Method for predicting residual bearing capacity of RC pier column with cracks in salt freezing environment

Abstract

The invention discloses a method for predicting the residual bearing capacity of an RC pier column with cracks in a salt freezing environment. The method comprises the steps that 1, historical data of the bearing capacity of the RC pier column with cracks in the salt freezing environment are acquired; 2, constructing a bearing capacity neural network model of the RC pier stud with the crack; 3, training the bearing capacity neural network model of the RC pier stud with the crack; 4, obtaining a relational expression between the salt freezing cycle number and the plastic dissipation energy density; and 5, predicting the bearing capacity of the RC pier stud with the crack under the follow-up salt freezing cycle times. The method is simple in step, a relational expression between the salt freezing cycle times and the plastic dissipated energy density is obtained based on a neural network model and a finite element model of the RC pier stud with the crack, then the salt freezing cycle load is equivalent to the monotonous cooling load through the plastic dissipated energy density under the subsequent salt freezing cycle times, and salt freezing cycle simulation is achieved. And obtaining a bearing capacity predicted value and a damage position.

Description

Technical Field

[0001] The invention belongs to the technical field of RC pier column durability assessment, and in particular relates to a method for predicting the residual bearing capacity of a cracked RC pier column in a salt-freezing environment. Background Art

[0002] During the service life of concrete bridges, the performance of various load-bearing components continues to degrade under the influence and erosion of external environmental factors. RC piers (i.e. reinforced concrete columns) are key components that bear loads, and the degradation of their bearing performance will directly affect the safe operation of the entire bridge structure. In a salt-freezing environment, RC piers are affected by freeze-thaw cycles and corrosive ions at the same time, and cracks caused by various factors will accelerate the degradation of the residual bearing capacity of RC piers. Therefore, it is of great significance to predict the residual bearing capacity of cracked RC piers in a salt-freezing environment.

[0003] In the durability assessment study of RC piers, three methods are usually used: experimental testing, numerical simulation, and intelligent algorithm prediction to obtain the residual bearing capacity of RC piers. In the experimental test, the quick freezing method is usually used to place the RC pier in a salt solution for a salt-freezing cycle test, and then the residual bearing capacity of the RC pier is obtained through a static loading test. However, this method requires a lot of manpower and material resources, and the test cycle is long; in the numerical simulation, the impact of the RC pier under the salt-freezing cycle is usually simulated based on the finite element model of the RC pier, and the freeze-thaw cycle process is also simulated based on the number of salt-freezing cycles, which has low calculation efficiency and high calculation cost; in the intelligent algorithm prediction, neural networks are usually used to predict the residual bearing capacity of RC piers in a salt-freezing environment, which is also predicted based on the number of salt-freezing cycles, and the damage location of the RC pier cannot be obtained.

[0004] Therefore, there is currently a lack of an accurate and efficient method for predicting the residual bearing capacity of cracked RC piers under salt-freezing environments. Based on the neural network model and the finite element model of cracked RC piers, the relationship between the number of salt-freezing cycles and the plastic dissipated energy density is obtained. Then, under the subsequent number of salt-freezing cycles, the plastic dissipated energy density is used to equate the salt-freezing cycle load to a monotonic cooling load to realize salt-freezing cycle simulation, and the bearing capacity prediction value and damage location are obtained, thereby solving the problems of current experimental research, numerical simulation, and intelligent algorithm methods. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for predicting the residual bearing capacity of a RC pier with cracks under a salt-freezing environment in view of the deficiencies in the above-mentioned prior art. The method has simple steps and reasonable design. Based on a neural network model and a finite element model of a RC pier with cracks, a relationship between the number of salt-freezing cycles and the plastic dissipated energy density is obtained. Then, under the subsequent number of salt-freezing cycles, the plastic dissipated energy density is used to equate the salt-freezing cycle load to a monotonic cooling load to realize salt-freezing cycle simulation, and the bearing capacity prediction value and damage location are obtained, thereby solving the problems of current experimental research, numerical simulation, and intelligent algorithm methods.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for predicting the residual bearing capacity of a cracked RC pier column under a salt-freezing environment, characterized in that the method comprises the following steps:

[0007] Step 1: Obtain historical data on the bearing capacity of cracked RC piers under salt-freezing environment:

[0008] Step 101, collecting bearing capacity test data corresponding to different salt-freezing cycles of cracked RC piers in a salt-freezing environment;

[0009] Step 102, using cubic Hermite interpolation to interpolate the collected bearing capacity test data to obtain the interpolated number of salt-freezing cycles and corresponding bearing capacity data;

[0010] Step 103, obtaining the influencing indicators of each cracked RC pier: cross-sectional size, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction and Na2CO3 volume fraction;

[0011] Step 104, record the data of different cross-sectional dimensions, heights, crack widths, water-cement ratios, fly ash proportions, initial bearing capacities, reinforcement ratios, NaCl volume fractions, Na2SO4 volume fractions, Na2CO3 volume fractions, salt-freezing cycles, and bearing capacities as bearing capacity data, and use the bearing capacity data as a training set;

[0012] Step 2: Construct a neural network model of the bearing capacity of cracked RC piers;

[0013] Constructing a neural network model of the bearing capacity of a cracked RC pier column; wherein the neural network model of the bearing capacity of a cracked RC pier column comprises a CNN network module, an SE module, a Flatten layer and a BiLSTM network module, a third fully connected layer and an output layer;

[0014] The CNN network module includes a first convolutional layer, a first ReLU activation function layer, a second convolutional layer, and a second ReLU activation function layer;

[0015] The SE module includes a global average pooling layer, a first fully connected layer, a third ReLU activation function layer, a second fully connected layer, and a Sigmoid activation function layer;

[0016] Step 3: Training of the neural network model for the bearing capacity of cracked RC pier columns:

[0017] Step 301, using a computer to record each group of cross-sectional dimensions, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction, number of salt-freezing cycles and bearing capacity of the training set as a training sample;

[0018] Step 302: using a computer to input the cross-sectional size, height, crack width, water-cement ratio, fly ash ratio, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles in any training sample, and output the corresponding bearing capacity, and input it into the neural network model of the bearing capacity of the cracked RC pier column for training, so as to obtain a trained neural network model of the bearing capacity of the cracked RC pier column;

[0019] Step 4: Obtain the relationship between the number of salt-freezing cycles and the plastic dissipated energy density:

[0020] The finite element model of the RC pier with cracks was established by computer using ABAQUS finite element software. Based on the simulation value of the RC pier with cracks to be studied and the reference value of the trained neural network model of the bearing capacity of the RC pier with cracks, the relationship between the number of salt-freezing cycles and the plastic dissipated energy density was obtained:

[0021] Step 5: Prediction of the bearing capacity of the cracked RC pier under subsequent salt freezing cycles:

[0022] Step 501, input the subsequent salt-freezing cycle number corresponding to the load of the cracked RC pier column to be studied into the relationship between the salt-freezing cycle number and the plastic dissipated energy density, and obtain the required plastic dissipated energy density;

[0023] Step 502, using a computer and using ABAQUS finite element software to perform cooling simulation again on the RC pier column with cracks to be studied, until the plastic dissipation energy density meets the required plastic dissipation energy density;

[0024] Then, ABAQUS finite element software is used to carry out uniaxial compression simulation under the required plastic dissipation energy density to obtain the bearing capacity simulation value and damage location, and the bearing capacity simulation value is recorded as the predicted value of the residual bearing capacity under the subsequent salt-freezing cycle number corresponding to the load of the cracked RC pier to be studied.

[0025] The above-mentioned method for predicting the residual bearing capacity of a cracked RC pier column under a salt-freezing environment is characterized in that: step 302, the specific process is as follows:

[0026] Step 3021, taking the cross-sectional size, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles of any training sample as input features;

[0027] Step 3022: Input the features into the first convolutional layer to obtain the features output by the first convolutional layer;

[0028] Step 3023: The features output by the first convolutional layer are sequentially passed through the first ReLU activation function layer, the second convolutional layer, and the second ReLU activation function layer to obtain the first intermediate output features;

[0029] Step 3024: The features output by the first convolutional layer are sequentially passed through a global average pooling layer, a first fully connected layer, a third ReLU activation function layer, a second fully connected layer, and a Sigmoid activation function layer to obtain a second intermediate output feature.

[0030] Step 3025: Use the second intermediate output feature as a weight, multiply the first intermediate output feature by the second intermediate output feature, and obtain a weighted feature;

[0031] Step 3026: Input the weighted features into the Flatten layer for processing to obtain a one-dimensional feature vector;

[0032] Step 3027: Input the one-dimensional feature vector into the BiLSTM network module for processing to obtain an output feature vector, and output the output feature vector through the third fully connected layer to obtain an output prediction value;

[0033] Step 3028: Obtain a mean square error loss based on the output prediction value and the true value of the bearing capacity corresponding to the training sample;

[0034] Step 3029, the computer uses the Adam optimization algorithm to iteratively optimize the neural network model of the bearing capacity of the cracked RC pier column using the mean square error loss until all the training sets are trained and one iteration training is completed;

[0035] Step 302A, repeating step 3029 for iterative training until a preset number of iterative training times is met, thereby obtaining a trained neural network model of the bearing capacity of cracked RC pier columns.

[0036] The above-mentioned method for predicting the residual bearing capacity of a cracked RC pier column under a salt-freezing environment is characterized in that: Step 2 is as follows:

[0037] The number of convolution kernels in the first convolution layer is 32, the size of the convolution kernel is 3×1, and the sliding step is 1;

[0038] The number of neurons in the first fully connected layer is 16, the number of neurons in the second fully connected layer is 64; the number of neurons in the third fully connected layer and the output layer is 1;

[0039] The number of convolution kernels in the second convolution layer is 64, the size of the convolution kernel is 3×1, and the sliding step is 1;

[0040] The number of neurons in the hidden layer of the BiLSTM network module is 10.

[0041] The above-mentioned method for predicting the residual bearing capacity of a cracked RC pier column under a salt-freezing environment is characterized in that: Step 4, the specific process is as follows:

[0042] Step 401: according to the actual structure of the RC pier column with cracks to be studied, a simulation model of the RC pier column with cracks is established by using ABAQUS finite element software on a computer; wherein the simulation model of the RC pier column with cracks is the same as the actual structure of the RC pier column with cracks to be studied;

[0043] Step 402: Using a computer and ABAQUS finite element software to perform a cooling simulation on the established RC pier column simulation model with cracks, obtain the plastic dissipation energy at a set time interval; and divide the plastic dissipation energy at the set time interval by the volume of the RC pier column simulation model with cracks, to obtain the plastic dissipation energy density Dx at the set time interval;

[0044] Step 403: Using a computer and using ABAQUS finite element software, perform uniaxial compression simulation at various plastic dissipated energy densities Dx to obtain various bearing capacity simulation values ​​Ps;

[0045] Step 404: using a computer, input the cross-sectional dimensions, height, crack width, water-cement ratio, fly ash ratio, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles corresponding to the actual structure of the RC pier with cracks to be studied into the trained neural network model of the bearing capacity of the RC pier with cracks, and obtain a bearing capacity reference value Pref corresponding to the actual structure of the RC pier with cracks to be studied;

[0046] Step 405: using a computer, when the difference between the bearing capacity reference value Pref and the bearing capacity simulation value Ps is the smallest, obtain the number of salt-freezing cycles corresponding to the bearing capacity reference value Pref and the plastic dissipated energy density corresponding to the bearing capacity simulation value Ps;

[0047] Step 406, repeating steps 404 and 405 multiple times to obtain the number of salt-freezing cycles and the corresponding plastic dissipated energy density;

[0048] Step 407: Use a computer to perform quadratic polynomial fitting on each salt-freezing cycle number and each corresponding plastic dissipated energy density to obtain a relationship between the salt-freezing cycle number and the plastic dissipated energy density.

[0049] Compared with the prior art, the present invention has the following advantages:

[0050] 1. The method of the present invention has simple steps and reasonable design, and solves the problems of current experimental research, numerical simulation, and intelligent algorithm methods.

[0051] 2. The present invention uses the bearing capacity collection data of RC pier columns with cracks under salt-freezing environment to train the neural network model of the bearing capacity of RC pier columns with cracks, and obtains a trained neural network model of the bearing capacity of RC pier columns with cracks, which is convenient for obtaining the bearing capacity reference value according to the number of salt-freezing cycles in the future.

[0052] 3. The present invention establishes a finite element model simulation of a cracked RC pier column, obtains the number of salt-freezing cycles and the corresponding plastic dissipated energy density based on the simulation value of the cracked RC pier column to be studied and the reference value of the trained neural network model of the bearing capacity of the cracked RC pier column, and then obtains the relationship between the number of salt-freezing cycles and the plastic dissipated energy density.

[0053] 4. The number of subsequent salt-freezing cycles corresponding to the load of the cracked RC pier to be studied in the present invention is input into the relationship between the number of salt-freezing cycles and the plastic dissipated energy density to obtain the required plastic dissipated energy density, and then the salt-freezing cycle load is equivalent to the monotonic cooling load using the plastic dissipated energy density to realize the salt-freezing cycle simulation, and the bearing capacity simulation value and the damage location are obtained. The bearing capacity simulation value is the predicted value of the remaining bearing capacity under the subsequent number of salt-freezing cycles of the cracked RC pier to be studied, which improves the efficiency and accuracy of the bearing capacity prediction value and can obtain the damage location.

[0054] In summary, the method of the present invention has simple steps and reasonable design. Based on the neural network model and the finite element model of the RC pier with cracks, the relationship between the number of salt-freezing cycles and the plastic dissipated energy density is obtained. Then, under the subsequent number of salt-freezing cycles, the plastic dissipated energy density is used to equate the salt-freezing cycle load to a monotonic cooling load to realize the salt-freezing cycle simulation, and the bearing capacity prediction value and damage location are obtained, thereby solving the problems of the current experimental research, numerical simulation, and intelligent algorithm methods.

[0055] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The figure is a flowchart of the method of the present invention.

[0057] Figure 2It is a structural schematic diagram of the neural network model of the bearing capacity of cracked RC pier columns according to the present invention. DETAILED DESCRIPTION

[0058] like Figure 1 As shown, the method for predicting the residual bearing capacity of a cracked RC pier column under a salt-freezing environment of the present invention comprises the following steps:

[0059] Step 1: Obtain historical data on the bearing capacity of cracked RC piers under salt-freezing environment:

[0060] Step 101, collecting bearing capacity test data corresponding to different salt-freezing cycles of cracked RC piers in a salt-freezing environment;

[0061] Step 102, using cubic Hermite interpolation to interpolate the collected bearing capacity test data to obtain the interpolated number of salt-freezing cycles and corresponding bearing capacity data;

[0062] Step 103, obtaining the influencing indicators of each cracked RC pier: cross-sectional size, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction and Na2CO3 volume fraction;

[0063] Step 104, record the data of different cross-sectional dimensions, heights, crack widths, water-cement ratios, fly ash proportions, initial bearing capacities, reinforcement ratios, NaCl volume fractions, Na2SO4 volume fractions, Na2CO3 volume fractions, salt-freezing cycles, and bearing capacities as bearing capacity data, and use the bearing capacity data as a training set;

[0064] Step 2: Construct a neural network model of the bearing capacity of cracked RC piers;

[0065] Constructing a neural network model of the bearing capacity of a cracked RC pier column; wherein the neural network model of the bearing capacity of a cracked RC pier column comprises a CNN network module, an SE module, a Flatten layer and a BiLSTM network module, a third fully connected layer and an output layer;

[0066] The CNN network module includes a first convolutional layer, a first ReLU activation function layer, a second convolutional layer, and a second ReLU activation function layer;

[0067] The SE module includes a global average pooling layer, a first fully connected layer, a third ReLU activation function layer, a second fully connected layer, and a Sigmoid activation function layer;

[0068] Step 3: Training of the neural network model for the bearing capacity of cracked RC pier columns:

[0069] Step 301, using a computer to record each group of cross-sectional dimensions, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction, number of salt-freezing cycles and bearing capacity of the training set as a training sample;

[0070] Step 302: using a computer to input the cross-sectional size, height, crack width, water-cement ratio, fly ash ratio, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles in any training sample, and output the corresponding bearing capacity, and input it into the neural network model of the bearing capacity of the cracked RC pier column for training, so as to obtain a trained neural network model of the bearing capacity of the cracked RC pier column;

[0071] Step 4: Obtain the relationship between the number of salt-freezing cycles and the plastic dissipated energy density:

[0072] The finite element model of the RC pier with cracks was established by computer using ABAQUS finite element software. Based on the simulation value of the RC pier with cracks to be studied and the reference value of the trained neural network model of the bearing capacity of the RC pier with cracks, the relationship between the number of salt-freezing cycles and the plastic dissipated energy density was obtained:

[0073] Step 5: Prediction of the bearing capacity of the cracked RC pier under subsequent salt freezing cycles:

[0074] Step 501, input the subsequent salt-freezing cycle number corresponding to the load of the cracked RC pier column to be studied into the relationship between the salt-freezing cycle number and the plastic dissipated energy density, and obtain the required plastic dissipated energy density;

[0075] Step 502, using a computer and using ABAQUS finite element software to perform cooling simulation again on the RC pier column with cracks to be studied, until the plastic dissipation energy density meets the required plastic dissipation energy density;

[0076] Then, ABAQUS finite element software is used to carry out uniaxial compression simulation under the required plastic dissipation energy density to obtain the bearing capacity simulation value and damage location, and the bearing capacity simulation value is recorded as the predicted value of the residual bearing capacity under the subsequent salt-freezing cycle number corresponding to the load of the cracked RC pier to be studied.

[0077] In this embodiment, step 302, the specific process is as follows:

[0078] Step 3021, taking the cross-sectional size, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles of any training sample as input features;

[0079] Step 3022: Input the features into the first convolutional layer to obtain the features output by the first convolutional layer;

[0080] Step 3023: The features output by the first convolutional layer are sequentially passed through the first ReLU activation function layer, the second convolutional layer, and the second ReLU activation function layer to obtain the first intermediate output features;

[0081] Step 3024: The features output by the first convolutional layer are sequentially passed through a global average pooling layer, a first fully connected layer, a third ReLU activation function layer, a second fully connected layer, and a Sigmoid activation function layer to obtain a second intermediate output feature.

[0082] Step 3025: Use the second intermediate output feature as a weight, multiply the first intermediate output feature by the second intermediate output feature, and obtain a weighted feature;

[0083] Step 3026: Input the weighted features into the Flatten layer for processing to obtain a one-dimensional feature vector;

[0084] Step 3027: Input the one-dimensional feature vector into the BiLSTM network module for processing to obtain an output feature vector, and output the output feature vector through the third fully connected layer to obtain an output prediction value;

[0085] Step 3028: Obtain a mean square error loss based on the output prediction value and the true value of the bearing capacity corresponding to the training sample;

[0086] Step 3029, the computer uses the Adam optimization algorithm to iteratively optimize the neural network model of the bearing capacity of the cracked RC pier column using the mean square error loss until all the training sets are trained and one iteration training is completed;

[0087] Step 302A, repeating step 3029 for iterative training until a preset number of iterative training times is met, thereby obtaining a trained neural network model of the bearing capacity of cracked RC pier columns.

[0088] In this embodiment, step 2 is specifically as follows:

[0089] The number of convolution kernels in the first convolution layer is 32, the size of the convolution kernel is 3×1, and the sliding step is 1;

[0090] The number of neurons in the first fully connected layer is 16, the number of neurons in the second fully connected layer is 64; the number of neurons in the third fully connected layer and the output layer is 1;

[0091] The number of convolution kernels in the second convolution layer is 64, the size of the convolution kernel is 3×1, and the sliding step is 1;

[0092] The number of neurons in the hidden layer of the BiLSTM network module is 10.

[0093] In this embodiment, step 4, the specific process is as follows:

[0094] Step 401: according to the actual structure of the RC pier column with cracks to be studied, a simulation model of the RC pier column with cracks is established by using ABAQUS finite element software on a computer; wherein the simulation model of the RC pier column with cracks is the same as the actual structure of the RC pier column with cracks to be studied;

[0095] Step 402: Using a computer and ABAQUS finite element software to perform a cooling simulation on the established RC pier column simulation model with cracks, obtain the plastic dissipation energy at a set time interval; and divide the plastic dissipation energy at the set time interval by the volume of the RC pier column simulation model with cracks, to obtain the plastic dissipation energy density Dx at the set time interval;

[0096] Step 403: Using a computer and using ABAQUS finite element software, perform uniaxial compression simulation at various plastic dissipated energy densities Dx to obtain various bearing capacity simulation values ​​Ps;

[0097] Step 404: using a computer, input the cross-sectional dimensions, height, crack width, water-cement ratio, fly ash ratio, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles corresponding to the actual structure of the RC pier with cracks to be studied into the trained neural network model of the bearing capacity of the RC pier with cracks, and obtain a bearing capacity reference value Pref corresponding to the actual structure of the RC pier with cracks to be studied;

[0098] Step 405: using a computer, when the difference between the bearing capacity reference value Pref and the bearing capacity simulation value Ps is the smallest, obtain the number of salt-freezing cycles corresponding to the bearing capacity reference value Pref and the plastic dissipated energy density corresponding to the bearing capacity simulation value Ps;

[0099] Step 406, repeating steps 404 and 405 multiple times to obtain the number of salt-freezing cycles and the corresponding plastic dissipated energy density;

[0100] Step 407: Use a computer to perform quadratic polynomial fitting on each salt-freezing cycle number and each corresponding plastic dissipated energy density to obtain a relationship between the salt-freezing cycle number and the plastic dissipated energy density.

[0101] In this embodiment, in step 402, the time interval is set to 300 in the ABAQUS finite element software.

[0102] In this embodiment, the SE module is a Squeeze-and-Excitation Networks module.

[0103] In this embodiment, the BiLSTM network module is a Bidirectional Long Short-Term Memory network module.

[0104] In this embodiment, combined with Figure 2 The first convolutional layer is Conv 1, the first ReLU activation function layer is ReLU 1, the second convolutional layer is Conv 2, the second ReLU activation function layer is ReLU 2, the third ReLU activation function layer is ReLU 3, the first fully connected layer is FC 1, the second fully connected layer is FC 2, and the third fully connected layer is FC 3.

[0105] In this embodiment, the size of the feature is expressed by the number of channels × length × width, the size of the input feature is 11 × 1 × 1, the feature size output by the first convolutional layer is 9 × 1 × 32, the feature output by the first convolutional layer is 9 × 1 × 32 after the first ReLU activation function layer outputs the feature, the feature output by the first ReLU activation function layer is 7 × 1 × 64 after the second convolutional layer outputs the feature, the feature output by the second convolutional layer is 7 × 1 × 64 after the second ReLU activation function layer outputs the feature, that is, the first intermediate output feature;

[0106] The features output by the first convolutional layer are 1×1×32 after the global average pooling layer outputs, the features output by the global average pooling layer are 1×1×16 after the first fully connected layer outputs, the features output by the first fully connected layer are 1×1×16 after the third ReLU activation function layer outputs; the features output by the third ReLU activation function layer are 1×1×64 after the second fully connected layer outputs, i.e., the second intermediate output features;

[0107] The second intermediate output feature is taken as the weight, and the first intermediate output feature is multiplied by the weight output by the second fully connected layer to obtain a weighted feature. The size of the weighted feature is 7×1×64; the one-dimensional feature vector obtained after processing by the Flatten layer is 448×1.

[0108] In this embodiment, during the specific implementation, the actual structure of the RC pier with cracks to be studied is as follows: the cross-sectional size is 100mm×100mm, the height is 400mm, the crack width is 0.1mm, the water-cement ratio is 0.39, the fly ash ratio is 21.95%, the initial bearing capacity is 435.45kN, and the reinforcement ratio is 3.14%.

[0109] In this embodiment, during the specific implementation, in step 404, the number of salt-freezing cycles is 1-75, the volume fraction of NaCl is 3.50%, the volume fraction of Na2SO4 is 0, and the volume fraction of Na2CO3 is 0, and the bearing capacity reference value is obtained by inputting the trained neural network model of the bearing capacity of the RC pier with cracks. In this way, the difference between the bearing capacity reference value Pref and the bearing capacity simulation value Ps is minimized, so that the bearing capacity simulation value and the bearing capacity simulation value Ps are close to the same, thereby realizing the salt-freezing environment simulation in the finite element simulation.

[0110] In this embodiment, during the specific implementation, a cooling simulation is performed in step 402, specifically as follows:

[0111] Using ABAQUS finite element software, add an initial temperature of 0°C to the predefined field, drop it to -180°C, and set a smoothing analysis step, time 0-1 second, amplitude 0-1;

[0112] In step 403, uniaxial compression simulation is performed, specifically as follows: the bottom end of the simulation model of the RC pier column with cracks is fixed, and displacement loading is performed along the Z direction at the top end of the simulation model of the RC pier column with cracks; wherein the Z direction is the height direction.

[0113] In this embodiment, during the specific implementation, the relationship between the number of salt-freezing cycles n and the plastic dissipated energy density Dx is obtained through fitting in step 406 as follows: x =0.0011n 2 +0.4926n+42.5338.

[0114] In this embodiment, during specific implementation, the temperature reduction simulation and uniaxial compression simulation in step 502 may refer to steps 402 and 403 .

[0115] In this embodiment, during specific implementation, the preset number of iterative training in step 302A is 2000-10000.

[0116] In summary, the method of the present invention has simple steps and reasonable design. Based on the neural network model and the finite element model of the RC pier with cracks, the relationship between the number of salt-freezing cycles and the plastic dissipated energy density is obtained. Then, under the subsequent number of salt-freezing cycles, the plastic dissipated energy density is used to equate the salt-freezing cycle load to a monotonic cooling load to realize the salt-freezing cycle simulation, and the bearing capacity prediction value and damage location are obtained, thereby solving the problems of the current experimental research, numerical simulation, and intelligent algorithm methods.

[0117] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A method for predicting the residual bearing capacity of RC piers with cracks in a salt-freezing environment, characterized in that: The method comprises the following steps: Step 1: Obtain historical data on the bearing capacity of cracked RC piers under salt-freezing environment: Step 101, collecting bearing capacity test data corresponding to different salt-freezing cycles of cracked RC piers in a salt-freezing environment; Step 102, using cubic Hermite interpolation to interpolate the collected bearing capacity test data to obtain the interpolated number of salt-freezing cycles and corresponding bearing capacity data; Step 103, obtaining the influencing indicators of each cracked RC pier: cross-sectional size, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction and Na2CO3 volume fraction; Step 104, record the data of different cross-sectional dimensions, heights, crack widths, water-cement ratios, fly ash proportions, initial bearing capacities, reinforcement ratios, NaCl volume fractions, Na2SO4 volume fractions, Na2CO3 volume fractions, salt-freezing cycles, and bearing capacities as bearing capacity data, and use the bearing capacity data as a training set; Step 2: Construct a neural network model of the bearing capacity of cracked RC piers; Constructing a neural network model of the bearing capacity of a cracked RC pier column; wherein the neural network model of the bearing capacity of a cracked RC pier column comprises a CNN network module, an SE module, a Flatten layer and a BiLSTM network module, a third fully connected layer and an output layer; The CNN network module includes a first convolutional layer, a first ReLU activation function layer, a second convolutional layer, and a second ReLU activation function layer; The SE module includes a global average pooling layer, a first fully connected layer, a third ReLU activation function layer, a second fully connected layer, and a Sigmoid activation function layer; Step 3: Training of the neural network model for the bearing capacity of cracked RC pier columns: Step 301, using a computer to record each group of cross-sectional dimensions, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction, number of salt-freezing cycles and bearing capacity of the training set as a training sample; Step 302: using a computer to input the cross-sectional size, height, crack width, water-cement ratio, fly ash ratio, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles in any training sample, and output the corresponding bearing capacity, and input it into the neural network model of the bearing capacity of the cracked RC pier column for training, so as to obtain a trained neural network model of the bearing capacity of the cracked RC pier column; Step 4: Obtain the relationship between the number of salt-freezing cycles and the plastic dissipated energy density: The finite element model of the RC pier with cracks was established by computer using ABAQUS finite element software. Based on the simulation value of the RC pier with cracks to be studied and the reference value of the trained neural network model of the bearing capacity of the RC pier with cracks, the relationship between the number of salt-freezing cycles and the plastic dissipated energy density was obtained: Step 5: Prediction of the bearing capacity of the cracked RC pier under subsequent salt freezing cycles: Step 501, input the subsequent salt-freezing cycle number corresponding to the load of the cracked RC pier column to be studied into the relationship between the salt-freezing cycle number and the plastic dissipated energy density, and obtain the required plastic dissipated energy density; Step 502, using a computer and using ABAQUS finite element software to perform cooling simulation again on the RC pier column with cracks to be studied, until the plastic dissipation energy density meets the required plastic dissipation energy density; Then, ABAQUS finite element software is used to carry out uniaxial compression simulation under the required plastic dissipation energy density to obtain the bearing capacity simulation value and damage location, and the bearing capacity simulation value is recorded as the predicted value of the residual bearing capacity under the subsequent salt-freezing cycle number corresponding to the load of the cracked RC pier to be studied.

2. A method for predicting the residual bearing capacity of a cracked RC pier column in a salt-freezing environment according to claim 1, characterized in that: Step 302, the specific process is as follows: Step 3021, taking the cross-sectional size, height, crack width, water-cement ratio, fly ash proportion, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles of any training sample as input features; Step 3022: Input the features into the first convolutional layer to obtain the features output by the first convolutional layer; Step 3023: The features output by the first convolutional layer are sequentially passed through the first ReLU activation function layer, the second convolutional layer, and the second ReLU activation function layer to obtain the first intermediate output features; Step 3024: The features output by the first convolutional layer are sequentially passed through a global average pooling layer, a first fully connected layer, a third ReLU activation function layer, a second fully connected layer, and a Sigmoid activation function layer to obtain a second intermediate output feature. Step 3025: Use the second intermediate output feature as a weight, multiply the first intermediate output feature by the second intermediate output feature, and obtain a weighted feature; Step 3026: Input the weighted features into the Flatten layer for processing to obtain a one-dimensional feature vector; Step 3027: Input the one-dimensional feature vector into the BiLSTM network module for processing to obtain an output feature vector, and output the output feature vector through the third fully connected layer to obtain an output prediction value; Step 3028: Obtain a mean square error loss based on the output prediction value and the true value of the bearing capacity corresponding to the training sample; Step 3029, the computer uses the Adam optimization algorithm to iteratively optimize the neural network model of the bearing capacity of the cracked RC pier column using the mean square error loss until all the training sets are trained and one iteration training is completed; Step 302A, repeating step 3029 for iterative training until a preset number of iterative training times is met, thereby obtaining a trained neural network model of the bearing capacity of cracked RC pier columns.

3. A method for predicting the residual bearing capacity of a cracked RC pier column in a salt-freezing environment according to claim 1, characterized in that: Step 2 is as follows: The number of convolution kernels in the first convolution layer is 32, the size of the convolution kernel is 3×1, and the sliding step is 1; The number of neurons in the first fully connected layer is 16, the number of neurons in the second fully connected layer is 64; the number of neurons in the third fully connected layer and the output layer is 1; The number of convolution kernels in the second convolution layer is 64, the size of the convolution kernel is 3×1, and the sliding step is 1; The number of neurons in the hidden layer of the BiLSTM network module is 10.

4. A method for predicting the residual bearing capacity of a cracked RC pier column in a salt-freezing environment according to claim 1, characterized in that: Step 4: The specific process is as follows: Step 401: according to the actual structure of the RC pier column with cracks to be studied, a simulation model of the RC pier column with cracks is established by using ABAQUS finite element software on a computer; wherein the simulation model of the RC pier column with cracks is the same as the actual structure of the RC pier column with cracks to be studied; Step 402: Using a computer and ABAQUS finite element software to perform a cooling simulation on the established RC pier column simulation model with cracks, obtain the plastic dissipation energy at a set time interval; and divide the plastic dissipation energy at the set time interval by the volume of the RC pier column simulation model with cracks, to obtain the plastic dissipation energy density Dx at the set time interval; Step 403: Using a computer and using ABAQUS finite element software, perform uniaxial compression simulation at various plastic dissipated energy densities Dx to obtain various bearing capacity simulation values ​​Ps; Step 404: using a computer, input the cross-sectional dimensions, height, crack width, water-cement ratio, fly ash ratio, initial bearing capacity, reinforcement ratio, NaCl volume fraction, Na2SO4 volume fraction, Na2CO3 volume fraction and number of salt-freezing cycles corresponding to the actual structure of the RC pier with cracks to be studied into the trained neural network model of the bearing capacity of the RC pier with cracks, and obtain a bearing capacity reference value Pref corresponding to the actual structure of the RC pier with cracks to be studied; Step 405: using a computer, when the difference between the bearing capacity reference value Pref and the bearing capacity simulation value Ps is the smallest, obtain the number of salt-freezing cycles corresponding to the bearing capacity reference value Pref and the plastic dissipated energy density corresponding to the bearing capacity simulation value Ps; Step 406, repeating steps 404 and 405 multiple times to obtain the number of salt-freezing cycles and the corresponding plastic dissipated energy density; Step 407: Use a computer to perform quadratic polynomial fitting on each salt-freezing cycle number and each corresponding plastic dissipated energy density to obtain a relationship between the salt-freezing cycle number and the plastic dissipated energy density.

Images