridge fatigue crack repair post-life prediction system and method. BACKGROUND

[0002] The performance evaluation of the bridge structure after fatigue crack repair lacks effective means, the performance of the repaired structure is affected by the load history, crack repair quality and stress distribution characteristics, the existing method lacks accurate description of the dynamic changes of the performance of the repaired area, and it is difficult to quantify the repair effect.

[0003] Traditional crack propagation rate calculation mainly depends on a single fracture mechanics formula, which has a certain theoretical basis, but ignores the complexity of the time sequence characteristics in the crack propagation process, and cannot dynamically adapt to the changes of the bridge operation state.

[0004] In summary, the steel structure bridge fatigue crack repair post-life prediction field mainly faces the following problems: lack of accurate evaluation of the dynamic performance of the repaired area, insufficient crack propagation rate prediction accuracy, and low reliability of the life prediction model. The existence of these problems not only restricts the scientific evaluation of the safety state of the repaired bridge, but also poses a potential threat to the subsequent maintenance and safe operation of the bridge. Therefore, a scientific, systematic and accurate fatigue crack repair post-life prediction method is urgently needed to comprehensively improve the safety management level of the bridge structure.

[0005] In view of this, the application provides a steel structure bridge fatigue crack repair post-life prediction system and method. SUMMARY

[0006] To achieve the above-mentioned purpose, the application provides a steel structure bridge fatigue crack repair post-life prediction system and method, and the specific technical solutions are as follows: a steel structure bridge fatigue crack repair post-life prediction method, comprising:

[0007] Collecting load data and stress data in the operation process of the bridge structure after repairing the fatigue crack;

[0008] Based on the finite element analysis method, the stress distribution characteristics of the fatigue crack repair area are combined to extract dynamic response characteristic parameters;

[0009] The dynamic response characteristic parameters are subjected to multi-dimensional data fusion and preprocessing to construct a data set for training a deep learning algorithm;

[0010] Combining the deep learning algorithm and the fracture mechanics theory, a crack propagation rate prediction model and a residual life prediction model are constructed;

[0011] According to the performance degradation law of the bridge operation state and the repair area, the residual life of the repaired structure is predicted, and multi-level warning is performed.

[0012] Preferably, strain sensors and load sensors are arranged on the bridge structure after repairing the fatigue crack to collect strain data and load data of the bridge structure in real time during operation;

[0013] The collected load data and strain data are preprocessed, including denoising and outlier rejection; the preprocessed load data and strain data are constructed into time series data.

[0014] Preferably, according to the collected load data, load conditions of the finite element model are set; the time history of the load is the same as the collected load time series, and the loading frequency of the load is consistent with the sampling frequency; multiple load conditions are set, respectively corresponding to different load levels and loading modes;

[0015] For each load condition, a finite element static analysis is performed to obtain a stress distribution cloud map of the fatigue crack repair area;

[0016] Based on the stress distribution characteristics of the fatigue crack repair area, dynamic response characteristic parameters are extracted; the dynamic response characteristic parameters include stress amplitude, stress ratio, stress gradient, and stress triaxiality.

[0017] Preferably, the collected load time series data and strain time series data are fused with the extracted dynamic response characteristic parameters in multiple dimensions to construct a fusion data set;

[0018] The fusion data set is preprocessed, including data normalization, data enhancement, and data division;

[0019] Based on the preprocessed fusion data set, a training deep learning data set for training a deep learning algorithm is constructed, and the training deep learning data set is divided into a training data set, a validation data set, and a test data set.

[0020] Preferably, a crack propagation rate prediction model is constructed based on a long short-term memory neural network (LSTM) model, and a network structure of the LSTM model is designed, including an input layer, an LSTM layer, a fully connected layer, and an output layer; the input layer is used to input the training data set; the LSTM layer is composed of multiple LSTM units, each LSTM unit contains an input gate, a forget gate, an output gate, and a memory unit, and is used to learn and memorize features of time series data; the fully connected layer is used to extract features output by the LSTM layer and map the features to predicted values of the crack propagation rate; the output layer adopts a linear activation function to output the predicted crack propagation rate value;

[0021] The LSTM model is trained by using a training data set, parameters of the model are optimized, a mean square error is used as a loss function, in the training process of the LSTM model, the model is verified by using a verification data set, and the LSTM model meeting performance requirements is used as a crack propagation rate prediction model.

[0022] Preferably, a linear elastic fracture mechanics theory is introduced to establish a relationship between the crack propagation rate and a stress intensity factor range, a Paris formula is used to describe the relationship of the crack propagation rate, and the stress intensity factor range is calculated.

[0023] Based on the Paris formula and a calculation formula of the stress intensity factor range, a residual life prediction model is constructed, the predicted crack propagation rate is substituted into the Paris formula, the cumulative cycle number of crack propagation is obtained by integration, and a residual life prediction value of the repaired structure is obtained.

[0024] Preferably, according to the residual life prediction value, a multi-level early warning of the safety state of the repaired structure is performed, a warning threshold and an alarm threshold are set, a warning signal is sent when the residual life prediction value is lower than the warning threshold, and an alarm signal is sent when the residual life prediction value is lower than the alarm threshold.

[0025] A steel structure bridge fatigue crack residual life prediction system is used to implement the steel structure bridge fatigue crack residual life prediction method, and comprises a data acquisition module, a feature extraction module, a data fusion module, a residual life prediction module and a multi-level early warning module.

[0026] The data acquisition module is used to acquire load data and stress data in the running process of the bridge structure after the fatigue crack is repaired.

[0027] The feature extraction module extracts dynamic response characteristic parameters based on a finite element analysis method and in combination with stress distribution characteristics of the fatigue crack repair area.

[0028] The data fusion module performs multi-dimensional data fusion and preprocessing on the dynamic response characteristic parameters, and constructs a data set for training a deep learning algorithm.

[0029] The residual life prediction module combines a deep learning algorithm and a fracture mechanics theory to construct a crack propagation rate prediction model and a residual life prediction model.

[0030] The multi-level early warning module predicts the residual life of the repaired structure according to the bridge running state and the performance degradation law of the repair area, and performs multi-level early warning.

[0031] An electronic device comprises a processor and a memory, wherein the memory stores a computer program that can be invoked by the processor; the processor executes the steel structure bridge fatigue crack repair after life prediction method by invoking the computer program stored in the memory.

[0032] A computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steel structure bridge fatigue crack repair after life prediction method.

[0033] The beneficial effects of the present application: the present application collects load and stress data, which can accurately reflect the bridge operation state, provides basic data support for subsequent analysis, and ensures the reliability and accuracy of model input.

[0034] The present application extracts dynamic response characteristic parameters through finite element analysis, which can accurately characterize the stress distribution characteristics of the repair area and provide key physical property descriptions for crack propagation prediction.

[0035] The present application improves the consistency and effectiveness of data through data fusion and preprocessing, constructs a high-quality data set, provides high-quality training samples for deep learning algorithms, and improves model accuracy.

[0036] The present application uses deep learning to mine time series features and combines fracture mechanics theory to construct a model that can dynamically predict crack propagation behavior, significantly improving the accuracy of residual life prediction.

[0037] The present application predicts the remaining life through performance degradation law and combines a multi-level warning mechanism to timely detect potential hazards, provide scientific guidance for bridge maintenance, and improve operational safety.

[0038] The method of the present application combines finite element analysis, deep learning algorithms, and fracture mechanics theory to comprehensively improve the accuracy of performance evaluation and life prediction of repaired bridge structures, can real-time alert potential risks, optimize bridge maintenance and management strategies, and ensure the safety and reliability of bridge operation. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 A steel structure bridge fatigue crack repair after life prediction method flowchart is provided for the present application;

[0040] Figure 2 A crack propagation rate prediction model construction flowchart of the steel structure bridge fatigue crack repair after life prediction method is provided for the present application;

[0041] Figure 3 A residual life prediction and warning flowchart of the steel structure bridge fatigue crack repair after life prediction method is provided for the present application;

[0042] Figure 4 A steel structure bridge fatigue crack repair life prediction system structure diagram is provided. DETAILED DESCRIPTION

[0043] In order to make the above objectives, features and advantages of the present application more apparent, a detailed description of the specific embodiments of the present application will be given below with reference to the accompanying drawings.

[0044] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0045] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is separate or alternative to other embodiments.

[0046] Embodiment 1

[0047] Reference Figures 1 to 3 The first embodiment of the present application provides a steel structure bridge fatigue crack repair life prediction method.

[0048] Step 1: Collecting load data and stress data of the bridge structure after repairing fatigue cracks during operation.

[0049] Strain sensors and load sensors are arranged on the bridge structure after repairing fatigue cracks to collect strain data and load data of the bridge during operation in real time;

[0050] For example, according to the stress distribution characteristics of the fatigue crack repair area, appropriate strain sensors and load sensors are selected; the strain sensor is a resistance strain gauge, the model is BE120-05AA, the range is ±5000 micro-strain, the sensitivity coefficient is 2.0±0.1, and the resistance value is 120±0.5Ω; the load sensor is a pressure load sensor, the model is CYB-YB-50KN, the range is 050kN, the overall accuracy is 0.02%FS, and the output signal is 05V.

[0051] Exemplarily, the strain gauges are arranged at key positions in and near the fatigue crack repair area, including the crack tip, the crack propagation path, the stress concentration area, etc.; at least 3 strain gauges are arranged at each key position, respectively arranged along the longitudinal direction, the transverse direction and the 45° direction of the bridge, to measure the strain response in different directions; the load sensors are arranged at the key sections of the bridge to measure the load acting on the bridge.

[0052] Exemplarily, a wireless sensor network system is built, which is composed of wireless strain nodes, wireless load nodes, wireless relay nodes and wireless base stations; the wireless strain nodes and the wireless load nodes are connected with the strain gauges and the load sensors respectively, to collect strain data and load data, and transmit the data to the wireless relay nodes through wireless communication; the wireless relay nodes transmit the data to the wireless base stations, and the wireless base stations are connected with the upper computer to realize data storage and processing.

[0053] Exemplarily, the data collection frequency of the wireless sensor network is set to 200 Hz to ensure that the dynamic change characteristics of the load and the strain can be accurately captured; the data collection time T is determined according to actual needs.

[0054] The collected load data P(t) and strain data are preprocessed, including denoising and outlier rejection; the wavelet transform method is used for denoising the original data, the wavelet base function is selected as sym5, and the decomposition layer number is 5; the denoised data is detected and rejected for outliers.

[0055] The preprocessed load data and strain data are constructed into time series data, represented as P(t)|t=1,2,...,m and n0=1,2,...,n, where m is the number of sampling points, m=f s T, f s is the sampling frequency, and n is the number of strain sensors;

[0056] This step collects the strain data of the fatigue crack repair area and the load data of the bridge by arranging strain sensors and load sensors in real time, which provides reliable data support for subsequent fatigue fracture analysis and life prediction; the collected data is preprocessed, including denoising and outlier rejection, which improves the quality of the data and lays a foundation for subsequent data analysis; time series data representing the strain response of the fatigue crack repair area and the load action of the bridge are formed, which provides data input for subsequent feature parameter extraction, fatigue crack propagation rate analysis and life prediction.

[0057] Step 2: Based on the finite element analysis method, the dynamic response characteristic parameters are extracted in combination with the stress distribution characteristics of the fatigue crack repair area.

[0058] According to the bridge design drawings and the fatigue crack repair scheme, a finite element model of the bridge structure is established by using the finite element software ANSYS; the model considers the geometric size, material properties, boundary conditions and characteristics of the fatigue crack repair area of the bridge; the eight-node solid element SOLID185 is used to simulate the bridge structure, and the six-node triangular element PLANE183 is used to simulate the fatigue crack repair area; the mesh near the fatigue crack repair area is densified to improve the calculation accuracy.

[0059] According to the load data P(t)|t=1, 2,..., m collected in step 1, the load working condition of the finite element model is set; the load is applied at the key section of the bridge, and the size and direction are consistent with the measured load data; the time history of the load is the same as the collected load time series, and the loading frequency is consistent with the sampling frequency; multiple load working conditions are set to correspond to different load levels and loading methods to simulate the load action in the actual bridge operation process.

[0060] For each load working condition, a finite element static analysis is performed to obtain the stress distribution cloud diagram of the fatigue crack repair area; the stress distribution cloud diagram reflects the stress distribution characteristics of the fatigue crack repair area, including stress concentration and stress gradient; the stress tensor of the fatigue crack repair area is σ ij =[σ xx ,σ yy ,σ zz ,τ xy ,τ yz ,τ zx ] ij ,σ xx ,σ yy ,σ zz are normal stress components, and τ xy ,τ yz ,τ zx are shear stress components.

[0061] Based on the stress distribution characteristics of the fatigue crack repair area, dynamic response characteristic parameters are extracted; the dynamic response characteristic parameters include stress amplitude, stress ratio, stress gradient and stress triaxiality.

[0062] Stress amplitude: at the load peak and valley time, the maximum principal stress σ1 and the minimum principal stress σ3 of the fatigue crack repair area are extracted, and the stress amplitude Δσ=σ1-σ3 is calculated.

[0063] Stress ratio: at the load peak time, the maximum principal stress σ1 and the minimum principal stress σ3 of the fatigue crack repair area are extracted, and the stress ratio R=σ3 / σ1 is calculated.

[0064] Stress gradient: at the load peak time, the stress tensors σ ij and σi'j' , the stress gradient G = |σ ij -σ i'j' | / d is calculated, where d is the distance between the centers of two units.

[0065] Stress triaxiality: at the moment of the load peak, the stress tensor σ ij of the unit in the fatigue crack repair area is extracted, and the stress triaxiality Y T is calculated, where σ m is the average normal stress, and the calculation formula of σ eq is: σ m = (σ m + σ m + σ xx ) / 3. yy zz

[0066] σ eq is the equivalent stress, and the calculation formula of σ eq is:

[0067]

[0068] Suppose that l dynamic response characteristic parameters are extracted, and the dynamic response characteristic parameters under the i-th load condition can be expressed as X i = (x i1 , x i2 ,..., x il ).

[0069] In this step, the stress distribution characteristics of the fatigue crack repair area under actual load are obtained through finite element analysis, which provides a basis for the extraction of dynamic response characteristic parameters; the dynamic response characteristic parameters representing the fatigue crack propagation characteristics are extracted, including stress amplitude, stress ratio, stress gradient, and stress triaxiality, etc. These characteristic parameters are closely related to the fatigue crack propagation rate and are important influencing factors for fatigue life prediction; the dynamic response characteristic parameters extracted under different load conditions are constructed into a characteristic parameter set, which provides data support for subsequent data fusion, preprocessing, and deep learning model training.

[0070] Step 3: Multi-dimensional data fusion and preprocessing are performed on the dynamic response characteristic parameters to construct a data set for training a deep learning algorithm.

[0071] The load time series data P(t)|t = 1, 2,..., m and the strain time series data n0 = 1, 2,..., n collected in step 1 are subjected to multi-dimensional data fusion with the dynamic response characteristic parameters X i extracted in step 2 to construct a fusion data set D = {(P j , E j , X​​j P(tj), P(tj+1), …, P(tj+k)}, j = 1, 2, …, N} ; P j = (P(t1), P(t2), …, P(t k ) is the load sequence of the jth sample; E j = (ε1(t1), …, ε1(t k ), ε2(t1), …, ε2(t k ), …, ε(t1), …, ε n (t k ) is the strain sequence of the jth sample, X j = (X i1 , X i2 , …, X il ) is the dynamic response characteristic parameter corresponding to the jth sample, N is the total number of samples, and k is the time step of each sample; by multi-dimensional data fusion, the load, strain, and dynamic response characteristic parameter are integrated into a high-dimensional data set, improving the information density and relevance of the data.

[0072] The fusion data set D is preprocessed, including data normalization, data enhancement, and data division.

[0073] The data normalization includes: using the maximum and minimum value normalization method, the load sequence P j , the strain sequence E j , and the dynamic response characteristic parameter X j are normalized to the interval [0, 1] respectively, that is:

[0074]

[0075] wherein, and are the normalized load, strain, and dynamic response characteristic parameter respectively, x iq is the unnormalized dynamic response characteristic parameter, min(·) and max(·) represent the minimum value and the maximum value respectively; the normalization process can eliminate the influence of different dimensions and orders of magnitude, and improve the comparability of the data.

[0076] The data enhancement includes: using data expansion technology, the fusion data set D is expanded to generate new training samples; the data expansion method includes: adding random noise to the load sequence P j and the strain sequence E j , randomly perturbing the dynamic response characteristic parameter X j , randomly cropping and splicing the samples; data enhancement can effectively increase the number and diversity of training samples, and improve the generalization ability and robustness of the deep learning model.

[0077] The data division includes: dividing the pre-processed fusion data set D into a training set and a validation set, wherein the training set is used to train the deep learning model, and the validation set is used to adjust the model hyperparameters and prevent overfitting.

[0078] Based on the pre-processed fusion data set, a training deep learning data set for training the deep learning algorithm is constructed; each sample is represented in the form of (S j ,y j ), wherein is the input feature of the jth sample, and are the normalized load sequence, strain sequence and dynamic response feature parameter sequence respectively, y j is the crack propagation rate or residual life corresponding to the jth sample; the crack propagation rate and residual life are determined according to actual monitoring data or theoretical calculation formula, as the training label of the deep learning model, to obtain the training data set D train ={(S j ,y j )|j=1,2,...,N1} and the validation data set D val ={(Sj,y j )|j=N1+1,N1+2,...,N2}, wherein N1 and N2 are the number of samples in the training set and the validation set respectively.

[0079] This step integrates the load, strain and dynamic response feature parameters into a high-dimensional data set through multi-dimensional data fusion, fully utilizes various monitoring data and finite element analysis results, improves the information density and relevance of the data, and provides rich feature information for the training of the deep learning model; data normalization, data enhancement and data division are used for standardization, expansion and optimization of the fusion data set, which improves the quality and applicability of the data and enhances the generalization ability and robustness of the deep learning model; a data set for training the deep learning algorithm is constructed, the input features are mapped with the crack propagation rate or residual life, which provides direct data support for the training and prediction of the deep learning model, and lays the data foundation for fatigue life prediction.

[0080] Step 4: Combine deep learning algorithm and fracture mechanics theory to construct crack propagation rate prediction model and residual life prediction model.

[0081] Long short-term memory neural network (LSTM) is selected as the basic algorithm of the crack propagation rate prediction model; LSTM is a special recurrent neural network (RNN) that can effectively process time series data and capture long-term dependencies in data; LSTM overcomes the gradient vanishing and gradient explosion problems of traditional RNN by introducing a gating mechanism, and has stronger modeling ability and more stable training process.

[0082] Design the network structure of the LSTM model, including an input layer, an LSTM layer, a fully connected layer, and an output layer; the input layer receives the training dataset D constructed in step 3. train ={(S j ,y j )|j=1,2,...,N1}, where Let y be the input feature of the j-th sample. j The label is the crack propagation rate of the j-th sample; the LSTM layer consists of multiple LSTM units, each containing an input gate, a forget gate, an output gate, and a memory unit, which can adaptively learn and remember the features of the time series data; the fully connected layer is used to extract the high-level features of the LSTM layer output and map them to the predicted crack propagation rate; the output layer uses a linear activation function to output the predicted crack propagation rate value.

[0083] Using training dataset D train The LSTM model is trained and its parameters are optimized; mean squared error (MSE) is used as the loss function, i.e.:

[0084]

[0085] Among them, y j The label for the crack propagation rate of the j-th sample. Label the crack propagation rate predicted by the LSTM model; employ the Adam optimization algorithm to minimize the loss function, adaptively adjusting the model's weights and biases; during training, utilize the validation dataset D. val The model is validated and its generalization performance is monitored to avoid overfitting. When the performance on the validation set reaches its optimum, the model parameters are saved as the final crack propagation rate prediction model.

[0086] By introducing the theory of linear elastic fracture mechanics (LEFM), a relationship is established between the crack propagation rate and the stress intensity factor range ΔK; the Paris formula is used to describe the crack propagation rate da / dN. z The relationship with ΔK is as follows: Where a is the crack size, N z To load the loop count, C and is a material constant.

[0087] The stress intensity factor range ΔK is related to the stress amplitude Δσ and the crack size a, and can be expressed as: Where Y is a correction factor related to crack shape and crack size.

[0088] Based on the Paris formula and the calculation formula of the stress intensity factor range, a residual life prediction model is constructed; the crack propagation rate da / dN is predicted by using the LSTM model z ; the predicted crack propagation rate is substituted into the Paris formula to obtain: The integral of the above formula is obtained to obtain the cumulative cycle number N of crack propagation f , that is: where a0 is the initial crack size, a c is the critical crack size;

[0089] The residual life N r can be expressed as: N r =N f -N0; where N0 is the current loading cycle number; by solving the above integral equation, the residual life prediction value of the repaired structure is obtained.

[0090] In this step, the LSTM neural network is used to construct a crack propagation rate prediction model, which fully utilizes the characteristics of time series data, captures the dynamic change law of the crack propagation process, and improves the accuracy of crack propagation rate prediction; the theory of fracture mechanics is introduced to establish a physical relationship between the crack propagation rate and the stress intensity factor range, which enhances the interpretability and rationality of the residual life prediction model; by combining deep learning algorithm and fracture mechanics theory, a residual life prediction model is constructed, which comprehensively considers the load, strain and dynamic response characteristics, and improves the reliability and applicability of life prediction; by solving the integral equation, the quantitative life prediction result is obtained, which provides a direct basis for subsequent warning and decision-making.

[0091] Step 5: According to the bridge operation state and the performance degradation law of the repaired area, the residual life of the repaired structure is predicted, and multi-level warning is carried out.

[0092] According to the residual life prediction value N r of the repaired structure obtained in step 4, the safety state of the structure is warned in multiple levels; the warning threshold N warn and the alarm threshold N alarm are set, when the residual life prediction value N r is lower than the warning threshold, a warning signal is sent out, prompting the need to strengthen monitoring and detection, and evaluating the safety performance of the structure; when the residual life prediction value N r is lower than the alarm threshold, an alarm signal is sent out, prompting the need to take measures such as load limiting and reinforcement to ensure the safe use of the structure.

[0093] The warning threshold N warn and the alarm threshold N alarmThe threshold value can be determined according to a design file, an evaluation criterion or expert experience, and is generally a certain percentage of the design life, such as 80% and 90%; the specific threshold value can be adjusted according to actual needs.

[0094] The step realizes real-time monitoring and risk prompting of the safety state of the repaired structure, and provides a scientific basis for operation management and maintenance decision-making through multi-level early warning; this helps to improve the safety and reliability of bridge operation, prolong the service life of the bridge and reduce the life cycle cost.

[0095] Embodiment 2

[0096] With reference to Figure 4 The second embodiment of the present application provides a steel structure bridge fatigue crack repaired life prediction system.

[0097] The system comprises a data acquisition module, a feature extraction module, a data fusion module, a residual life prediction module and a multi-level early warning module.

[0098] The data acquisition module is configured to acquire load data and stress data in the operation process of the bridge structure after repairing the fatigue crack.

[0099] The feature extraction module extracts dynamic response characteristic parameters based on the finite element analysis method and in combination with the stress distribution characteristics of the fatigue crack repair area.

[0100] The data fusion module performs multi-dimensional data fusion and preprocessing on the dynamic response characteristic parameters, and constructs a data set for training a deep learning algorithm.

[0101] The residual life prediction module combines a deep learning algorithm and a fracture mechanics theory to construct a crack propagation rate prediction model and a residual life prediction model.

[0102] The multi-level early warning module predicts the residual life of the repaired structure according to the bridge operation state and the performance degradation law of the repair area, and performs multi-level early warning.

[0103] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative and not limiting, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments without departing from the scope of the present application and the scope protected by the claims, and these are all within the protection scope of the present application.

Claims

1. A method for predicting the life of a steel structure bridge after fatigue crack repair, characterized by, The application relates to a method for predicting the crack propagation rate and the residual life of a bridge structure after repairing a fatigue crack. Load data and stress data of the bridge structure in operation after repairing the fatigue crack are collected; Based on the finite element analysis method, dynamic response characteristic parameters are extracted in combination with the stress distribution characteristics of the fatigue crack repair area; The dynamic response characteristic parameters are subjected to multidimensional data fusion and pretreatment, and a data set for training a deep learning algorithm is constructed; In combination with a deep learning algorithm and a fracture mechanics theory, a crack propagation rate prediction model and a residual life prediction model are constructed; A crack propagation rate prediction model is constructed based on a long short-term memory (LSTM) neural network model, and a network structure of the LSTM model is designed, wherein the network structure of the LSTM model comprises an input layer, an LSTM layer, a full connection layer and an output layer; the input layer is used for inputting a training data set; the LSTM layer is composed of a plurality of LSTM units, each of which comprises an input gate, a forgetting gate, an output gate and a memory unit and is used for learning and memorizing features of time sequence data; the full connection layer is used for extracting features output by the LSTM layer and mapping the features to a predicted value of the crack propagation rate; and the output layer adopts a linear activation function and outputs a predicted crack propagation rate value; The LSTM model is trained by using the training data set, and parameters of the model are optimized; a mean square error is adopted as a loss function; in the training process of the LSTM model, the model is verified by using a verification data set; and the LSTM model meeting performance requirements is taken as the crack propagation rate prediction model; The linear elastic fracture mechanics theory is introduced to establish a relationship between the crack propagation rate and a stress intensity factor range; and the Paris formula is adopted to describe the relationship of the crack propagation rate and to calculate the stress intensity factor range; Based on the Paris formula and a calculation formula of the stress intensity factor range, a residual life prediction model is constructed; the predicted crack propagation rate is substituted into the Paris formula, and the residual life prediction value of the repaired structure is obtained through integration of the predicted crack propagation rate. According to the bridge operation state and the performance degradation law of the repair area, the residual life of the repaired structure is predicted, and multi-level early warning is carried out.

2. The method of claim 1, wherein the method is characterized by: Strain sensors and load sensors are arranged on the bridge structure after repairing the fatigue crack, and strain data and load data of the bridge structure in operation are collected in real time; The collected load data and strain data are pretreated, including denoising and abnormal value elimination; The pretreated load data and strain data are constructed into time sequence data.

3. The method for predicting the life of a steel structure bridge after repairing fatigue cracks according to claim 2, characterized in that, According to the collected load data, load working conditions of a finite element model are set; the time history of the load is the same as the collected load time sequence, and the loading frequency of the load is consistent with the sampling frequency; A plurality of load working conditions are set, which correspond to different load levels and loading modes respectively; For each load working condition, a finite element static analysis is carried out to obtain a stress distribution cloud diagram of the fatigue crack repair area; Dynamic response characteristic parameters are extracted based on the stress distribution characteristics of the fatigue crack repair area. The dynamic response characteristic parameters include a stress amplitude, a stress ratio, a stress gradient and a stress triaxiality.

4. The method of claim 3, wherein the method further comprises: The collected load time series data and strain time series data are fused with the extracted dynamic response characteristic parameters in multiple dimensions to construct a fusion data set; The fusion data set is preprocessed, including data normalization, data enhancement and data division; Based on the preprocessed fusion data set, a training deep learning data set for training a deep learning algorithm is constructed, and the training deep learning data set is divided into a training data set, a validation data set and a test data set.

5. The method of claim 4, wherein the method is characterized by: According to the residual life prediction value, the safety state of the repaired structure is multi-level warned; The warning threshold and the alarm threshold are set, when the residual life prediction value is lower than the warning threshold, a warning signal is sent; when the residual life prediction value is lower than the alarm threshold, an alarm signal is sent.

6. A system for predicting the remaining life of a steel structure bridge after repairing a fatigue crack, which is used to implement the method for predicting the remaining life of a steel structure bridge after repairing a fatigue crack according to any one of claims 1 to 5, characterized in that, It comprises: a data acquisition module, a feature extraction module, a data fusion module, a residual life prediction module and a multi-level warning module; The data acquisition module is used to acquire load data and stress data in the operation process of the bridge structure after repairing the fatigue crack; The feature extraction module extracts dynamic response characteristic parameters based on the finite element analysis method and in combination with the stress distribution characteristics of the fatigue crack repair area; The data fusion module performs multi-dimensional data fusion and preprocessing on the dynamic response characteristic parameters to construct a data set for training a deep learning algorithm; The residual life prediction module combines a deep learning algorithm and a fracture mechanics theory to construct a crack propagation rate prediction model and a residual life prediction model; The multi-level warning module predicts the residual life of the repaired structure according to the bridge operation state and the performance degradation law of the repair area, and performs multi-level warning.

7. An electronic device, comprising: It comprises: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; the processor executes the steel structure bridge fatigue crack repair life prediction method of any one of claims 1 to 5 by calling the computer program stored in the memory.

8. A computer-readable storage medium, characterized in that: The instructions are stored, and when the instructions run on the computer, the computer executes the steel structure bridge fatigue crack repair life prediction method of any one of claims 1 to 5.

Images