A neural network-based tank hot buckling failure prediction method

Abstract

The application provides a kind of neural network-based tank thermal buckling failure prediction method.The method can realize the rapid prediction and evaluation of tank thermal buckling failure, lay the foundation for the safety detection of tank, and ensure the safety of oil and gas storage: the method mainly includes five steps: step one, collect the characteristic parameters of the tank;Step two, use finite element software to simulate the thermal buckling of the tank, so as to obtain the minimum critical temperature under different tank characteristic parameters;Step three, collect the characteristic parameters of the tank in the actual working condition;Step four, establish the neural network model of the tank after the failure of thermal buckling behavior, use the characteristic parameters of the tank in simulation as input, use the minimum critical temperature obtained by simulation as output, and train the neural network;Step five, the minimum critical temperature can be used to predict the safety of the tank failure;The step process of the method is shown in the abstract drawing.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of tank safety research, and particularly relates to a tank thermal buckling failure prediction method based on a neural network. BACKGROUND

[0002] A tank is a main storage mode of oil and natural gas, and as an important storage device, the tank is widely applied in many oil storage bases. With the continuous development of the petroleum and chemical industry in China, it means that the tank is large-scale and the scale of the tank farm is continuously expanded, and the safety of the tank becomes particularly important.

[0003] With the continuous increase of oil and gas consumption, large steel tanks as important facilities for industrial and strategic energy storage have been widely applied. According to the different geometric shapes, the tank is usually divided into vertical cylindrical tanks, horizontal cylindrical tanks and spherical tanks. Among them, the steel vertical cylindrical tank has the characteristics of small land occupation and low cost, and is widely applied in various chemical industry parks. The integrated design of tank farm buildings and facilities has become the direction of the construction and development of Chinese oil depots. If a serious fire accident occurs in a large oil and gas tank, it may trigger a domino effect, and the tank is easy to be affected by the synergistic effect of fire radiation and explosion shock wave.

[0004] Because the duration of the fire load and the explosion load is different in orders of magnitude, there is only a spatial synergistic effect without a time synergistic effect. The reason is that the duration of the explosion load is only a few milliseconds to a few hundred milliseconds, while the duration of the fire load is much longer, and the fire radiation has a delayed damage effect, so it is of great significance to determine the vulnerability and failure of the tank under the fire radiation for the safety of oil and gas.

[0005] Currently, the research methods for thermal buckling of storage tanks under fire radiation are mainly numerical simulation and CFD simulation. Numerical simulation is to solve the numerical value by using finite element method, linear buckling analysis is carried out by using boundary element method and equivalent geometry analysis method, the purpose is to evaluate the buckling mode and critical temperature value, and the buckling temperature predicted by geometric nonlinear analysis. CFD simulation is to use one of the most commonly used CFD software, fire dynamics simulator (FDS) to model the fire near the target tank, and then use the finite element software ANSYS Workbench to carry out thermal structure analysis. By using the implicit method of transient heat transfer, the finite element model (FEM) for thermal structure analysis realized in ANSYS Workbench, an experimental platform is built to realize the combined loading of adjacent pool fire and explosion shock wave on thin-walled steel tank. Or use the ABAQUS finite element transient heat transfer model to determine the temperature change of the side wall of the target tank as a function of fire passing time, so as to predict the thermal buckling behavior of the storage tank. However, these two prediction methods require a large amount of calculation time, and for the finite element method, the mesh division will also affect the prediction result, and when the thermal buckling behavior of the storage tank occurs, the damaged structure cannot be accurately predicted and positioned.

[0006] Based on the above analysis, it is shown that in the field of tank safety research, a simple, fast and accurate method for predicting the thermal buckling failure of the storage tank is urgently needed to ensure the safety of oil and gas storage.

[0007] Therefore, the present application proposes a storage tank thermal buckling failure prediction method based on neural network. The storage tank occurs thermal buckling under the action of fire, with the increase of fire temperature, the thermal deformation of the storage tank is intensified, which leads to the damage and failure of the storage tank. The present application uses finite element analysis software to calculate the temperature field distribution of the target tank, solves the thermal post-buckling behavior of the tank by artificial damping method, based on this principle, combined with the method of artificial neural network learning, the thermal buckling failure of the storage tank can be quickly predicted. This method does not need to measure the thermal radiation data of the storage tank, the operation is simple, the analysis is rapid, and it has important significance for the safety evaluation of the storage tank. SUMMARY

[0008] The application provides a neural network-based tank thermal buckling failure prediction method. The method can realize rapid prediction and evaluation of tank thermal buckling failure, lay a foundation for safety detection of the tank, and guarantee the safety of oil and gas storage. The neural network-based tank thermal buckling failure prediction method is characterized in that: firstly, the initial temperature of the environment is determined as 20℃, finite element analysis software is used to simulate the thermal buckling of the steel tank, the diameter, height, wall thickness, net distance between the tank and the flame, and the height and diameter of the flame are changed during the simulation process, and tank data under different working conditions are collected; secondly, sample data of different tanks are collected according to the characteristic parameters of the tanks in the database; thirdly, the sample data of the tank are taken as input, the thermal buckling result simulated by the finite element software is taken as output, and the neural network model is trained, and then the thermal buckling failure of the tank is accurately predicted based on the trained neural network model.

[0009] A neural network-based tank thermal buckling failure prediction method mainly includes the following contents:

[0010] (1) Collect the characteristic parameters of the tank. The characteristic parameters of the tank have an important influence on the finite element thermal buckling simulation and must be confirmed before simulation, including the material of the tank, the diameter, height, wall thickness of the tank, the net distance between the tank and the flame, and the height and diameter of the flame.

[0011] (2) Simulate the finite element thermal buckling of the tank. By changing the characteristic parameters of the tank, the thermal buckling behavior of the tank under different characteristic parameters is obtained. The tank thermal buckling will produce nonlinear large deformation. The lowest critical temperature under different tank characteristic parameters is counted, and the temperature fitting curve under different characteristic parameters is obtained by fitting with the finite element software.

[0012] (3) Collect the characteristic parameters of the tank in the actual working condition, analyze the tank characteristic parameters used in the finite element software simulation, and obtain the database of the characteristic parameters of the actual tank.

[0013] (4) Establish a neural network model of the tank thermal buckling behavior failure. According to the results obtained by the finite element software simulation, the tank characteristic parameters used in the simulation are taken as input, and the lowest critical temperature obtained by the simulation is taken as output. The neural network is trained. The input of the neural network is composed of the material of the tank, the diameter, height, wall thickness of the tank, the net distance between the tank and the flame, and the height and diameter of the flame.

[0014] (5) Failure prediction. Based on the established neural network model, the characteristic parameters of the tank under the actual working condition, i.e. the data of the tank characteristic parameters to be predicted, are taken as input, and the lowest critical temperature of the tank when thermal buckling occurs can be obtained. According to the lowest critical temperature, the safety of the tank failure can be predicted, and protective measures can be taken in advance. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a schematic diagram of a neural network structure Figure 1 Figure 2 is a flow chart of a method for predicting tank thermal buckling failure

[0016] Figure 3 is a schematic diagram of a neural network structure Figure 2 Figure 4 is a comparison chart of simulated and predicted minimum critical temperatures of a tank

[0017] Figure 3 Figure 5 is a comparison chart of simulated and predicted minimum critical temperatures of a tank

[0018] Figure 6 is a comparison chart of simulated and predicted minimum critical temperatures of a tank Figure 4 Figure 7 is a schematic diagram of a neural network structure DETAILED DESCRIPTION

[0019] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings, so that the advantages, features and application steps of the present application can be easily understood and accepted by those skilled in the art, and the protection scope of the present application can be more clearly defined.

[0020] A neural network-based tank thermal buckling failure prediction method mainly includes five steps, and the flow chart is shown in the accompanying drawings. The specific steps are as follows:

[0021] Step 1: Collect the characteristic parameters of the tank, including the material of the tank, the diameter, height, wall thickness of the tank, the net distance between the tank and the flame, and the height and diameter of the flame. The characteristic parameters of the tank can be obtained from the design data, construction data and completion data of the tank.

[0022] Step 2: Simulate the thermal buckling of the tank by using the finite element software, so as to obtain the minimum critical temperature under different characteristic parameters of the tank; the finite element software adopts the ADM algorithm (Artificial Damping Method), which is effective for thermal buckling and local buckling phenomena. When considering the total external load and internal force of the structure, the ADM algorithm also increases the "artificial damping force" to establish the balance of the three. When the model is stable, the damping ratio is small enough, and the artificial damping energy will not affect the overall balance of the model; when local buckling occurs in some areas of the structure, the change of the node displacement in the area will cause the damping energy to change with the corresponding node velocity, so when the damping ratio value changes rapidly, it indicates that buckling occurs in the local area, i.e. when the artificial damping ratio changes suddenly, the tank exhibits thermal buckling behavior, and the corresponding temperature at this time is the minimum critical temperature. During the simulation process, the data should be collected every 4m for the distance between the tank and the flame, every 2m for the diameter of the flame, and every 4m for the height of the flame position, so as to obtain a set of characteristic data (X, Y, Z) of the tank. Then, the corresponding minimum critical temperature (W) is simulated by the finite element software. The artificial damping ratio must be used as the critical measurement standard for the occurrence of tank thermal buckling during the finite element simulation.

[0023] ​Step three: collect the characteristic parameters of the storage tank in the actual working condition, analyze the characteristic parameters of the storage tank simulated by the finite element software, and obtain the basic characteristic parameters of the storage tank to be predicted; the sample parameters can be obtained by finite element software simulation, including the distance between the storage tank and the flame, the diameter of the flame, the position height of the flame and the minimum critical temperature. And the sample data of the storage tank with the same characteristic parameters is at least 10 groups. In the thermal buckling failure prediction of the storage tank, at least 10 groups of sample data with the same characteristic parameters of the storage tank to be predicted are needed to form the data structure shown in Table 1.

[0024] Table 1 data structure

[0025]

[0026] Step four: establish a neural network model of the storage tank after thermal buckling behavior failure, use 13 groups of data as training samples, use the characteristic parameters of the storage tank used in simulation as input, and use the minimum critical temperature obtained by simulation as output to train the neural network; according to the data structure in Table 1, the input layer of the neural network contains 3 nodes corresponding to the distance between the storage tank and the flame, the diameter of the flame and the position height of the flame, and the output layer contains 1 node corresponding to the minimum critical temperature of the storage tank when the thermal buckling behavior occurs; the middle layer of the neural network is set to 1 layer, and the number of nodes is set to 10; the structure of the neural network is shown in the accompanying Figure 2 ; 13 groups of data are used as 13 test data; the training function is set to Levenberg-Marquardt method.

[0027] Step five: failure prediction. Based on the neural network model established in step 5, the characteristic parameters of the storage tank in the actual working condition, i.e. the data of the characteristic parameters of the storage tank to be predicted, are used as input, and the minimum critical temperature of the storage tank when the thermal buckling occurs can be obtained, and the safety of the storage tank failure can be predicted according to the minimum critical temperature.

[0028] The following will combine a storage tank to predict its failure after thermal buckling by using the method described in the present application:

[0029] First, according to the method described in step one, collect the characteristic parameters of the storage tank. The material of the storage tank is Q345, and the target tank is in a windless state; the diameter of the target tank is 20m; the distance between the target tank and the flame is 40m; the detailed parameters are shown in the accompanying Figure 4 .

[0030] Second, according to the method described in step two, change the distance between the storage tank and the flame, the diameter of the flame and the position height of the flame, and simulate the thermal buckling of the storage tank by using the finite element software to obtain the minimum critical temperature under different characteristic parameters of the storage tank.

[0031] Thirdly, the characteristic parameters of the storage tank in actual working condition are collected to obtain the basic characteristic parameters of the storage tank to be predicted according to the method in step three.

[0032] Fourthly, the neural network is trained according to the method in step four, and the trained network is saved. The fitting curve of the minimum critical temperature of the storage tank is as shown in Figure 3

[0033] Fifthly, the minimum buckling temperature of the storage tank is predicted, and the failure is predicted by selecting the distance between the storage tank and the flame, the diameter of the flame and the position height of the flame.​

Claims

1. A neural network-based prediction method for hot buckling failure of a storage tank, characterized by The method for realizing the rapid prediction and evaluation of the thermal buckling failure of the storage tank by using the finite element software comprises the following steps: Step one: collecting the characteristic parameters of the storage tank, including the material of the storage tank, the diameter, height, wall thickness of the storage tank, the net distance between the storage tank and the flame, the height and diameter of the flame; the characteristic parameters of the storage tank are obtained from the design data, construction data and completion data of the storage tank; Step two: simulating the thermal buckling of the storage tank by using the finite element software, so as to obtain the minimum critical temperature under different characteristic parameters of the storage tank; the finite element software adopts the ADM algorithm (Artificial Damping Method), which is effective for the thermal buckling and local buckling phenomena; the ADM algorithm increases the "artificial damping force" when considering the total external load and internal force of the structure, and establishes the balance of the three; when the model is stable, the damping ratio is small enough, and the artificial damping energy does not affect the overall balance of the model; when local buckling occurs in some areas of the structure, the change of the node displacement in the area causes the damping energy to change with the corresponding node velocity, so when the value of the damping ratio changes rapidly, it indicates that buckling occurs in the local area, that is, when the artificial damping ratio changes suddenly, the storage tank exhibits thermal buckling behavior, and the corresponding temperature is the minimum critical temperature; during the simulation process, the distance between the storage tank and the flame should be collected every 4m, the diameter of the flame should be collected every 2m, and the height of the flame position should be collected every 4m, so as to obtain a set of characteristic data (X, Y, Z) of the storage tank, and then the corresponding minimum critical temperature (W) is simulated by using the finite element software; the artificial damping ratio must be used as the critical measurement standard for the occurrence of the thermal buckling of the storage tank during the finite element simulation; Step three: collecting the characteristic parameters of the storage tank in the actual working condition, analyzing the characteristic parameters of the storage tank used in the finite element software simulation, so as to obtain the basic characteristic parameters of the storage tank to be predicted; the sample parameters can be obtained by the finite element software simulation, including the distance between the storage tank and the flame, the diameter of the flame, the position height of the flame and the minimum critical temperature, and the sample data of the storage tank with the same characteristic parameters are at least 10 groups; in the prediction of the thermal buckling failure of the storage tank, at least 10 groups of sample data with the same characteristic parameters as the storage tank to be predicted are extracted, and the data structure shown in Table 1 is formed; Table 1 data structure Step four: a neural network model of the failure of the tank thermal buckling behavior is established, according to the results obtained by the finite element software simulation, 13 groups of data are taken as training samples, the characteristic parameters of the tank used in the simulation are taken as input, the lowest critical temperature obtained by simulation is taken as output, and the neural network is trained; according to the data structure in table 1, the input layer of the neural network contains 3 nodes, corresponding to the distance between the tank and the flame, the diameter of the flame and the position height of the flame, and the output layer contains 1 node, corresponding to the lowest critical temperature of the tank thermal buckling behavior; the middle layer of the neural network is set to 1 layer, and the node number is set to 10; 13 groups of data are also taken as 13 test data; the training function is set to Levenberg-Marquardt method (Levenberg-Marquardt algorithm); Step five: failure prediction; based on the neural network model established in step 5, the characteristic parameters of the tank under actual working conditions, i.e. the data of the characteristic parameters of the tank to be predicted, are taken as input, and the lowest critical temperature of the tank thermal buckling can be obtained, and the safety of the tank failure can be predicted according to the lowest critical temperature.

Images