A suction bucket soil failure mode prediction method based on a gan-bp algorithm

Abstract

The application provides a suction bucket uplift soil body failure mode prediction method based on a GAN-BP algorithm, and comprises the following steps: establishing a suction anchor uplift visualization experiment device; using the experiment device to perform suction anchor uplift experiments under different working conditions, and obtaining laser speckle image original videos of the suction anchor in different loading rate processes; processing and calculating the laser speckle image original videos to obtain soil body displacement contour maps around the suction bucket at the maximum uplift force; measuring the soil body crack width and depth through the soil body displacement contour maps, and establishing a soil body “factor-failure size” database; establishing a GAN-BP machine learning algorithm to perform small sample data prediction; the GAN-BP machine learning algorithm is established by considering the small sample database, and a suction bucket uplift soil body failure mode prediction model is established based on the algorithm; the measured data are obtained through the experiment device, and small sample expansion is performed, so that the influence law of the length, diameter and uplift rate of the suction bucket on the soil body crack size is obtained.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of suction caisson foundation, and particularly relates to a suction caisson uplift soil failure mode prediction method based on a GAN-BP algorithm. BACKGROUND

[0002] Offshore wind farms will develop in deeper waters in the future, and floating wind turbines are the inevitable choice for future development. Offshore wind power will move from nearshore to offshore and from shallow sea to deep sea. Floating wind turbines are connected to deep-sea suction anchors by anchor chains to withstand the uplift tension caused by wind and waves. The marine environment is complex and variable, often accompanied by severe typhoons and complex sea conditions. When storm surges arrive, the overturning moment of offshore wind power foundation is large, and the corresponding uplift force of the anchor cable transmitted to the suction anchor is large. Therefore, it is necessary to study and determine the uplift ultimate bearing capacity of the suction anchor to provide a theoretical basis for the installation and design of deep-sea wind power foundations, and the study of the mechanical properties of deep-sea suction anchors under storm surges has considerable engineering significance. Although a lot of research has been done on the uplift bearing characteristics of suction anchors, few people have focused on the deformation and fracture surface size of the inner and outer soil of the suction anchor during the uplift process. SUMMARY

[0003] The purpose of the present application is to provide a prediction method that can focus on the deformation and fracture surface size of the inner and outer soil of the suction anchor during the uplift process.

[0004] To achieve the above-mentioned purpose, the present application provides a suction caisson uplift soil failure mode prediction method based on a GAN-BP algorithm, which comprises the following steps:

[0005] S1: Establish a suction anchor uplift visualization experimental device;

[0006] S2: Use the experimental device to perform suction anchor uplift experiments under different working conditions, and obtain the original video of the laser speckle image of the suction anchor during the different loading rates;

[0007] S3: Process and calculate the original video of the laser speckle image to obtain the soil displacement contour map around the suction caisson at the maximum uplift force;

[0008] S4: Measure the width and depth of the soil fracture surface through the soil displacement contour map, and establish a soil “factor-damage size” database;

[0009] S5: Establish a GAN-BP machine learning algorithm to predict small sample data;

[0010] S6: According to the prediction result, analyze the influence trend of different factors on the foundation failure surface.

[0011] Further, in step S1, the suction anchor pullout visualization experiment device comprises a suction anchor system, an image acquisition system and a control mechanism: the suction anchor system comprises a hollow cylindrical suction anchor with a closed top end and an open bottom end, a valve-equipped water and air exhaust hole is arranged on the top end of the hollow cylinder, to discharge water and air trapped in the anchor during sinking, before pulling out, the valve is closed to seal the water and air exhaust hole. A pull rod is connected to the center of the anchor cover, and the pull rod is connected to a driving mechanism; a tension sensor is installed on the pull rod; the hollow cylinder is fixed in a glass box filled with transparent sand; the glass box is arranged in a counterforce frame; the driving mechanism, tension sensor and counterforce frame constitute a loading system; the control mechanism is signal connected with the loading system and the image acquisition system.

[0012] Further, the image acquisition system comprises a camera mechanism and a light source mechanism, the light source mechanism comprises a laser emitter, a linear lens and a laser power supply group, the laser power supply group is connected with the laser emitter by a cable, the laser emitter generates a fan laser speckle image by a linear lens, which is opposite to the suction anchor; the camera mechanism is arranged outside the counterforce frame and opposite to the side of the glass box, for shooting the laser speckle image; the image acquisition system is connected with the control mechanism, the control mechanism comprises an image processing system, the image processing system processes pictures in an incremental sequence by using PIVlab software, after processing the pictures, all incremental picture sequence results of the cloud image before the anchor reaches the maximum pullout force are superimposed, to obtain the displacement field of the suction anchor at the moment when the pullout force is maximum.

[0013] Further, the counterforce frame is welded by square steel pipes, which plays a role in stabilizing and transmitting counterforce; the driving mechanism is a linear servo electric cylinder, which is used to stably provide different speeds to drive the pull rod to move, the pull rod lifting range is 0-300mm, the pull range is 0-1000N, and the moving speed range is 1-500mm / s; the control mechanism is a computer, and the moving speed and lifting position can be adjusted through the touch panel on the control box (computer). The center of the anchor cover is provided with a threaded hole, and the pull rod is connected through the threaded hole.

[0014] Further, step S2 is specifically:

[0015] S2.1: design working conditions, the parameters that can be changed in the working conditions include the inner diameter of the suction barrel, the outer diameter of the suction barrel, the length of the suction barrel and the pullout force loading rate;

[0016] S2.2: fill the configured transparent sand into the organic glass box;

[0017] S2.3: The driving mechanism is used to penetrate the suction anchor into the transparent sand in the glass box, specifically: first, open the suction anchor drain valve, adjust the position of the bottom glass box, so that the suction anchor is in the center of the plane of the glass box; second, use a servo cylinder to penetrate the suction anchor into the designated position in the transparent sand;

[0018] S2.4: After the suction anchor is penetrated into the transparent sand, it is left for a period of time;

[0019] S2.5: Land visualization system arrangement, start and adjust the image acquisition system, complete the preparation work of acquisition;

[0020] S2.6: Pull the suction anchor upwards, and record the original video of the laser speckle image and the reading of the tension sensor during the pulling process at different loading rates;

[0021] Further, in step S2.5, the land visualization system arrangement is specifically: adjust the position of the transparent glass box so that the suction anchor is directly opposite the center of the model box plane; adjust and arrange the distance between the laser emitter and the glass box so that the laser can completely illuminate the inside of the transparent sand after passing through the linear optical prism, producing a stable fan laser speckle image. Adjust the aperture size and ISO value of the camera lens to obtain clear and visible pictures.

[0022] Further, in step S2.5, the preparation work of acquisition is specifically: start the video recording of the camera mechanism, prepare to record the laser speckle image of the suction anchor during the process of different loading rates; connect the data acquisition instrument of the tension sensor to the computer, start the computer, start the data acquisition system, and prepare to record the pulling force of the top of the suction anchor in real time.

[0023] Further, in step S2.6, the way to pull the suction anchor is displacement control, the pulling speed is controlled by the driving mechanism, the suction anchor is pulled at different loading rates, and the reading of the tension sensor is recorded by the microcomputer.

[0024] Further, step S3 is specifically:

[0025] S3.1: After the laser speckle image original video is black and white, extract each frame of the video in batches;

[0026] S3.2: Import the extracted laser speckle slices at different times and working conditions into the image processing and analysis software for picture processing and calculation;

[0027] S3.3: After exporting the calculation data, draw the soil velocity field, displacement field and displacement vector diagram during the pulling process, and then calculate the maximum pulling force to obtain the soil displacement contour map around the suction anchor cylinder.

[0028] Further, in step S4, the factors of the "factor-rupture size" database include the length of the suction bucket, the diameter of the suction bucket, and the pull-up rate; and the rupture sizes of the "factor-rupture size" database include the width of the fracture surface and the depth of the fracture surface.

[0029] Further, in step S5, the GAN-BP machine learning algorithm includes a GAN module and a BP module; and the method for establishing the GAN-BP machine learning algorithm model is:

[0030] S5.1: generating high-quality data samples through the GAN module, so that the generated data is close to the real data distribution;

[0031] S5.2: combining the high-quality data samples generated by the GAN module with the original data to form a new combined training set;

[0032] S5.3: training the BP module using the combined training set to establish a GAN-BP hybrid model.

[0033] Further, the GAN module is a generative adversarial network, and the BP module is a BP neural network;

[0034] Further, in step S5.1, the method for generating high-quality data samples through the GAN module is:

[0035] S5.11: defining a generator G(z; θ G ) and a discriminator D(x; θ D );

[0036] Generator G(z; θ G ): input random noise z ~ p z (z), generate sample x fake =G(z);

[0037] Discriminator D(x; θ D ): input data x, output the probability that the data is real D(x)

[0038] S5.12: train the discriminator D and the generator G by optimizing the loss function until the samples generated by the generator G are difficult to distinguish;

[0039] The goal of the GAN module is to maximize the ability of the discriminator D to distinguish real data and generated data, while minimizing the probability of being recognized by the discriminator G; therefore, the objective function is:

[0040]

[0041] By optimizing the above loss function, D and G are trained until the samples generated by the generator are difficult to distinguish;

[0042] S5.13: After training, use the generator G as input to random noise z to generate augmented samples x. augmented The expanded sample x augmented This is the expanded data sample of the "factor-damage size" database obtained in step S4;

[0043] The training method for the BP module is as follows:

[0044] S5.31: Prepare the dataset, X train =[X real ,X augmented ],Y train =[Y real ,Y augmented ];

[0045] S5.32: Define a BP neural network, which includes an input layer, a hidden layer, and an output layer;

[0046] enter:

[0047] Output:

[0048] Parameters: weight matrix W and bias b;

[0049] S5.33: Forward propagation, using the activation function f, calculates the values ​​of the hidden layer and the output layer:

[0050] h = f(W) hidden x+b hidden )

[0051]

[0052] In the formula, f: activation function, used to introduce nonlinearity, commonly such as Sigmoid, ReLU, etc.; h: output value of the hidden layer; W hidden : Weight matrix from input layer to hidden layer; b hidden : The bias vector of the hidden layer; W output : Weight matrix from hidden layer to output layer; b output : The bias vector of the output layer;

[0053] S5.34: Use the mean squared error (MSE) as the loss function:

[0054]

[0055] In the formula, Loss function; N: total number of training data samples; y i : The true value of the i-th sample; Predicted value of the i-th sample;

[0056] S5.35: Back propagation, optimize network parameters by gradient descent.

[0057]

[0058] In the formula, η: learning rate, control parameter update step size; Partial derivative of the loss function with respect to the weight matrix W; Partial derivative of the loss function with respect to the bias vector b.

[0059] Further, in step S6, the influence trend of different factors on the foundation failure surface is analyzed, and the development law of the foundation failure envelope surface is comprehensively judged by using image analysis and machine learning methods.

[0060] Compared with the prior art, the advantages of the present application are:

[0061] 1. The present application considers the small sample database to establish a GAN-BP machine learning algorithm, and based on the algorithm, a suction bucket uplift soil failure mode prediction model is established, the measured data is obtained through the suction anchor uplift visualization experimental device, and the small sample is expanded, so that the influence law of the length, diameter and uplift rate of the suction bucket on the soil failure surface size is obtained.

[0062] 2. The present application considers the limitations of experimental devices and conditions, and establishes a hybrid machine learning method for small sample database, which combines the advantages of good mapping relationship between multiple input and output of BP neural network algorithm and small sample expansion problem of GAN algorithm.

[0063] 3. The present application adopts machine learning method to comprehensively consider various factors to predict the foundation bearing capacity, and controls a single variable to explain the foundation bearing capacity, and explains the suction bucket uplift foundation bearing capacity from the single factor and multi-factor coupling angle.

[0064] 4. The present application sets up the suction anchor uplift visualization experimental device, and uses experimental data as machine learning sample library, which avoids the disadvantage that the prediction result does not match the actual result caused by using numerical simulation results as sample library. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 The flow chart of the suction bucket uplift soil failure mode prediction method based on the GAN-BP algorithm proposed in the embodiment of the present application is shown in the figure;

[0066] Figure 2 The schematic diagram of the suction anchor uplift visualization experimental device in the embodiment of the present application is shown in the figure;

[0067] Figure 3Figure 2 is a sectional view of a barrel suction anchor structure of a visual test device in the method of an embodiment of the present application;

[0068] Figure 4 Figure 6 is a soil displacement contour map around a suction barrel corresponding to different loading speeds of a suction anchor in the method of an embodiment of the present application;

[0069] Figure 5 Figure 7 is a schematic diagram of a local failure mode and a global failure mode of a barrel suction anchor in the method of an embodiment of the present application;

[0070] Figure 6 Figure 8 is a schematic diagram of a comparison result of a predicted value and an actual value of a fracture surface width B and a fracture surface depth H by a GAN-BP algorithm in the method of an embodiment of the present application;

[0071] Figure 7 Figure 9 is a schematic diagram of a comparison result of a predicted value and an actual value of a fracture surface width B and a fracture surface depth H by a BP algorithm in the method of a comparative example. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be further described below.

[0073] In the following examples, BP refers to a BP neural network (Back Propagation Artificial Network), and GAN refers to a generative adversarial network (Generative Adversarial Network).

[0074] The present embodiment proposes a suction barrel uplift soil body failure mode prediction method based on a GAN-BP algorithm, as shown in Figure 1, which comprises the following steps: Figure 1

[0075] S1: Establish a suction anchor uplift visual test device;

[0076] In the present embodiment, as shown in Figure 2, the suction anchor uplift visual test device comprises a suction anchor system, an image acquisition system and a microcomputer 11: Figure 2

[0077] The suction anchor system comprises a hollow thin-walled barrel suction anchor 2 with a closed top end and an open bottom end, which is made of a series of acrylic organic glass, as shown in Figure 3: Figure 3 ​​As shown, the top end of the hollow barrel suction anchor 2 is provided with a level 23 and a valve-equipped drainage vent 21 for draining water and air trapped in the anchor during sinking, and the valve is closed to seal the drainage vent 21 before pulling up. A 8mm-diameter threaded hole 22 is provided at the center of the anchor cap, through which a pull rod is connected, and the top end of the pull rod is connected to the linear servo motor cylinder 6 through a flange 4; a tension sensor 5 is also installed on the pull rod, and the tension sensor 5 is connected to a tension sensor data acquisition instrument 8 through a cable; the suction anchor 2 is fixed in an 18cm x 18cm x 40cm acrylic organic glass box 1, the acrylic organic glass box 1 has a wall thickness of 1cm, and the inside is filled with transparent sandy soil 3. The glass box 1 is fixed in a metal counterforce frame 9, and the top end of the counterforce frame 9 is fixed with a guide rod 7 for controlling the relative position of the suction barrel in the soil.

[0078] Among them, the linear servo motor cylinder 6, the tension sensor 5 and the metal counterforce frame 9 constitute a loading system, and the microcomputer 11 is signal connected with the loading system and the image acquisition system.

[0079] In this embodiment, the counterforce frame 9 is welded from square steel pipes, which plays a role in stabilizing and transmitting counterforce; the linear servo motor cylinder 6 is drivingly connected with the pull rod for stably providing different speeds to drive the pull rod to move, and the pull rod has a lifting range of 0-300mm, a tension range of 0-1000N and a moving speed range of 1-500mm / s; the moving speed and the lifting position can be controlled and adjusted by the microcomputer 9; the tension sensor has a range of 500N and an accuracy of 0.01N, and the tension data acquisition card has a sampling frequency of 10Hz.

[0080] As shown in the figure, Figure 2 The image acquisition system includes a Canon camera 10 and a light source mechanism, and the light source mechanism includes a laser emitter 12, a linear lens 13 and a laser power supply group 14, wherein the laser power supply group 14 is connected with the laser emitter 12 through a cable, the laser emitter 12 passes through the linear lens 13 and directly faces the suction anchor 2 in the glass box 1 to generate a fan laser speckle image; the camera 10 is arranged on the side surface of the counterforce frame 9 outside and directly faces the side surface of the glass box 1 for shooting the laser speckle image generated by the light source mechanism. The image acquisition system is signal controlled and connected with the microcomputer 11, and the control system of the microcomputer 11 includes an image processing system, which carries out 1g suction anchor high-speed pulling model test soil visualization research based on transparent soil and PIV technology, processes pictures in the form of incremental sequence by using PIVlab software, superimposes all incremental picture sequence result cloud maps of the suction anchor 2 before reaching the maximum pulling force to obtain the displacement field of the suction anchor 2 at the moment when the pulling force is maximum.

[0081] S2: Using the above experimental device, the suction anchor pull-out experiment under different working conditions is carried out, and the original video of the laser speckle image of the suction anchor in the process of different loading rates is obtained.

[0082] The specific implementation of step S2 is:

[0083] S2.1: Design the working condition, and the parameters that can be changed include the inner diameter of the suction bucket, the outer diameter of the suction bucket, the length of the suction bucket and the pull-out force loading rate;

[0084] S2.2: Fill the configured transparent sand 3 into the organic acrylic glass box 1;

[0085] S2.3: Use the linear servo cylinder 6 to penetrate the suction anchor 2 into the transparent sand 3 in the glass box 1, the specific method is: first, open the suction anchor drain valve 21, adjust the position of the bottom glass box 1, so that the suction anchor 2 is opposite to the center of the bottom plane of the glass box 1; second, use the servo cylinder 6 to penetrate the suction anchor 2 into the designated position in the transparent sand 3,

[0086] In this embodiment, five suction anchors are selected, which are numbered M-1, M-2, M-3, M-4 and M-5, and the average diameter L, the side wall thickness t1 and the top wall thickness t2 parameters are shown in Table 1,

[0087] Table 1 Suction anchor size parameter table

[0088]

[0089] S2.4: After the suction anchor 2 penetrates into the transparent sand 3, it is left for a period of time;

[0090] S2.5: Land visualization system arrangement, start and adjust the image acquisition system, complete the acquisition preparation work.

[0091] In this embodiment, the land visualization system arrangement method is: adjust the position of the transparent glass box 1, so that the suction anchor 2 is opposite to the center of the bottom plane of the glass box 1; adjust and arrange the distance between the laser emitter 12 and the glass box 1, so that the laser can completely illuminate the inside of the transparent sand 3 after passing through the linear optical prism 13, thereby generating stable fan laser speckle images. Adjust the lens aperture size and ISO value of the camera 10, so that the pictures obtained are clear and visible;

[0092] The acquisition preparation work is specifically: start the video recording of the camera 10, prepare to record the laser speckle images of the suction anchor 2 in the process of different loading rates; connect the tension sensor data acquisition instrument 8 to the computer 10 through the cable, start the computer 10, start the data acquisition system, and prepare to record the top pull-out force of the suction anchor 2 in real time during the pull-out process.

[0093] S2.6: Pull up the suction anchor 2 and record the real-time video of the pulling process of the suction anchor 2, and record the original video of the laser speckle image and the reading of the tension sensor during the different loading rates. Among them, the way to pull the suction anchor 2 is displacement control, and the pulling speed is controlled by a linear servo cylinder 6. The suction anchor 2 is pulled up at different loading rates, and the reading of the tension sensor 5 is recorded by the microcomputer 10, as shown in the following Table 2:

[0094] Table 2: Pulling loading rate record table of different suction anchors

[0095]

[0096] S3: Process and calculate the original video of the laser speckle image to obtain the soil displacement contour map around the suction anchor barrel at the maximum pulling force;

[0097] S3.1: After obtaining the original video of the laser speckle image by recording the pulling process of the suction anchor with the camera 10, the original video of the laser speckle image is black and white by using the Pr editing software, and then each frame of the video is extracted in batches by using the FreeVideoToJPGConverter software;

[0098] S3.2: The extracted laser speckle slices at different times and working conditions are imported into the image processing and analysis software PIVlab for picture processing and calculation;

[0099] S3.3: After exporting the calculation data, the soil velocity field, displacement field and displacement vector diagram during the pulling process are drawn by using the Origin software, and the maximum pulling force F max is obtained by programming calculation by using MATLAB, and the soil displacement contour map around the suction anchor barrel at the maximum pulling force is obtained.

[0100] As shown in Figure 4 , the soil displacement contour map around the suction barrel corresponding to the suction anchor with a diameter D = 50 mm and a length L = 25 mm at different loading speeds, wherein, Figure 4 (a) is the soil displacement contour map around the suction barrel corresponding to the suction anchor at a loading speed v = 1 mm / s; Figure 4 (b) is the soil displacement contour map around the suction barrel corresponding to the suction anchor at a loading speed v = 12.5 mm / s; Figure 4 (c) is the soil displacement contour map around the suction barrel corresponding to the suction anchor at a loading speed v = 25 mm / s; Figure 4 (d) is the soil displacement contour map around the suction barrel corresponding to the suction anchor at a loading speed v = 50 mm / s; Figure 4(e) is a contour map of the land displacement around the suction bucket when the suction anchor is loaded at a speed of v = 100 mm / s.

[0101] S4: Measure the width and depth of the land fracture surface using land displacement contour maps, and establish a "factor-damage size" database. The "factors" in this database include the length of the suction barrel, the diameter of the suction barrel, and the upward pull rate; the "damage size" includes the fracture surface width B and the fracture surface depth H.

[0102] Table 3 below records the "factor-damage size" data for suction anchor M-3 (diameter D=50mm, length L=25mm) and suction anchor M-1 (average diameter D=50mm, length L=50mm) at pull-out rates V=12.5mm, V=25mm, V=50mm, and V=100mm, respectively.

[0103] Table 3. "Factor-Damage Size" Database (Partial)

[0104]

[0105] S5: Establish the GAN-BP machine learning algorithm for prediction of small sample data;

[0106] In this embodiment, the GAN-BP machine learning algorithm includes a GAN module and a BP module. The method for establishing the GAN-BP machine learning algorithm model is as follows:

[0107] S5.1: Generate high-quality data samples using the GAN module to make the generated data closely resemble the distribution of real data. The method is as follows:

[0108] S5.11: Define the generator G(z; θ) G ) and discriminator D(x; θ D );

[0109] Generator G(z; θ) G Input random noise z~p z (z), generate sample x fake =G(z);

[0110] Discriminator D(x; θ) D ): Input data x, output probability D(x) that the data is true.

[0111] S5.12: Train the discriminator D and generator G by optimizing the loss function until the samples generated by the generator G are difficult to distinguish;

[0112] The goal of the GAN module is to maximize the discriminator D's ability to distinguish between real and generated data, while minimizing the probability of the generator G being detected by the discriminator; therefore, its objective function is:

[0113]

[0114] By optimizing the above loss function, D and G are trained until the samples generated by the generator are difficult to distinguish;

[0115] S5.13: After training, use the generator G to input random noise z to generate augmented samples x augmented , the augmented sample x augmented is the augmented data sample obtained in step S4, and part of the data is shown in Table 4:

[0116] Table 4 GAN augmented samples (part)

[0117] Diameter D (mm) Length L (mm) Loading rate v (mm / s) Fracture surface width B (mm) Fracture surface depth H (mm) 44.31518173 68.41151428 28.2653923 84.93772888 90.00748444 54.74298096 57.27037048 22.36504173 83.10439301 84.56375885 45.3184166 61.41879654 11.4386301 66.49040985 81.75670624 40.28317261 60.88147736 27.43096733 83.2386322 76.64796448 52.46657562 73.30093384 80.52485657 78.88573456 112.2959061 51.99222183 50.20989227 21.87613678 79.08688354 79.97119904 43.28781128 52.85329819 21.70853233 79.03794861 81.4047699 47.32524109 47.0251503 52.17623138 88.63056183 86.86373138 44.13137436 61.69446182 20.72089386 81.01996613 80.69803619 57.11587906 49.82167053 22.88331604 83.21014404 82.46383667 57.58080292 54.06962585 12.35421562 73.38467407 80.63221741 43.01916122 62.54060745 21.55706215 82.27147675 77.31987 58.78965378 49.02116776 39.20677185 89.68482971 90.51895142 47.80777359 35.50246429 95.53018951 81.99420929 69.41500092 43.06548309 51.82941055 28.85625458 85.93470764 75.40726471 55.61429977 60.53258133 10.71063805 73.97647858 83.76602173 54.35240173 19.05724716 99.8663559 65.19730377 54.34417725 49.18434906 56.70740509 24.64125061 83.2882843 79.88928223 47.5256691 71.38578796 40.03570557 70.89803314 103.9092407 44.07012558 35.29263306 56.65210724 67.41448975 57.59306717

[0118] S5.2: Combine the high-quality data samples generated by the GAN module with the original data to form a new combined training set X train =[X real ,X augmented ],Y train =[Y real ,Y augmented ] X train is the input data, including the length, diameter of the suction bucket and the pull-up rate; Y train is the output data, including the width and depth of the fracture surface;

[0119] S5.3: Train the BP module using the above combined training set to establish a GAN-BP hybrid model, the method being as follows:

[0120] 1) Define a BP neural network, which includes an input layer, a hidden layer and an output layer;

[0121] Input:

[0122] Output:

[0123] Parameters: weight matrix W and bias b;

[0124] 2) Forward propagation, calculate the values of the hidden layer and the output layer through the activation function f:

[0125] h=f(W hidden x+b hidden )

[0126]

[0127] In the formula, f: activation function, used to introduce nonlinearity, commonly such as Sigmoid, ReLU, etc.; h: output value of the hidden layer; Whidden : weight matrix from input layer to hidden layer; b hidden : bias vector of hidden layer; W output : weight matrix from hidden layer to output layer; b output : bias vector of output layer;

[0128] 3) Mean Squared Error (MSE) is used as the loss function:

[0129]

[0130] where, loss function; N: total number of training data samples; y i : true value of the i-th sample; : predicted value of the i-th sample;

[0131] 4) Backpropagation, optimize network parameters by gradient descent.

[0132]

[0133] where, η: learning rate, control the step size of parameter update; partial derivative of the loss function with respect to the weight matrix W; partial derivative of the loss function with respect to the bias vector b.

[0134] S6: According to the prediction results, analyze the influence trend of different factors on the foundation failure surface, and comprehensively judge the development law of the foundation failure envelope surface by image analysis and machine learning methods.

[0135] From the displacement contour maps under different working conditions as shown in Figure 4 , it can be seen that when the suction anchor is pulled up at different loading rates, the soil failure mode can be roughly divided into overall failure mode and local failure mode. When the pulling rate v = 1 mm / s, the soil around the suction bucket has basically no vertical displacement, and the failure mode of the suction bucket is local failure mode, and a clear shear fracture zone along the bucket wall can be seen, as shown in Figure 5 (a). When the pulling rate of the suction bucket v = 100 mm / s, the soil plug in the suction bucket and part of the foundation soil at the bottom of the bucket will be pulled out with the anchor, as shown in Figure 5 (b), the failure mode of the anchor is overall failure mode, at this time, the suction bucket is in overall failure mode, and the peak pulling force reaches the undrained ultimate bearing capacity. When 1 mm / s < v < 100 mm / s, the suction bucket is in partial drainage state, and with the increase of the loading rate, the soil flow range caused by the bucket pulling also increases.

[0136] The prediction results of the GAN-BP machine learning algorithm model in the method of the embodiment are compared with the measured data, and the sample number obtained by sampling any sample in the database is taken as the abscissa, and the GAN-BP algorithm output value and the true value y i are taken as the ordinate, as shown in the comparison result schematic diagram shown in Figure 6 , wherein, Figure 6 (a) is the comparison result of the prediction value and the measured value of the fracture surface width B by the GAN-BP algorithm, Figure 6 (b) is the comparison result of the prediction value and the measured value of the fracture surface depth H by the GAN-BP algorithm, and the mean square error of the test set is 0.0189, and the average absolute error is 0.1161.

[0137] Comparative example

[0138] In the comparative example, the original sample without GAN data expansion is directly used for BP neural network fitting, and the comparison result of the prediction value and the measured value is as shown in Figure 7 , wherein, Figure 7 (a) is the comparison result of the prediction value and the measured value of the fracture surface width B by the BP algorithm, Figure 7 (b) is the comparison result of the prediction value and the measured value of the fracture surface depth H by the BP algorithm, and the mean square error of the test set is 0.0910, and the average absolute error is 0.2617.

[0139] Therefore, it can be seen that the GAN-BP hybrid algorithm greatly improves the robustness and accuracy of the model, and can better explain the mapping relationship between the diameter, length and loading rate of the suction bucket and the size of the foundation failure envelope, and provide a reference for the design of the foundation bearing capacity.

[0140] The above is only a preferred embodiment of the present application, and does not limit the present application in any way. Any person skilled in the art can make any form of equivalent replacement, modification or change of the technical solutions and technical contents disclosed in the present application without departing from the scope of the technical solutions of the present application, and still belongs to the protection scope of the present application.

Claims

1. A method for predicting the failure mode of soil body on suction caisson based on GAN-BP algorithm, characterized in that, The method comprises the following steps: S1: establishing a suction anchor pulling visualization experimental device; S2: using the experimental device to carry out suction anchor pulling experiments under different working conditions, and obtaining original videos of laser speckle images of the suction anchor in the process of different loading rates; S3: processing and calculating the original videos of the laser speckle images to obtain soil displacement contour maps around the suction bucket at the maximum pulling force; S4: measuring the width and depth of the soil rupture surface through the soil displacement contour maps, and establishing a soil factor-damage size database; In step S4, the factors of the soil factor-damage size database include the length of the suction bucket, the diameter of the suction bucket, and the pulling rate; and the damage size of the soil factor-damage size database includes the width of the rupture surface and the depth of the rupture surface; S5: establishing a GAN-BP machine learning algorithm for small sample data prediction; In step S5, the GAN-BP machine learning algorithm comprises a GAN module and a BP module; and the method for establishing the GAN-BP machine learning algorithm model is as follows: S5.1: generating high-quality data samples through the GAN module so that the generated data is close to the distribution of the real data; S5.2: combining the high-quality data samples generated by the GAN module with the original data to form a new combined training set; S5.3: training the BP module using the combined training set to establish a GAN-BP hybrid model; S6: analyzing the influence trend of different factors on the foundation rupture surface according to the prediction result.

2. The suction bucket soil failure mode prediction method of claim 1, wherein, In step S1, the suction anchor pulling visualization experimental device comprises a suction anchor system, an image acquisition system, and a control mechanism: The suction anchor system comprises a hollow cylindrical suction anchor with a closed top end and an open bottom end, a drainage and exhaust hole with a valve is arranged on the top end cover of the hollow cylindrical suction anchor, a pull rod is connected to the center of the top end cover, and the pull rod is connected to a driving mechanism; a tension sensor is installed on the pull rod; the hollow cylindrical suction anchor is fixed in a glass box, the glass box is filled with transparent sand soil; and the glass box is arranged in a counterforce frame; The driving mechanism, the tension sensor, and the counterforce frame constitute a loading system; The control mechanism is signal connected with the loading system and the image acquisition system.

3. The suction bucket soil failure mode prediction method of claim 2, wherein, The image acquisition system comprises a camera mechanism and a light source mechanism, the light source mechanism comprises a laser emitter, a linear lens, and a laser power supply group, the laser power supply group is connected with the laser emitter through a cable, the laser emitter generates a fan laser speckle image through the linear lens and directly faces the suction anchor; the camera mechanism is arranged on the side of the counterforce frame and directly faces the glass box, and is used for shooting the laser speckle image; the image acquisition system is connected with the control mechanism, the control mechanism comprises an image processing system, the image processing system processes pictures in an incremental sequence manner by using PIVlab software to obtain a displacement field of the suction anchor at the moment when the pulling force is maximum.

4. The suction bucket soil failure mode prediction method of claim 2, wherein, The counterforce frame is welded by square steel pipes; the driving mechanism is a linear servo electric cylinder, which is used to provide different speeds to drive the pull rod to move, the lifting range of the pull rod is 0-300mm, the pulling force range is 0-1000N, and the moving speed range is 1-500mm / s; the control mechanism is a computer.

5. The suction bucket soil failure mode prediction method of claim 2, wherein, Step S2 is specifically: S2.1: design working conditions, the parameters that can be changed include the inner diameter of the suction bucket, the outer diameter of the suction bucket, the length of the suction bucket and the uplift force loading rate; S2.2: fill the configured transparent sand into the plexiglass box; S2.3: use the driving mechanism to penetrate the suction anchor into the transparent sand in the glass box; S2.4: after the suction anchor penetrates into the transparent sand, stand for a period of time; S2.5: land visualization system arrangement, turn on and adjust the image acquisition system, complete the acquisition preparation work; S2.6: pull up the suction anchor, and record the laser speckle image original video and the tension sensor reading of the suction anchor in the process of different loading rates.

6. The suction bucket soil failure mode prediction method of claim 5, wherein, In step S2.6, the method of pulling the suction anchor is displacement control, the pulling speed is controlled by the driving mechanism, and the suction anchor is pulled up at different loading rates.

7. The suction bucket soil failure mode prediction method of claim 1, wherein, Step S3 is specifically: S3.1: after the laser speckle image original video is black and white, extract each frame of the video in batches; S3.2: import the extracted laser speckle slices at different times and under different working conditions into the image processing analysis software for picture processing and calculation; S3.3: after exporting the calculation data, draw the soil velocity field, displacement field and displacement vector diagram in the pulling process, and then calculate the maximum uplift force to obtain the soil displacement contour map.

8. The suction bucket soil failure mode prediction method of claim 1, wherein, In step S5.1, the method for generating high-quality data samples by the GAN module is: S5.11: Define the generator G(z; 0 G ) and the discriminator D(x; 0 D ). S5.12: train the discriminator D and the generator G by optimizing the loss function until the samples generated by the generator G are difficult to distinguish; S5.13: After training, use the generator G as input to random noise z to generate augmented samples x. augmented The expanded sample x augmented This is the expanded data sample of the "Factor-Damage Size" database in step S4; The training method of the BP module is: S5.31 : Prepare dataset, X train = [X real , X augmented ], Y train = [Y real , Y augmented ]; S5.32: define a BP neural network, which includes an input layer, a hidden layer and an output layer; S5.33: forward propagation, calculate the values of the hidden layer and the output layer through the activation function f; S5.34: use mean square error MSE as the loss function; S5.35: optimize the network parameters by gradient descent.

Images