Bridge damper performance detection method and system based on deep learning

Abstract

The invention belongs to the technical field of damper performance detection, and discloses a method and a system for detecting the performance of a damper for a bridge based on deep learning. The method comprises the following steps: constructing a bridge damper performance detection model and a risk item performance detection verification guidance model; corresponding real-time performance detection working condition parameters and real-time performance detection environment parameters are set for the real-time attribute parameters of the damper for the bridge; detecting and predicting the performance of the damper for the bridge by using the performance detection model of the damper for the bridge; performing risk item performance detection verification guidance by using the risk item performance detection verification guidance model; carrying out risk item performance detection verification on the damper for the bridge; and combining the real-time performance detection prediction result and the real-time risk item performance detection verification result to obtain a real-time performance detection final result of the damper for the bridge. According to the invention, the problems of poor performance detection effect, efficiency and cost limitation and low detection precision in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of damper performance detection, and particularly relates to a method and system for detecting the performance of dampers for bridges based on deep learning. Background Art

[0002] Dampers are important components in bridge structures, used to reduce the vibration of bridges under external forces such as wind and earthquakes. Through performance detection, it can be ensured that the dampers work properly, thus guaranteeing the structural safety of the bridges. Regularly performing performance detection on dampers can timely detect potential problems, such as damper wear and aging, thus preventing accidents caused by damper failures. Through performance detection, the state of the dampers can be evaluated, and maintenance and replacement can be carried out in a timely manner, thus extending the service life of the bridges.

[0003] The existing technologies for detecting the performance of dampers for bridges have the following defects:

[0004] 1) Poor performance detection effect: The performance detection of dampers for many bridges still relies on traditional manual detection methods, which are difficult to comprehensively and accurately evaluate the actual performance of dampers for bridges, and often can only detect specific types of dampers or specific working conditions, unable to comprehensively cover various complex environments and working conditions, resulting in poor performance detection effects;

[0005] 2) Limitations in efficiency and cost: The existing methods for detecting the performance of dampers for bridges are time-consuming and laborious, with low efficiency, difficult to meet the detection requirements of large-scale and high-frequency, with high costs, and as the number of bridges increases, the maintenance costs increase exponentially;

[0006] 3) Low detection accuracy: The existing methods for detecting the performance of dampers for bridges mainly rely on manual detection and simple instrument equipment, with low performance detection accuracy and inaccurate results. As the complexity of the structure, material, and characteristics of dampers for bridges increases, the traditional performance detection technologies can no longer meet the actual needs, and some hidden and deep defects cannot be detected in a timely manner, posing potential safety hazards. Summary of the Invention

[0007] In order to solve the problems of poor performance detection effect, limitations in efficiency and cost, and low detection accuracy existing in the prior art, the purpose of the present invention is to provide a method and system for detecting the performance of dampers for bridges based on deep learning.

[0008] The technical solution adopted by the present invention is as follows:

[0009] A method for detecting the performance of dampers for bridges based on deep learning, comprising the following steps:

[0010] Using deep learning algorithms and reinforcement learning algorithms, construct a performance detection model for dampers used in bridges and a risk item performance detection verification guidance model;

[0011] According to the performance detection requirements, set corresponding real-time performance detection condition parameters and real-time performance detection environment parameters for the real-time damper attribute parameters of the damper used in the bridge to obtain real-time input data;

[0012] According to the real-time input data, use the performance detection model for dampers used in bridges to conduct performance detection prediction of the dampers used in bridges to obtain real-time performance detection prediction results;

[0013] According to the real-time performance detection prediction results, use the risk item performance detection verification guidance model to conduct risk item performance detection verification guidance to obtain real-time risk item performance detection verification strategies;

[0014] According to the real-time risk item performance detection verification strategies, conduct risk item performance detection verification on the dampers used in bridges to obtain real-time risk item performance detection verification results;

[0015] Combine the real-time performance detection prediction results and the real-time risk item performance detection verification results to obtain the final real-time performance detection results of the dampers used in bridges.

[0016] Furthermore, using deep learning algorithms and reinforcement learning algorithms to construct a performance detection model for dampers used in bridges and a risk item performance detection verification guidance model includes the following steps:

[0017] Use CAE simulation software to build a performance detection simulation environment, and based on the performance detection simulation environment, set different historical damper attribute parameters, historical performance detection condition parameters, and historical performance detection environment parameters for the damper used in the bridge, conduct performance detection simulation, and obtain a number of historical performance detection simulation results;

[0018] Combine the historical damper parameters, historical performance detection condition parameters, historical performance detection environment parameters, and historical performance detection simulation results of the same performance detection process to obtain a number of model training samples, and perform dimensionality reduction processing on the number of model training samples to obtain a number of dimensionality-reduced model training samples;

[0019] According to a number of dimensionality-reduced model training samples, use deep learning algorithms to construct a performance detection model for dampers used in bridges to obtain a number of historical performance detection prediction results, and according to the number of historical performance detection prediction results, use reinforcement learning algorithms to construct a risk item performance detection verification guidance model.

[0020] Furthermore, the historical damper attribute parameters include the historical elastic modulus, historical Poisson's ratio, historical density, historical damping coefficient, historical stiffness, historical large damping force, and historical stroke of the damper used in the bridge;

[0021] The real-time property parameters of the damper for bridges include the real-time elastic modulus, real-time Poisson's ratio, real-time density, real-time damping coefficient, real-time stiffness, real-time large damping force, and real-time stroke of the damper for bridges;

[0022] The historical performance detection condition parameters include the historical static load, historical dynamic load, historical impact load, historical vehicle load, and historical self-weight of the bridge of the damper for bridges;

[0023] The real-time performance detection condition parameters include the real-time static load, real-time dynamic load, real-time impact load, real-time vehicle load, and real-time self-weight of the bridge of the damper for bridges;

[0024] The historical performance detection environment parameters include the historical environmental temperature, historical environmental humidity, historical environmental wind speed, historical environmental rainfall, and historical environmental sunshine of the simulated environment;

[0025] The real-time performance detection environment parameters include the real-time environmental temperature, real-time environmental humidity, real-time environmental wind speed, real-time environmental rainfall, and real-time environmental sunshine of the simulated environment.

[0026] Furthermore, combining the historical damper parameters for bridges, historical performance detection condition parameters, historical performance detection environment parameters, and historical performance detection simulation results of the same performance detection process, a number of model training samples are obtained, and dimensionality reduction processing is performed on the number of model training samples to obtain a number of dimensionality-reduced model training samples, including the following steps:

[0027] Combine the historical damper parameters for bridges, historical performance detection condition parameters, and historical performance detection environment parameters of the same performance detection process to obtain a number of historical input data;

[0028] Use the historical performance detection simulation results as the true labels of the corresponding historical input data to obtain a number of model training samples;

[0029] Use the RF algorithm to obtain the importance scores of a number of features of the historical input data in the model training samples for the true labels, and use the top several features with importance scores as the key features;

[0030] According to a number of key features, perform dimensionality reduction processing on the historical input data of each model training sample to obtain a number of dimensionality-reduced model training samples.

[0031] Furthermore, the damper performance detection model for bridges is constructed based on the LSTM-MLP algorithm, and the damper performance detection model for bridges includes a data feature extraction module constructed based on the LSTM algorithm and a performance detection prediction module constructed based on the MLP algorithm connected in sequence.

[0032] Furthermore, the risk item performance detection and verification guidance model is constructed based on the MPO-MOGRPO algorithm, and the verification strategy generation model includes a meta-strategy optimization module constructed based on the MPO algorithm and a verification strategy generation module constructed based on the MOGRPO algorithm. The verification strategy generation module includes an objective function set, a policy network, an experience replay pool, and an agent. The agent is connected to the objective function set and the policy network respectively, and the meta-strategy optimization module is connected to the policy network.

[0033] Furthermore, according to a number of dimension-reduced model training samples, a performance detection model for bridge dampers is constructed using a deep learning algorithm, and a number of historical performance detection prediction results are obtained. Then, according to the number of historical performance detection prediction results, a risk item performance detection and verification guidance model is constructed using a reinforcement learning algorithm, including the following steps:

[0034] Use the LSTM-MLP algorithm to construct an initial performance detection model for bridge dampers, and use the MPO-MOGRPO algorithm to construct an initial risk item performance detection and verification guidance model;

[0035] Use the ISGA algorithm to optimize the model parameters of the initial performance detection model for bridge dampers and the initial risk item performance detection and verification guidance model to obtain an optimized performance detection model for bridge dampers and an optimized risk item performance detection and verification guidance model;

[0036] Input a number of dimension-reduced model training samples, train the optimized performance detection model for bridge dampers to obtain a final performance detection model for bridge dampers and a number of historical performance detection prediction results;

[0037] According to a number of historical performance detection prediction results, train the optimized risk item performance detection and verification guidance model to obtain a final risk item performance detection and verification guidance model and a number of historical verification strategy generation experiences, and store the number of historical verification strategy generation experiences in the experience replay pool.

[0038] Furthermore, according to the real-time input data, use the performance detection model for bridge dampers to perform performance detection prediction of bridge dampers to obtain real-time performance detection prediction results, including the following steps:

[0039] Preprocess the real-time input data to obtain preprocessed real-time input data, and input the preprocessed real-time input data into the performance detection model for bridge dampers;

[0040] Use the data feature extraction module of the performance detection model for bridge dampers to extract the real-time data features of the preprocessed real-time input data;

[0041] Using the performance detection and prediction module of the damper performance detection model for bridges, according to the real-time data characteristics, perform the performance detection and prediction of the damper for bridges to obtain the real-time performance detection and prediction results.

[0042] Furthermore, according to the real-time performance detection and prediction results, use the risk item performance detection verification guidance model to conduct risk item performance detection verification guidance to obtain the real-time risk item performance detection verification strategy, including the following steps:

[0043] According to the real-time performance detection and prediction results, use the meta-strategy optimization module of the risk item performance detection verification guidance model to update the policy network of the verification strategy generation module to obtain the updated policy network;

[0044] Randomly extract several historical verification strategies from the experience replay pool to generate experiences, and update the agent of the verification strategy generation module according to the experiences generated by several historical verification strategies and the real-time performance detection and prediction results to obtain the updated agent;

[0045] Select the real-time objective function from the set of objective functions, and based on the real-time objective function, use the verification strategy generation module with the updated policy network and the updated agent to generate the verification strategy to obtain the real-time risk item performance detection verification strategy.

[0046] A damper performance detection system for bridges based on deep learning is used to implement the damper performance detection method for bridges. The system includes a model construction unit, an input data setting unit, a performance detection and prediction unit, a verification guidance unit, a verification result collection unit, and a final result generation unit.

[0047] The beneficial effects of the present invention are:

[0048] The present invention discloses a damper performance detection method and system for bridges based on deep learning. By adopting deep learning and reinforcement learning algorithms, it can more comprehensively and accurately evaluate the actual performance of dampers for bridges, covering not only various types of dampers but also adapting to various complex environments and working conditions, significantly improving the performance detection effect; realizing automated and intelligent performance detection, greatly reducing manual intervention, improving the detection efficiency, while reducing the labor cost. In addition, after large-scale application, the growth of the maintenance cost will be effectively controlled to achieve cost optimization; using advanced deep learning models, it can accurately identify the tiny performance changes and hidden defects of dampers, including deep defects, thereby improving the detection accuracy and reducing potential safety hazards; according to the real-time risk item performance detection verification strategy, conduct risk item performance detection verification on the dampers for bridges, further improving the accuracy of performance detection.

[0049] Other beneficial effects of the present invention will be further described in the specific implementation manner. Brief Description of the Drawings

[0050] Figure 1 is a flowchart of the method for detecting the performance of bridge dampers based on deep learning in the present invention.

[0051] Figure 2 is a structural block diagram of the system for detecting the performance of bridge dampers based on deep learning in the present invention. Detailed Embodiments

[0052] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0053] Embodiment 1:

[0054] As Figure 1 shown, this embodiment provides a method for detecting the performance of bridge dampers based on deep learning, including the following steps:

[0055] S1: Using deep learning algorithms and reinforcement learning algorithms, construct a performance detection model for bridge dampers and a risk item performance detection verification guidance model, including the following steps:

[0056] S1-1: Use computer-aided engineering (CAE) simulation software to build a performance detection simulation environment, and based on the performance detection simulation environment, set different historical bridge damper attribute parameters, historical performance detection condition parameters, and historical performance detection environment parameters, perform performance detection simulations, and obtain a number of historical performance detection simulation results;

[0057] Perform performance detection simulations through CAE simulation software in the performance detection simulation environment to obtain a large amount of performance detection-related data, providing data support for subsequent model training;

[0058] The historical bridge damper attribute parameters include the historical elastic modulus, historical Poisson's ratio, historical density, historical damping coefficient, historical stiffness, historical large damping force, and historical stroke of the bridge damper;

[0059] The historical performance detection condition parameters include the historical static load, historical dynamic load, historical impact load, historical vehicle load, and historical bridge self-weight of the bridge damper;

[0060] The historical performance detection environment parameters include the historical environmental temperature, historical environmental humidity, historical environmental wind speed, historical environmental rainfall, and historical environmental sunshine of the simulation environment;

[0061] S1-2: Combine the historical damper parameters for bridges, historical performance detection condition parameters, historical performance detection environment parameters, and historical performance detection simulation results of the same performance detection process to obtain a number of model training samples, and perform dimensionality reduction processing on the number of model training samples to obtain a number of dimension-reduced model training samples, including the following steps:

[0062] S1-2-1: Combine the historical damper parameters for bridges, historical performance detection condition parameters, and historical performance detection environment parameters of the same performance detection process to obtain a number of historical input data;

[0063] S1-2-2: Use the historical performance detection simulation results as the true labels of the corresponding historical input data to obtain a number of model training samples;

[0064] S1-2-3: Use the Random Forest (RF) algorithm to obtain the importance scores of a number of features of the historical input data in the model training samples for the true labels, and use the top several features with importance scores as the key features;

[0065] S1-2-4: According to a number of key features, perform dimensionality reduction processing on the historical input data of each model training sample to obtain a number of dimension-reduced model training samples;

[0066] S1-3: According to a number of dimension-reduced model training samples, use a deep learning algorithm to construct a performance detection model for dampers used in bridges to obtain a number of historical performance detection prediction results, and use a reinforcement learning algorithm according to a number of historical performance detection prediction results to construct a performance detection verification guidance model for risk items, including the following steps:

[0067] S1-3-1: Use the Long Short-Term Memory (LSTM)-Multilayer Perceptron (MLP) algorithm to construct an initial performance detection model for dampers used in bridges, and use the Meta-Policy Optimization (MPO)-Multi-Objective Group Relative Policy Optimization (MOGRPO) algorithm to construct an initial performance detection verification guidance model for risk items;

[0068] The performance detection model for dampers used in bridges is constructed based on the LSTM-MLP algorithm, and the performance detection model for dampers used in bridges includes a data feature extraction module constructed based on the LSTM algorithm and a performance detection prediction module constructed based on the MLP algorithm that are connected in sequence;

[0069] The LSTM can capture the complex relationship between the data and the performance detection results, improve the model's ability to understand the data, provide more accurate input features for the performance detection prediction module, and enhance the accuracy of performance detection prediction; the performance detection prediction module MLP further processes and analyzes the extracted data features, learns the complex mapping relationship between the input features and the output labels, and utilizes the powerful non-linear modeling ability of the MLP to improve the accuracy of performance detection prediction;

[0070] The risk item performance detection verification guidance model is constructed based on the MPO-MOGRPO algorithm, and the verification strategy generation model includes a meta-strategy optimization module constructed based on the MPO algorithm and a verification strategy generation module constructed based on the MOGRPO algorithm. The verification strategy generation module includes an objective function set, a policy network, an experience replay pool, and an agent. The agent is respectively connected to the objective function set and the policy network, and the meta-strategy optimization module is connected to the policy network;

[0071] The meta-strategy optimization module is used to verify the network parameters of the policy network in the verification strategy generation module so that these parameters can quickly adapt to new and unseen performance detection prediction results, improving the generalization ability of the model. Even in the case of unseen performance detection prediction results, the policy network can be updated based on previous learning experiences, improving the adaptability of the verification strategy generation model; the objective function set of the verification strategy generation module can handle multiple conflicting objectives, such as minimizing the performance detection cost, maximizing the performance detection accuracy, maximizing the performance detection efficiency, etc., and generate verification strategies that balance these objectives. The agent learns historical verification strategies through the experience replay pool, continuously optimizing its own strategy generation ability. The agent controls the policy network according to the learned experience to generate more effective verification strategies. The design of the experience replay pool and the agent enables the model to continuously learn and optimize, improving the quality of strategy generation. Since the verification strategy generation module adopts a group exploration method, it can avoid falling into local optimal solutions to a certain extent. The policy network outputs the distribution probability of actions in a given state, and the verification strategy generation module directly updates the policy network through gradients, eliminating the Critic model in traditional reinforcement learning and making the algorithm structure more concise;

[0072] S1-3-2: Use the Improved Snow Geese Algorithm (ISGA) to optimize the model parameters of the initial performance detection model of bridge dampers and the initial risk item performance detection verification guidance model, and obtain the optimized performance detection model of bridge dampers and the optimized risk item performance detection verification guidance model, including the following steps:

[0073] S1-3-2-1: Integrate the first initial model parameters of the initial bridge damper performance detection model and the second initial model parameters of the initial risk item performance detection and verification guidance model to obtain comprehensive initial model parameters;

[0074] S1-3-2-2: Encode the comprehensive initial model parameters into the individual vectors of the ISGA individuals in the ISGA algorithm, and set the ISGA population parameters and the maximum number of iterations;

[0075] S1-3-2-3: Take minimizing the error value as the optimization goal, and set the fitness function of the swarm intelligence optimization algorithm according to the optimization goal;

[0076] The formula is:

[0077] Fit(P)=minMSN(P)

[0078] Where Fit(P) is the fitness function; MSN(P) is the mean square error function; P is the ISGA individual;

[0079] S1-3-2-4: Based on the individual vectors of the ISGA individuals and the ISGA population parameters, the Circle chaotic mapping sequence is used to generate initial solutions and obtain several initial solutions; each initial solution corresponds to an initial ISGA individual, and several initial ISGA individuals constitute the initial ISGA individual population;

[0080] The formula is:

[0081]

[0082] Where, P i The initial ISGA individual generated by the Circle chaotic map sequence, that is, the initial solution; P o is a randomly generated ISGA individual; i is the ISGA individual indicator; mod(*) is the remainder function;

[0083] Compared with the randomly distributed population, the initial position distribution of the improved ISGA population generated by the Circle chaotic mapping sequence is more uniform, which expands the search range of the ISGA population in space and increases the diversity of group positions. To a certain extent, it improves the defect of the algorithm that it is easy to fall into local extreme values, thereby improving the optimization efficiency of the algorithm.

[0084] S1-3-2-5: Use the fitness function to obtain the initial fitness value of each initial ISGA individual in the initial ISGA population, and take the initial ISGA individual with the lowest fitness value as the leader goose;

[0085] S1-3-2-6: Enter the exploration stage, introduce the leading goose rotation mechanism, the call guidance mechanism, and the dynamic reverse mechanism, iterate and update the initial ISGA population, obtain the once-updated ISGA population, and retain the optimal individual;

[0086] The leading goose rotation mechanism: In each iteration, compete according to the fitness value of the ISGA individuals to select a new leading goose. This mechanism can prevent the leading goose from falling into local optimality prematurely and enhance the global search ability of the algorithm;

[0087] The formula is:

[0088]

[0089] In the formula, P i t+1 is the once-updated leading goose; is the ISGA individual with the third-to-last fitness value in the initial ISGA population at the t-th and (t + 1)-th iteration times; is the ISGA individual with the fifth-to-last fitness value in the initial ISGA population at the t-th iteration time; t is the current iteration time; is the optimal individual; a is the first weight factor; rand is a random number; P i t is the initial leading goose;

[0090] The call guidance mechanism: According to the distance between the ISGA individuals and the leading goose, use the sound wave propagation attenuation model to adjust the individual position update. For the ISGA individuals with a relatively short distance, their position update is more affected by the leading goose and can quickly approach the optimal solution. For the ISGA individuals with a relatively long distance, their position update is less affected by the leading goose and can maintain a certain exploration ability. This mechanism can prevent the group from over-aggregating or dispersing and improve the local search accuracy of the algorithm;

[0091] The formula is:

[0092]

[0093] In the formula, is the once-updated ISGA individual; is the initial ISGA individual at the t-th iteration time; is the initial ISGA individual the received sound intensity; is the initial ISGA individual at the t-th iteration time the corresponding sound intensity parameter; L WA is the initial sound intensity; L low is the lowest receivable sound intensity; a" is the convergence factor; is the initial ISGA individual with the farthest distance; r' is a random parameter; B(d) is a Brownian motion function; d is a Brownian motion parameter; is the exclusive OR processing symbol;

[0094]

[0095] In the formula, a" is the convergence factor; tanh(.) is the hyperbolic tangent function; t is the current iteration number; t max is the maximum iteration number; a max and a min are the maximum and minimum values of the convergence factor respectively; λ is the decreasing rate parameter, k' is the decreasing period parameter, λ = -2π, k' = π;

[0096] The dynamic reverse mechanism performs dynamic reverse on the initial ISGA individual to improve the diversity of exploration directions and avoid falling into local optima;

[0097] The formula is:

[0098]

[0099] In the formula, is the reverse ISGA individual updated once; γ is the decreasing inertia coefficient; L max and L min are the maximum and minimum values of the vector space respectively;

[0100] Integrate the leading goose updated once, several ISGA individuals updated once, and several reverse ISGA individuals updated once to obtain the ISGA population updated once, and retain the ISGA individual with the lowest fitness value as the optimal individual;

[0101] S1-3-2-7: Enter the development stage, introduce the abnormal boundary strategy and the Gaussian mutation mechanism to perform a second update on the ISGA population updated once to obtain the ISGA population updated twice, and retain the optimal individual;

[0102] The abnormal boundary strategy calculates the difference between the fitness value of each ISGA individual updated once and the average fitness value of the group. For ISGA individuals with fitness values much higher than the average value of the group, their position update methods will be adjusted, such as using the Gaussian mutation mechanism, a larger step size or a smaller step size. This mechanism can help individuals avoid falling into local optima and improve the convergence speed and accuracy of the algorithm;

[0103] The formula is:

[0104]

[0105] In the formula, is the ISGA individual updated twice; The ISGA individual for one update; Fit(*) is the fitness function; Fit avg is the average fitness value of the population; is the ISGA individual with the highest fitness value; a' and e are the second weight factor and the third weight factor; G(1,1) is the parameter of the Gaussian mutation mechanism;

[0106] S1-3-2-8: If the number of iterations is greater than or equal to the iteration threshold or the fitness value of the optimal individual is less than the fitness threshold, then output the optimal individual as the optimal solution;

[0107] S1-3-2-9: Decode the individual vector of the optimal individual to obtain the optimal comprehensive initial model parameters, and optimize the model parameters of the initial performance detection model of the bridge damper and the initial performance detection verification guidance model of the risk item according to the optimal comprehensive initial model parameters to obtain the optimized performance detection model of the bridge damper and the optimized performance detection verification guidance model of the risk item;

[0108] S1-3-3: Input several model training samples after dimensionality reduction, and train the optimized performance detection model of the bridge damper to obtain the final performance detection model of the bridge damper and several historical performance detection prediction results;

[0109] S1-3-4: Train the optimized performance detection verification guidance model of the risk item according to several historical performance detection prediction results to obtain the final performance detection verification guidance model of the risk item and several historical verification strategy generation experiences, and store the several historical verification strategy generation experiences in the experience replay pool, including the following steps:

[0110] S1-3-4-1: Take the verification strategy generation optimization goal as the scenario of meta-strategy optimization, and train the optimized meta-strategy optimization module of the performance detection verification guidance model of the risk item according to several historical performance detection prediction results to obtain the final meta-strategy optimization module;

[0111] S1-3-4-2: Use the final meta-strategy optimization module to initialize the policy network of the initial verification strategy generation module in different scenarios to obtain the initialized policy network;

[0112] S1-3-4-3: According to different scenarios of the meta-strategy optimization module, that is, the optimization goal of the verification strategy, set the objective function set for the optimized performance detection verification guidance model of the risk item with the initialized policy network, set the experience replay pool, and set the action space and state space for the agent of the optimized verification strategy generation module;

[0113] S1-3-4-4: Use the verification strategy generation problem as a simulation environment, and based on the initialized policy network and the agent with the action space and state space set, obtain the updated verification strategy generation module;

[0114] S1-3-4-5: Traverse all the objective functions in the objective function set, and based on several historical performance detection prediction results, train the updated verification strategy generation module to obtain the final verification strategy generation module, and generate several historical verification strategy generation experiences;

[0115] S1-3-4-5: Integrate the final meta-policy optimization module and the final verification strategy generation module to obtain the final verification strategy generation model, and store several historical verification strategy generation experiences in the experience replay pool;

[0116] S2: According to the performance detection requirements, set the corresponding real-time performance detection working condition parameters and real-time performance detection environment parameters for the real-time damper property parameters of the bridge damper to obtain real-time input data;

[0117] The real-time damper property parameters of the bridge damper include the real-time elastic modulus, real-time Poisson's ratio, real-time density, real-time damping coefficient, real-time stiffness, real-time large damping force, and real-time stroke of the bridge damper;

[0118] The real-time performance detection working condition parameters include the real-time static load, real-time dynamic load, real-time impact load, real-time vehicle load, and real-time bridge self-weight of the bridge damper;

[0119] The real-time performance detection environment parameters include the real-time environmental temperature, real-time environmental humidity, real-time environmental wind speed, real-time environmental rainfall, and real-time environmental sunshine of the simulation environment;

[0120] S3: According to the real-time input data, use the bridge damper performance detection model to perform bridge damper performance detection prediction to obtain real-time performance detection prediction results, including the following steps:

[0121] S3-1: Preprocess the real-time input data to obtain the preprocessed real-time input data, and input the preprocessed real-time input data into the bridge damper performance detection model;

[0122] S3-2: Use the data feature extraction module of the bridge damper performance detection model to extract the real-time data features of the preprocessed real-time input data;

[0123] S3-3: Use the performance detection prediction module of the bridge damper performance detection model to perform bridge damper performance detection prediction according to the real-time data features to obtain real-time performance detection prediction results;

[0124] The real-time performance detection and prediction results include real-time voltage withstand performance prediction results, real-time slow speed and speed-related performance prediction results, real-time frequency-related performance prediction results, real-time temperature-related performance prediction results, real-time seismic action performance prediction results, real-time wind vibration load performance prediction results, and real-time fatigue and wear resistance performance prediction results;

[0125] S4: According to the real-time performance detection and prediction results, use the risk item performance detection and verification guidance model to conduct risk item performance detection and verification guidance, and obtain the real-time risk item performance detection and verification strategy, including the following steps:

[0126] S4-1: According to the real-time performance detection and prediction results, use the meta-strategy optimization module of the risk item performance detection and verification guidance model to update the policy network of the verification strategy generation module, and obtain the updated policy network;

[0127] S4-2: Randomly extract a number of historical verification strategies from the experience replay pool to generate experiences, and update the agent of the verification strategy generation module according to the experiences generated by a number of historical verification strategies and the real-time performance detection and prediction results, and obtain the updated agent, including the following steps:

[0128] S4-2-1: Randomly extract a number of historical verification strategies from the experience replay pool to generate experiences, generate a number of possible verification actions according to the experiences generated by a number of historical verification strategies, and update the action space of the agent of the verification strategy generation module according to the number of possible verification actions, and obtain the updated action space;

[0129] S4-2-2: Analyze the real-time performance detection and prediction results to obtain a number of real-time performance detection and prediction states, and update the state space of the agent of the verification strategy generation module according to the real-time performance detection and prediction states, and obtain the updated state space;

[0130] S4-2-3: Integrate the updated state space and the updated action space of the agent to obtain the updated agent;

[0131] S4-3: Select the real-time objective function from the set of objective functions, and based on the real-time objective function, use the verification strategy generation module set with the updated policy network and the updated agent to conduct verification strategy generation, and obtain the real-time risk item performance detection and verification strategy, including the following steps:

[0132] S4-3-1: Select the real-time objective function from the set of objective functions of the verification strategy generation module, and based on the real-time objective function, use the updated agent of the verification strategy generation module to control the updated policy network, and generate the probability distribution of all possible verification actions in the updated action space corresponding to each real-time performance detection and prediction state in the updated state space;

[0133] S4-3-2: Use the possible verification action with the highest probability distribution in the updated action space as the execution verification action for the corresponding real-time performance detection prediction state;

[0134] S4-3-3: Integrate the execution verification actions of all real-time performance detection prediction states in the updated state space to obtain a real-time risk item performance detection verification strategy;

[0135] The real-time risk item performance detection verification strategy includes real-time risk item selection decision-making, real-time risk item performance detection verification decision-making, real-time performance detection working condition parameter adjustment decision-making, and real-time performance detection environment parameter adjustment decision-making;

[0136] S5: According to the real-time risk item performance detection verification strategy, conduct risk item performance detection verification on the bridge damper to obtain the real-time risk item performance detection verification result;

[0137] S6: Combine the real-time performance detection prediction result and the real-time risk item performance detection verification result to obtain the final real-time performance detection result of the bridge damper.

[0138] Embodiment 2:

[0139] As Figure 2 shown, this embodiment provides a performance detection system for bridge dampers based on deep learning, which is used to implement the performance detection method for bridge dampers. The system includes a model construction unit, an input data setting unit, a performance detection prediction unit, a verification guidance unit, a verification result acquisition unit, and a final result generation unit;

[0140] The model construction unit is used to construct a performance detection model for bridge dampers and a risk item performance detection verification guidance model using deep learning algorithms and reinforcement learning algorithms;

[0141] The input data setting unit is used to set corresponding real-time performance detection working condition parameters and real-time performance detection environment parameters for the real-time bridge damper attribute parameters of the bridge damper according to the performance detection requirements to obtain real-time input data;

[0142] The performance detection prediction unit is used to perform bridge damper performance detection prediction using the bridge damper performance detection model according to the real-time input data to obtain a real-time performance detection prediction result;

[0143] The verification guidance unit is used to perform risk item performance detection verification guidance using the risk item performance detection verification guidance model according to the real-time performance detection prediction result to obtain a real-time risk item performance detection verification strategy;

[0144] A verification result acquisition unit, configured to acquire real-time risk item performance detection and verification results obtained by performing risk item performance detection and verification on a bridge damper according to a real-time risk item performance detection and verification strategy;

[0145] A final result generation unit, configured to combine real-time performance detection prediction results and real-time risk item performance detection and verification results to obtain the final real-time performance detection result of the bridge damper.

[0146] The present invention discloses a method and system for performance detection of a bridge damper based on deep learning. By adopting deep learning and reinforcement learning algorithms, it can more comprehensively and accurately evaluate the actual performance of the bridge damper, covering not only various types of dampers but also adapting to a variety of complex environments and working conditions, significantly improving the performance detection effect; realizing automated and intelligent performance detection, greatly reducing manual intervention, improving detection efficiency, while reducing labor costs. In addition, after large-scale application, the growth of maintenance costs will be effectively controlled, achieving cost optimization; using an advanced deep learning model, it can accurately identify small performance changes and hidden defects of the damper, including deep defects, thereby improving detection accuracy and reducing potential safety hazards; according to the real-time risk item performance detection and verification strategy, performing risk item performance detection and verification on the bridge damper further improves the accuracy of performance detection.

[0147] The present invention is not limited to the above optional embodiments, and anyone can obtain other various forms of products under the inspiration of the present invention. The above specific embodiments should not be construed as limiting the protection scope of the present invention, and the protection scope of the present invention should be defined by the claims, and the description can be used to interpret the claims.

Claims

1. A performance detection method for bridge dampers based on deep learning, characterized in that: It includes the following steps: Using deep learning algorithms and reinforcement learning algorithms, construct a performance detection model for bridge dampers and a risk item performance detection verification guidance model; According to the performance detection requirements, set corresponding real-time performance detection working condition parameters and real-time performance detection environment parameters for the real-time bridge damper attribute parameters of the bridge damper to obtain real-time input data; According to the real-time input data, use the performance detection model for bridge dampers to conduct performance detection prediction of bridge dampers to obtain real-time performance detection prediction results; According to the real-time performance detection prediction results, use the risk item performance detection verification guidance model to conduct risk item performance detection verification guidance to obtain real-time risk item performance detection verification strategies; According to the real-time risk item performance detection verification strategy, conduct risk item performance detection verification on the bridge damper to obtain real-time risk item performance detection verification results; Combining the real-time performance detection prediction results and the real-time risk item performance detection verification results, obtain the final real-time performance detection results of the bridge damper.

2. The performance detection method of a damper for bridges based on deep learning according to claim 1, characterized in that: Using deep learning algorithms and reinforcement learning algorithms to construct a performance detection model for bridge dampers and a risk item performance detection verification guidance model includes the following steps: Use CAE simulation software to build a performance detection simulation environment, and based on the performance detection simulation environment, set different historical bridge damper attribute parameters, historical performance detection working condition parameters, and historical performance detection environment parameters, and conduct performance detection simulation to obtain several historical performance detection simulation results; Combining the historical bridge damper parameters, historical performance detection working condition parameters, historical performance detection environment parameters, and historical performance detection simulation results of the same performance detection process, obtain several model training samples, and perform dimensionality reduction processing on the several model training samples to obtain several dimensionality-reduced model training samples; According to several dimensionality-reduced model training samples, use deep learning algorithms to construct a performance detection model for bridge dampers to obtain several historical performance detection prediction results, and according to the several historical performance detection prediction results, use reinforcement learning algorithms to construct a risk item performance detection verification guidance model.

3. The method for detecting the performance of a damper for bridges based on deep learning according to claim 2, wherein: The historical bridge damper attribute parameters include the historical elastic modulus, historical Poisson's ratio, historical density, historical damping coefficient, historical stiffness, historical large damping force, and historical stroke of the bridge damper; The real-time bridge damper attribute parameters include the real-time elastic modulus, real-time Poisson's ratio, real-time density, real-time damping coefficient, real-time stiffness, real-time large damping force, and real-time stroke of the bridge damper; The historical performance detection working condition parameters include the historical static load, historical dynamic load, historical impact load, historical vehicle load, and historical bridge self-weight of the bridge damper; The real-time performance detection working condition parameters include the real-time static load, real-time dynamic load, real-time impact load, real-time vehicle load, and real-time bridge self-weight of the bridge damper; The historical performance detection environment parameters include the historical environmental temperature, historical environmental humidity, historical environmental wind speed, historical environmental rainfall, and historical environmental sunshine of the simulation environment; The real-time performance detection environment parameters described above include the real-time environmental temperature, real-time environmental humidity, real-time environmental wind speed, real-time environmental rainfall, and real-time environmental sunshine of the simulation environment.

4. The performance detection method of a damper for bridges based on deep learning according to claim 3, characterized in that: Combining the historical damper parameters for bridges, historical performance detection condition parameters, historical performance detection environment parameters, and historical performance detection simulation results of the same performance detection process, obtaining a number of model training samples, and performing dimensionality reduction processing on the number of model training samples to obtain a number of dimension-reduced model training samples, including the following steps: Combining the historical damper parameters for bridges, historical performance detection condition parameters, and historical performance detection environment parameters of the same performance detection process to obtain a number of historical input data; Taking the historical performance detection simulation results as the true labels of the corresponding historical input data to obtain a number of model training samples; Using the RF algorithm to obtain the importance scores of a number of features of the historical input data in the model training samples for the true labels, and taking the top several features with importance scores as the key features; According to a number of key features, performing dimensionality reduction processing on the historical input data of each model training sample to obtain a number of dimension-reduced model training samples.

5. A performance detection method for a bridge damper based on deep learning according to claim 4, characterized in that: The performance detection model for dampers used in bridges described above is constructed based on the LSTM-MLP algorithm, and the performance detection model for dampers used in bridges includes a data feature extraction module constructed based on the LSTM algorithm and a performance detection prediction module constructed based on the MLP algorithm connected in sequence.

6. The performance detection method of a damper for bridges based on deep learning according to claim 5, wherein: The risk item performance detection verification guidance model described above is constructed based on the MPO-MOGRPO algorithm, and the verification strategy generation model includes a meta-strategy optimization module constructed based on the MPO algorithm and a verification strategy generation module constructed based on the MOGRPO algorithm. The verification strategy generation module includes a set of objective functions, a policy network, an experience replay pool, and an agent. The agent is respectively connected to the set of objective functions and the policy network, and the meta-strategy optimization module is connected to the policy network.

7. A performance detection method for a bridge damper based on deep learning according to claim 6, characterized in that: According to a number of dimension-reduced model training samples, using a deep learning algorithm to construct a performance detection model for dampers used in bridges, obtaining a number of historical performance detection prediction results, and according to a number of historical performance detection prediction results, using a reinforcement learning algorithm to construct a risk item performance detection verification guidance model, including the following steps: Using the LSTM-MLP algorithm to construct an initial performance detection model for dampers used in bridges, and using the MPO-MOGRPO algorithm to construct an initial risk item performance detection verification guidance model; Using the ISGA algorithm to optimize the model parameters of the initial performance detection model for dampers used in bridges and the initial risk item performance detection verification guidance model to obtain an optimized performance detection model for dampers used in bridges and an optimized risk item performance detection verification guidance model; Inputting a number of dimension-reduced model training samples to train the optimized performance detection model for dampers used in bridges to obtain a final performance detection model for dampers used in bridges and a number of historical performance detection prediction results; Based on the prediction results of several historical performance detections, train the optimized risk item performance detection and verification guidance model to obtain the final risk item performance detection and verification guidance model and several historical verification strategy generation experiences, and store the several historical verification strategy generation experiences in the experience replay pool.

8. A performance detection method for a bridge damper based on deep learning according to claim 7, characterized in that: According to the real-time input data, use the performance detection model of bridge dampers to perform performance detection and prediction of bridge dampers, and obtain the real-time performance detection prediction results, including the following steps: Preprocess the real-time input data to obtain the preprocessed real-time input data, and input the preprocessed real-time input data into the performance detection model of bridge dampers; Use the data feature extraction module of the performance detection model of bridge dampers to extract the real-time data features of the preprocessed real-time input data; Use the performance detection and prediction module of the performance detection model of bridge dampers to perform performance detection and prediction of bridge dampers according to the real-time data features, and obtain the real-time performance detection prediction results.

9. A performance detection method for a bridge damper based on deep learning according to claim 8, characterized in that: According to the real-time performance detection prediction results, use the risk item performance detection and verification guidance model to perform risk item performance detection and verification guidance, and obtain the real-time risk item performance detection and verification strategy, including the following steps: According to the real-time performance detection prediction results, use the meta-strategy optimization module of the risk item performance detection and verification guidance model to update the policy network of the verification strategy generation module to obtain the updated policy network; Randomly extract several historical verification strategy generation experiences from the experience replay pool, and update the agent of the verification strategy generation module according to the several historical verification strategy generation experiences and the real-time performance detection prediction results to obtain the updated agent; Select the real-time objective function from the set of objective functions, and based on the real-time objective function, use the verification strategy generation module set with the updated policy network and the updated agent to generate the verification strategy to obtain the real-time risk item performance detection and verification strategy.

10. A performance detection system for bridge dampers based on deep learning, which is used to implement the performance detection method for bridge dampers described in any one of claims 1-9, and is characterized in that: The described system includes a model construction unit, an input data setting unit, a performance detection and prediction unit, a verification guidance unit, a verification result collection unit, and a final result generation unit.

Images