Method for generating bond-slip model of interface between FRP sheet and concrete based on WGAN

Abstract

The application provides a WGAN-based FRP sheet and concrete interface bonding slip model generation method, which can automatically, quickly and accurately obtain the bonding slip model and has a wide application prospect. After the automatic and rapid acquisition of the bonding slip model, the bonding performance between the FRP sheet and the concrete can be accurately reflected, and a safe and reliable reinforcement component design calculation method can be established. The WGAN is used to predict the strain, so that the training process can be simplified and the training can be stable. The WGAN can replace the traditional experimental analysis to quickly and accurately establish a bonding strength model. The LSTM is used as the generator of the WGAN model, so that the gradient disappearance and explosion problems of the RNN are solved, and the strain data related to time can be more accurately predicted. The CNN is used as the discriminator of the WGAN model, so that the quality and convergence speed of the generated samples are improved.

Description

Technical Field

[0001] This invention belongs to the field of FRP sheet-concrete interface bonding performance analysis, and in particular relates to a method for generating an FRP sheet-concrete interface bonding slip model based on WGAN, which can be applied to the design of reinforced components. Background Technology

[0002] Externally bonded FRP (fiberglass reinforced polymer) reinforcement is widely used in reinforced concrete structure reinforcement projects due to its advantages such as applicability to corrosive environments, lightweight material, and suitability for reinforcing complex cross-section structures. Numerous experimental studies have shown that premature delamination at the FRP-concrete interface is a key factor leading to failure in externally bonded FRP-reinforced structures. Therefore, studying the interfacial bond performance between FRP and concrete and establishing a bond-slip model characterizing interfacial delamination failure behavior is of significant engineering importance for improving the reliability of structural reinforcement. Furthermore, it can help to automatically and quickly establish safe and reliable design and calculation methods for FRP-reinforced components. Summary of the Invention

[0003] This invention provides a method for generating a bond-slip model of the FRP sheet-concrete interface based on WGAN, which can automatically, quickly and accurately obtain the bond-slip model and has broad application prospects.

[0004] Technical solution of the present invention:

[0005] A method for generating a bond-slip model at the interface between FRP sheet and concrete based on WGAN, characterized by comprising:

[0006] Step 1: Acquisition of strain data samples corresponding to the location changes, and preprocessing method of training set;

[0007] Step 2, construct a prediction model based on WGAN;

[0008] Step 3: Evaluate the training and testing results of the network;

[0009] Step 4: Fit the predicted strain data at each location at the same time with the definition of the interface bond-slip model, and plot multiple curves corresponding to each definition in the bond-slip model, which can be applied to the design of safe and reliable reinforced components.

[0010] The method for generating the bond slip model of FRP sheet and concrete interface based on WGAN is characterized in that, step 1: the finite element software LS-DYNA is used to model and realize the data of strain data acquisition every two seconds, that is, the average strain ε;

[0011] In addition, five parameters were selected, namely the compressive strength f of the concrete. c Elastic modulus Es The thickness t of the FRP board f Ultimate strength f s Yield strength f d ;

[0012] Organize the data for the above six parameters and create the corresponding training and test sets for each position.

[0013] The method for generating a bond-slip model of FRP sheet-concrete interface based on WGAN is characterized in that step 2:

[0014] When building the WGAN model, an LSTM network was chosen as the generator and a CNN as the discriminator. The generator uses a sliding window method for prediction, predicting the strain data for the next 3 time series every 30 time series data. The strain data corresponding to the last 3 time series generated by the generator and the strain data corresponding to the first 30 time series of real time series are combined to form a fake dataset and the strain data corresponding to the real 33 time series, and then passed to the discriminator to identify the authenticity.

[0015] The WGAN model uses gradient penalty to enforce Lipschitz constraints.

[0016] The discriminator uses Leaky ReLU as its activation function.

[0017] The discriminator loss function is defined as follows:

[0018]

[0019] In equation (3), m represents the sample size, and x i y represents the input data of the generator. i G(x) represents the target value of the real dataset. i () represents the generated data value of the generator, D represents the discrimination calculation of the discriminator on the data, and λ is a penalty coefficient with a value of 10. Indicates a penalty item;

[0020] The generator loss function is defined as follows:

[0021]

[0022] The method for generating the bond-slip model of FRP sheet and concrete interface based on WGAN is characterized in that step 3: the root mean square error (RMSE) is used to evaluate the training and testing results of the network.

[0023] The method for generating the bond-slip model of FRP sheet and concrete interface based on WGAN is characterized by the following method for drawing each curve in the bond-slip model: the unknown parameter information is obtained by fitting the predicted strain data at each position at the same time with the formula (6) for the distribution of FRP strain with position, and the obtained parameter information is substituted into the definitions (7)(8)(9) of the bond-slip relationship to obtain each curve in the bond-slip model.

[0024]

[0025]

[0026]

[0027]

[0028] Advantages of the present invention

[0029] This invention proposes a method for generating a bond-slip model at the interface between FRP sheets and concrete based on WGAN, which has the following advantages:

[0030] 1. Using WGAN to predict strain simplifies the training process and achieves stable training. It can replace traditional experimental analysis to quickly and accurately establish a bond strength model.

[0031] Second, by using LSTM as the generator of the WGAN model, the gradient vanishing and exploding problems of RNN are solved, which can more accurately predict time-related strain data.

[0032] Third, by using CNN as the discriminator in the WGAN model, the quality of generated samples and the convergence speed were improved.

[0033] The application value of this invention:

[0034] Further applications: After automatically and quickly acquiring the bond slip model, this invention can accurately reflect the bond performance between FRP sheets and concrete, helping to establish a safe and reliable design and calculation method for reinforced components. Attached Figure Description

[0035] Figure 1 This is the overall flowchart of the present invention;

[0036] Figure 2 This is a schematic diagram of strain data acquisition according to the present invention;

[0037] Figure 3 The dataset format and content of this invention;

[0038] Figure 4This is a schematic diagram of the WGAN model.

[0039] Figure 5(a) is a graph showing the change of slip amount with position generated according to an embodiment of the present invention;

[0040] Figure 5(b) shows the variation of adhesive stress with location in an embodiment of the present invention;

[0041] Figure 5(c) is a bonding slip curve generated according to an embodiment of the present invention. Detailed Implementation

[0042] Its detailed process is as follows Figure 1 As shown. Detailed steps are as follows:

[0043] Step 1: Use the finite element method software LS-DYNA to model the bond behavior between the FRP sheet and concrete to collect strain data of the FRP sheet. Through simulation, data corresponding to the strain at each position every two seconds was obtained within a certain period after loading. The strain data acquisition method is as follows: Figure 2 As shown in the figure. Pn represents the strain data ε at each point selected at equal intervals on the FRP sheet under the action of an external load F, which is collected every two seconds.

[0044] The strain data ε obtained at each location over time is compared with the elastic modulus E of the concrete. s The compressive strength f of concrete c FRP thickness t f Ultimate strength f of FRP s The yield strength f of FRP d this Five kinds The parameters are combined into multiple datasets according to their different positions. The format and content of the datasets are as follows: Figure 3 As shown. In each dataset, the first 70% of the data in time series is used as the training set, and the last 30% is used as the test set.

[0045] Step 2: Build a machine learning prediction model based on WGAN. WGAN was proposed to solve the problem of unstable training in GANs, eliminating the need for careful balancing of the training levels of the generator and discriminator. The Wasserstein distance is WGAN's solution to this problem. The Wasserstein distance is defined as follows:

[0046]

[0047] In equation (1), W(P) r ,P g ) represents the probability distribution P r With P gThe distance between them, where x and y represent two random variables of the same dimension, ∏(P r ,P g ) represents P r and P g All possible joint probability distributions, γ represents the probability distribution in ∏(P r ,P g For any joint probability distribution in E, (x,y)～γ [‖xy‖] represents P r and P g Expected value of distance, This indicates the lower bound that the expected value can take.

[0048] When building the WGAN model, an LSTM network was chosen as the generator and a CNN as the discriminator. The structure of the WGAN model is as follows: Figure 4 As shown, a sliding window approach is used in the generator for prediction, where every 30 time series data points can predict the strain data for the next 3 time series. The fake dataset, consisting of the strain data from the last 3 time series generated by the generator and the strain data from the first 30 real time series, along with the real strain data from the 33 time series, is simultaneously fed to the discriminator for authentication. The discriminator uses the Leaky ReLU activation function, the expression of which is as follows:

[0049]

[0050] In equation (2), a is a number close to 0. In this invention, a = 0.01. This represents the input feature vector, which is the strain data with 33 time series passed to the discriminator.

[0051] The WGAN of this invention uses gradient penalty to enforce Lipschitz constraints. The discriminator loss function is defined as follows:

[0052]

[0053] In equation (3), m represents the sample size, and x i y represents the input data of the generator. i G(x) represents the target value of the real dataset. i () represents the generated data value of the generator, D represents the discrimination calculation of the discriminator on the data, and λ is a penalty coefficient with a value of 10. This indicates a penalty.

[0054] The generator loss function is defined as follows:

[0055]

[0056] Step 3: Train the model and obtain the predicted strain results. Train the corresponding model using the training set for each location, and then obtain the strain value at each location at the corresponding time using the test set. Simultaneously, the root mean square error (RMSE) is used to evaluate the training and testing results of the network. The RMSE is defined as:

[0057]

[0058] In equation (5), x i y represents the model's predicted value. i This represents the true value of the data, and n represents the number of samples.

[0059] Step 5: Obtain the bond-slip model. Organizing the strain values ​​obtained in Step 3 yields the strain values ​​at various locations on the FRP sheet at a given moment. Substituting these values ​​into the formula for FRP strain distribution with location, we obtain the parameters α, β, and d0, defined as follows:

[0060]

[0061] In equation (6), d represents the distance from the observation point to the loading end, and ε(d) represents the strain value at the observation point. Substituting the values ​​of α, β, and d0 into the definitions of slip as a function of position s(d), bond stress as a function of position τ(d), and bond stress as a function of slip τ(s), respectively, we can obtain the bond-slip model between the FRP sheet and the concrete interface, whose definitions are as follows:

[0062]

[0063]

[0064]

[0065] In equations (7)(8)(9), E f t represents the elastic modulus of FRP sheet. f Let represent the thickness of the FRP sheet, s represent the slippage of the FRP sheet, and τ represent the bond stress between the FRP sheet and the concrete. The corresponding relationship is shown in Figure 5(c), which includes the bond-slip curve most commonly used to analyze the bonding performance of the FRP-concrete interface. The calculations and result acquisition in step five are also automatically performed by code.

Claims

1. A method for generating a bond-slip model at the interface between FRP sheet and concrete based on WGAN, characterized in that, include: Step 1: Acquisition of strain data samples corresponding to the location changes, and preprocessing method of training set; Step 2, construct a prediction model based on WGAN; Step 3: Evaluate the training and testing results of the network; Step 4: Fit the predicted strain data at each location at the same time with the definition of the interface bond-slip model, and plot multiple curves corresponding to each definition in the bond-slip model, which can be applied to the design of safe and reliable reinforced components. Step 2: When building the WGAN model, an LSTM network was chosen as the generator and a CNN as the discriminator. The generator uses a sliding window method for prediction, predicting the strain data for the next 3 time series every 30 time series data. The strain data corresponding to the last 3 time series generated by the generator and the strain data corresponding to the first 30 time series of real time series are combined to form a fake dataset and the strain data corresponding to the real 33 time series, and then passed to the discriminator to identify the authenticity. The WGAN model uses gradient penalty to enforce Lipschitz constraints. The discriminator uses Leaky ReLU as its activation function. The discriminator loss function is defined as follows: In equation (3), Indicates the number of samples. This represents the input data of the generator. This represents the target value of the real dataset. This represents the generated data value of the generator. This indicates the discriminator's judgment calculation on the data. It is a penalty coefficient with a value of 10. Indicates a penalty item; The generator loss function is defined as follows:

2. The method for generating a bond-slip model of FRP sheet-concrete interface based on WGAN according to claim 1, characterized in that, Step 1: Use the finite element software LS-DYNA to model and collect strain data. The data corresponds to the position and strain every two seconds, i.e., the average strain. ; In addition, five parameters were selected, namely the compressive strength of concrete. Elastic modulus FRP board thickness Ultimate strength Yield strength ; Organize the data for the above six parameters and create the corresponding training and test sets for each position.

3. The method for generating a bond-slip model of FRP sheet-concrete interface based on WGAN according to claim 1, characterized in that, Step 3: Use the root mean square error (RMSE) to evaluate the training and testing results of the network.

4. The method for generating a bond-slip model of FRP sheet-concrete interface based on WGAN according to claim 1, characterized in that, Method for plotting curves in the bond-slip model: The unknown parameter information is obtained by fitting the predicted strain data at each position at the same time with the formula (6) for the distribution of FRP strain with position. The obtained parameter information is then substituted into the definitions (7)(8)(9) of the bond-slip relationship to obtain the curves in the bond-slip model.

Images