Tubular pile vertical bearing capacity prediction method based on deep learning

Abstract

The invention belongs to the field of building and foundation engineering, and discloses a tubular pile vertical bearing capacity prediction method based on deep learning. The detection precision of the vertical bearing capacity of the prestressed concrete pipe pile is improved. A convolutional neural network (CNN) and a long short-term memory network (LSTM) are introduced, sensor monitoring data and field test results are combined, a pile foundation bearing capacity prediction model is optimized, and the pile foundation bearing capacity is evaluated more accurately.

Description

Technical Field

[0001] The present invention belongs to the field of construction and foundation engineering, and specifically relates to a method for predicting the vertical bearing capacity of pipe piles based on deep learning. Background Art

[0002] Prestressed concrete piles (PHC piles) are commonly used in construction and foundation engineering. They are widely used in engineering construction fields such as building foundations and bridge foundations due to their strong bearing capacity, convenient construction and strong durability. As an important foundation support structure, the accurate detection of its vertical bearing capacity is crucial to the quality and safety of the project. At present, the test methods for the vertical bearing capacity of piles include static load test method and self-balancing method. The static load test method can directly and accurately obtain bearing capacity data, but its conventional loading method requires thousands of tons of piles, and the stacking and transportation of piles, etc., resulting in high costs, complex construction, and strict site requirements. The self-balancing method predicts the vertical bearing capacity based on the strain and deformation of the pile body. It is convenient to construct, but it has high requirements for measurement equipment, complex data processing, and it is difficult to ensure the accuracy of inversion analysis, which is easily affected. Non-experimental methods include empirical formula method and numerical simulation method. Among them, the empirical formula method is simple to calculate and low in cost, but this method relies on empirical formulas, and the accuracy is limited by the applicability of the formula used and the characteristics of the soil. The numerical simulation method builds a pile foundation model based on the finite element method and uses the transfer function of the pile side and pile end to determine the bearing capacity of the pipe pile. It can take into account complex soil and pile conditions and has strong flexibility. However, it requires high computing power and has high accuracy requirements for soil characteristics.

[0003] With the rise and development of artificial intelligence, the advantages of deep learning in data processing and pattern recognition have been applied in many industries. For example, the transportation industry trains traffic data such as vehicle flow to optimize traffic signals and reduce congestion. How to use big data technology to calibrate the vertical bearing capacity of pipe piles has become one of the key issues of concern in the engineering construction industry. Based on the self-balancing pile foundation monitoring technology, by using deep learning methods, potential bearing capacity information can be extracted from a large amount of data, thereby improving the accuracy and efficiency of pile foundation bearing capacity detection. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for predicting the vertical bearing capacity of pipe piles based on deep learning to solve the problems in the prior art. The technical solution adopted by the present invention is:

[0005] A method for predicting the vertical bearing capacity of pipe piles based on deep learning comprises the following steps:

[0006] Step 1: Obtain characteristic parameters of pipe piles and soil, and construct a basic data set S1 including soil layer parameters and pipe pile parameters;

[0007] Step 2: Preset the mechanical parameters in the basic data set S1 and preset the standard mechanical parameters corresponding to each type of soil layer;

[0008] Step 3: Calculate and obtain the estimated value of the theoretical vertical ultimate bearing capacity of the pile, and the obtained data is s1;

[0009] Step 4: Based on the soil layer parameters and pipe pile parameters in the basic geological database data set S1, a pipe pile numerical model is constructed using finite element software to calculate the ultimate vertical bearing capacity of the pipe pile, and the data obtained is s2;

[0010] Step 5: Use convolutional neural networks and long short-term memory networks to build a pile foundation bearing capacity prediction model, and combine the data set S2 to carry out prediction model training, optimize model parameters, and use the pile foundation bearing capacity prediction model in actual pipe pile testing to predict the ultimate bearing capacity of the pipe pile.

[0011] Furthermore, in step 1, the characteristic parameters of the pipe pile include geometric dimensions of the pile body, deadweight of the pile body, concrete grade, and concrete strength value.

[0012] Furthermore, in step 3, the following formula is used to calculate and obtain the estimated value of the theoretical vertical ultimate bearing capacity of the pipe pile:

[0013]

[0014] Where S is the ultimate vertical bearing capacity of a single pile, d 1i is the outer diameter of the pipe pile in the i-th soil layer; L is the length of the pipe pile, A p is the pile tip area of ​​the pipe pile, q pi is the ultimate lateral friction of the i-th soil layer on the pile body; q pi It is the pile tip resistance of the jth small area of ​​soil layer within the pile bottom range.

[0015] Furthermore, step five includes:

[0016] Step 100: Data collection: by installing sensors on site, real-time data of the pipe piles under different loads are collected to obtain the actual bearing capacity data of the pipe piles as a training set;

[0017] Step 200: Data preprocessing, preprocessing the collected sensor data and field test results by cleaning and standardizing them;

[0018] Step 300: deep learning model construction, using convolutional neural network to extract spatial features from sensor data, using long short-term memory network to model time series data, and constructing a pile foundation bearing capacity prediction model that comprehensively considers spatial and temporal features;

[0019] Step 400: Model training and optimization, using the preprocessed data to train the deep learning model;

[0020] Step 500: Bearing capacity assessment, applying the trained deep learning model to actual pipe pile detection, predicting the vertical bearing capacity of the pile foundation in real time based on sensor data, and comparing it with the field test results to adjust the parameters of the pile foundation bearing capacity prediction model;

[0021] Step 600: Determine matching learning model parameters, and predict the ultimate bearing capacity of the pipe pile according to inputting different soil layers and pipe pile parameters.

[0022] Furthermore, in step 300, the pile foundation bearing capacity prediction model includes:

[0023] The convolution layer 301 extracts features of the input soil layer parameters and pipe pile parameters in sequence and outputs multiple feature parameters;

[0024] Determine layer 302, based on the input soil layer mechanical parameters, determine whether each location-related feature included in the data is a feature of the location where the object of interest is located;

[0025] The synthesis layer 303 performs weighted synthesis processing on the multiple feature parameters output by the convolution layer 301 according to the output result of the determination layer 302;

[0026] The fully connected layer 304 outputs a predicted value of the vertical ultimate bearing capacity according to the multiple bearing capacity values ​​after weighted synthesis processing by the synthesis layer 303 .

[0027] The present invention has the following beneficial effects: The deep learning-based vertical bearing capacity calibration method of the present invention has the following advantages: Improved detection accuracy: The deep learning model comprehensively considers spatial and temporal characteristics, thereby improving the prediction accuracy of the vertical bearing capacity of the pipe piles. It reduces the complex operations and high costs required by traditional detection methods, and reduces the overall cost of detection. It realizes real-time evaluation of the bearing capacity of the pile foundation, providing a more timely and accurate means of engineering quality control. This method is not only applicable to prestressed concrete pipe piles, but can also be extended to other types of pile foundation detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic diagram of the process of predicting the ultimate bearing capacity of a pipe pile according to the present invention;

[0029] Figure 2 is a schematic diagram of a deep learning model of the present invention;

[0030] Figure 3 It is a basic structural diagram of the degree learning model of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the embodiments of the present invention Figure 1-Figure 3 , the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art.

[0032] The present invention provides a method for predicting the vertical bearing capacity of pipe piles based on deep learning, which can be applied to prestressed concrete pipe piles, etc. The method for predicting the vertical bearing capacity of pipe piles based on deep learning obtains theoretical values ​​and simulated values ​​of the vertical bearing capacity of pipe piles based on empirical parameter method and numerical simulation method, obtains a large amount of data, and then trains the data based on a deep learning model constructed by convolutional neural network and long short-term memory network (LSTM), compares the predicted value obtained by deep learning with the real data monitored, calculates the loss, performs back propagation according to the loss, adjusts the parameters of the deep learning model, determines the matching learning model parameters, and predicts the ultimate bearing capacity of pipe piles according to different soil layers and pipe pile parameters. Thus, the prefabricated pipe pile bearing capacity detection and evaluation method can achieve the technical effect of reducing the detection cost of pile foundation test and the danger of static load test.

[0033] A method for predicting the vertical bearing capacity of pipe piles based on deep learning comprises the following steps:

[0034] Step 1: Obtain the characteristic parameters of PHC piles and soil. The characteristic parameters of PHC piles include the geometric dimensions of the pile (inner diameter, outer diameter and length), pile deadweight, concrete grade, concrete strength value, etc. The soil of the pile includes the soil around the pile and the soil at the pile end. Construct a basic data set S1 containing soil layer parameters and pile parameters.

[0035] Optionally, soil characteristic parameter data are derived from local geological survey reports and based on experimental test data (indoor and in situ), and are classified according to soil type and soil layer burial depth distribution; basic physical and mechanical parameters of soil layers such as density, cohesion, internal friction angle, elastic modulus, etc.

[0036] Step 2: Use empirical parameters to preset the mechanical parameters in the basic data set S1, preset standard mechanical parameters corresponding to each type of soil layer, and convert the data into a shape suitable for CNN and LSTM input. The standard mechanical parameters include preset pile side friction resistance and preset pile end resistance. For monitoring the vertical settlement of the preset pile end and the preset pile end displacement, consider that each time point has different vertical settlements and corresponding bearing capacity values.

[0037] Step 3: According to the above preset data, the ultimate side friction resistance and pile end resistance of the pile body passing through each layer of soil are estimated by the empirical parameter method. Specifically, the estimated value of the theoretical vertical ultimate bearing capacity of the pile is obtained by the final empirical method. The obtained data is s1, and the empirical formula is as follows:

[0038]

[0039] Where S is the ultimate vertical bearing capacity of a single pile, d 1i is the outer diameter of the pipe pile in the i-th soil layer; L is the length of the pipe pile, A p is the pile end area of ​​the pipe pile, which can be calculated from the outer diameter and inner diameter. qpi is the ultimate lateral friction of the i-th soil layer on the pile body; q pi It is the pile tip resistance of the jth small area of ​​soil layer within the pile bottom range.

[0040] Step 4: Based on the soil layer parameters and pipe pile parameters in the basic data set S1, a PHC pipe pile numerical model is constructed using finite element software to calculate the ultimate vertical bearing capacity of the pipe pile. Specifically, a geometric model and mesh division are established, material parameters are defined / assigned, boundary conditions are set, and loads are applied for solution calculation to obtain the QS curve of the pipe pile and the numerical ultimate bearing capacity value of the pipe pile. The data is obtained as s2; finite element software such as ABAQUS, ANSYS, etc., but not limited to this, combines s1 and s2 to construct the actual bearing capacity training data set S2 of the pipe pile.

[0041] Step 5: Use convolutional neural network and long short-term memory network to build a pile foundation bearing capacity prediction model. Use convolutional neural network (CNN) to extract spatial features from sensor monitoring data, such as the vertical distribution of soil parameters around the pile. Specifically, you can use Keras in Python to build a combined model of CNN and LSTM. The code is as follows:

[0042] from keras.models import Sequential

[0043] from keras.layers import Dense,Dropout

[0044] model=Sequential()

[0045] model.add(Dense(64,activation='relu',input_dim=input_shape))

[0046] model.add(Dropout(0.2))

[0047] model.add(Dense(32,activation='relu'))

[0048] model.add(Dropout(0.2))

[0049] model.add(Dense(1,activation='linear'))

[0050] model.compile(optimizer='adam',loss='mean_squared_error')

[0051] model.fit(X_train,y_train,epochs=100,batch_size=32,validation_split=0.2)

[0052] Combined with the training set data set S2, the prediction model training is carried out, and the hyperparameters are optimized through grid search or random search methods, and the pile foundation bearing capacity prediction model is used in the actual pipe pile detection to predict the ultimate bearing capacity of the pipe pile;

[0053] refer to Figure 1 , including:

[0054] Step 100: Data collection: By installing sensors on site, real-time data such as strain and displacement of the pipe piles under different loads are collected. At the same time, standardized static load tests and dynamic tests are carried out to obtain the actual bearing capacity data of the pipe piles as a training set;

[0055] Step 200: Data preprocessing, cleaning and standardizing the collected sensor data and field test results to ensure data quality and consistency;

[0056] Step 300: Deep learning model construction, using convolutional neural network (CNN) to extract spatial features in sensor monitoring data, and long short-term memory network (LSTM) to model time series data.

[0057] Through the combination of these two networks, a pile foundation bearing capacity prediction model that comprehensively considers spatial and temporal characteristics is constructed;

[0058] Step 400: Model training and optimization, using the preprocessed data to train the deep learning model. Specifically,

[0059] Optimize model parameters to improve prediction accuracy. During the training process, cross-validation and hyperparameter tuning techniques are used to ensure the robustness and generalization ability of the model;

[0060] Step 500: Bearing capacity assessment, applying the trained deep learning model to actual pipe pile detection, predicting the vertical bearing capacity of the pile foundation in real time based on sensor data, and comparing it with the field test results to adjust the parameters of the deep learning model;

[0061] Step 600: Determine matching learning model parameters, and predict the ultimate bearing capacity of the pipe pile according to inputting different soil layers and pipe pile parameters.

[0062] The present invention provides a method for training a deep learning model using preprocessed data and optimizing model parameters to improve prediction accuracy. During the training process, cross-validation and hyperparameter tuning techniques are used to ensure the robustness and generalization ability of the model.

[0063] The present invention provides a method for applying a trained deep learning model to actual pipe pile detection, predicting the vertical bearing capacity of the pile foundation in real time based on sensor data, and comparing it with field test results to calibrate the prediction results.

[0064] refer to Figure 2 , the deep learning model, i.e. the pile foundation bearing capacity prediction model includes:

[0065] The convolution layer 301 extracts features of the input soil layer parameters and pipe pile parameters in sequence and outputs a plurality of feature parameters;

[0066] A determination layer 302, which determines whether each feature related to the location contained in the data is a feature of the location of the object of interest according to the input soil layer mechanical parameters;

[0067] A synthesis layer 303, which performs weighted synthesis processing on the features of the multiple outputs of the convolution layer 301 according to the output results of the determination layer 302, and the weight of the feature at the location of the object of interest is different from the weights of other features; and a fully connected layer 304, which outputs a vertical ultimate bearing capacity prediction value based on the multiple bearing capacity values ​​after the weighted synthesis processing by the synthesis layer 303.

[0068] The present invention provides a method for identifying a temporally continuous frame sequence by setting an LSTM layer in a deep learning model, thereby utilizing the characteristic of the LSTM layer of memorizing useful information and forgetting useless information over time.

[0069] The present invention provides, Figure 3 The schematic diagram of the convolutional neural network model shown in the figure consists of a five-layer structure, including the input layer (pile and soil characteristic parameters), the hidden layer and the output layer (vertical ultimate bearing capacity).

[0070] The embodiments described above are only descriptions of the preferred modes of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations, modifications, and substitutions made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for predicting the vertical bearing capacity of pipe piles based on deep learning, characterized in that: The following steps are involved: Step 1: Obtain characteristic parameters of pipe piles and soil, and construct a basic data set S1 including soil layer parameters and pipe pile parameters; Step 2: Preset the mechanical parameters in the basic data set S1 and preset the standard mechanical parameters corresponding to each type of soil layer; Step 3: Calculate and obtain the estimated value of the theoretical vertical ultimate bearing capacity of the pile, and the obtained data is s1; Step 4: Based on the soil layer parameters and pipe pile parameters in the basic geological data set S1, a pipe pile numerical model is constructed using finite element software to calculate the ultimate vertical bearing capacity of the pipe pile. The data obtained is s2. The s1 and s2 data are combined to construct the actual vertical bearing capacity training set data set S2 of the pipe pile; Step 5: Use convolutional neural networks and long short-term memory networks to build a pile foundation bearing capacity prediction model, and combine the data set S2 to carry out prediction model training, optimize model parameters, and use the pile foundation bearing capacity prediction model in actual pipe pile testing to predict the ultimate bearing capacity of the pipe pile.

2. The method for predicting the vertical bearing capacity of pipe piles based on deep learning according to claim 1, characterized in that: In step 1, the characteristic parameters of the pipe pile include the geometric dimensions of the pile body, the deadweight of the pile body, the concrete grade, and the concrete strength value.

3. The method for predicting the vertical bearing capacity of pipe piles based on deep learning according to claim 1, characterized in that: In step 3, the following formula is used to calculate and obtain the estimated value of the theoretical vertical ultimate bearing capacity of the pile: Where S is the ultimate vertical bearing capacity of a single pile, d 1i is the outer diameter of the pipe pile in the i-th soil layer; L is the length of the pipe pile, A p is the pile tip area of ​​the pipe pile, q pi is the ultimate lateral friction of the i-th soil layer on the pile body; q pi It is the pile tip resistance of the jth small area of ​​soil layer within the pile bottom range.

4. The method for predicting the vertical bearing capacity of a pipe pile based on deep learning according to claim 1, characterized in that: Step five includes: Step 100: Data collection: by installing sensors on site, real-time data of the pipe piles under different loads are collected to obtain the actual bearing capacity data of the pipe piles as a training set; Step 200: Data preprocessing, preprocessing the collected sensor data and field test results by cleaning and standardizing them; Step 300: deep learning model construction, using convolutional neural network to extract spatial features from sensor data, using long short-term memory network to model time series data, and constructing a pile foundation bearing capacity prediction model that comprehensively considers spatial and temporal features; Step 400: Model training and optimization, using the preprocessed data to train the deep learning model; Step 500: Bearing capacity assessment, applying the trained deep learning model to actual pipe pile detection, predicting the vertical bearing capacity of the pile foundation in real time based on sensor data, and comparing it with the field test results to adjust the parameters of the pile foundation bearing capacity prediction model; Step 600: Determine matching learning model parameters, and predict the ultimate bearing capacity of the pipe pile according to inputting different soil layers and pipe pile parameters.

5. A method for predicting the vertical bearing capacity of pipe piles based on deep learning according to claim 4, characterized in that: In step 300, the pile foundation bearing capacity prediction model includes: The convolution layer 301 extracts features of the input soil layer parameters and pipe pile parameters in sequence and outputs multiple feature parameters; Determine layer 302, based on the input soil layer mechanical parameters, determine whether each location-related feature included in the data is a feature of the location where the object of interest is located; The synthesis layer 303 performs weighted synthesis processing on multiple feature parameters output by the convolution layer 301 according to the output result of the determination layer 302; The fully connected layer 304 outputs a predicted value of the vertical ultimate bearing capacity according to the multiple bearing capacity values ​​after weighted synthesis processing by the synthesis layer 303 .

Images