CN-122021323-A - Dynamic response prediction method for deepwater explosion container

CN122021323ACN 122021323 ACN122021323 ACN 122021323ACN-122021323-A

Abstract

The invention discloses a method for predicting dynamic response of a deepwater explosion container, which relates to the technical field of deepwater structure safety protection, and is characterized in that a deep learning prediction model is established, material characteristic parameters and external load parameters are taken as inputs, container dynamic response indexes are taken as outputs, and model training, verification and testing are performed by using a sample set to obtain the deepwater explosion container dynamic response prediction model capable of rapidly predicting the container dynamic response. The method and the device construct a high-quality training data set through numerical simulation, ensure the physical reality and the prediction precision of the machine learning model, and the trained model can realize the rapid prediction of dynamic response, effectively solve the problem that the precision and the efficiency are difficult to balance in the existing method, and meet the requirements of rapid evaluation and real-time early warning in engineering. The data set covers the explosion load parameter and the deepwater environment parameter, so that the defect of narrow application range of the traditional empirical formula is avoided, and the engineering application cost is reduced.

Inventors

LI LINNA
HUANG YU
ZHU ZHEN
ZHANG ZHENG
XU XIAOXIAO
GAO HAN
TAO HAOHAO

Assignees

武汉科技大学

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The method for predicting the dynamic response of the deepwater explosion container is characterized by comprising the following steps of: Step 1, acquiring relevant experimental data of a deepwater explosion container, wherein the experimental data comprise container structure parameters, explosive loading quantity, environmental hydrostatic pressure, strain measuring point positions and corresponding actually measured strain values, and preprocessing the experimental data; Step 2, based on the preprocessed experimental data, constructing a three-dimensional dynamics simulation model of the deepwater explosion container by adopting finite element analysis software, and verifying the reliability of the model by comparing a numerical calculation result with measured strain data; Step 3, randomly sampling material characteristic parameters and external load parameters which influence the dynamic response of the container by using a Latin hypercube sampling method to generate a plurality of groups of parameter combination samples; Step 4, based on a plurality of groups of parameter combination samples, establishing a batch simulation calculation flow by utilizing an automatic script and finite element analysis software, and acquiring a dynamic response result of the container under multiple working conditions to expand a data set; step 5, fusing experimental actual measurement data with batch numerical simulation data, constructing a unified sample set after data preprocessing, and dividing the unified sample set into a training set, a verification set and a test set according to a preset proportion; Step 6, establishing a deep learning prediction model, taking material characteristic parameters and external load parameters as input, taking container dynamic response indexes as output, and performing model training, verification and test by using the sample set to obtain a deep water explosion container dynamic response prediction model capable of rapidly predicting container dynamic response; and 7, predicting the dynamic response of the deepwater explosion container for the collected experimental data related to the deepwater explosion container by adopting a deepwater explosion container dynamic response prediction model.
2. The method for predicting the dynamic response of the deepwater explosion container according to claim 1, wherein the container structure parameters in the step 1 comprise the length, the width, the height, the wall thickness and the diameter of the container, the data preprocessing comprises the steps of eliminating abnormal values, supplementing missing data and unifying data formats, wherein the abnormal values are identified and eliminated by a box line method, and the missing data is supplemented by a cubic spline interpolation method.
3. The method for predicting the dynamic response of the deepwater explosion container according to claim 2, wherein the finite element analysis software in the step 2 is LS-DYNA, and the specific process for constructing the three-dimensional dynamics simulation model comprises the following steps: The method comprises the steps of establishing a three-dimensional finite element model of a 1/8 container, carrying out mapping grid division, wherein the model comprises a container structure, explosives and a liquid medium, adding material definition and state equation cards for each part of the model through LS-PREPOST software, solving a steel container by adopting a solid Lagrange algorithm, solving the explosives and liquid water by adopting a fluid Euler algorithm, setting a hydrostatic pressure initialization card, constraint conditions and a fluid-solid coupling control card, determining solving step length and total calculating time, generating a K file, importing the K file into an LS-DYNA solver for numerical calculation, outputting a d3plot file, extracting a strain time curve by utilizing LS-PREPOST software, and comparing the strain time curve with measured data to verify model accuracy.
4. The method for predicting dynamic response of a deepwater explosion container according to claim 3, wherein the material characteristic parameters in the step 3 comprise yield strength and elastic modulus, the external load parameters comprise explosive amount and hydrostatic pressure, and the explosive amount and the hydrostatic pressure are uniformly distributed, and the yield strength and the elastic modulus are Gaussian distributed.
5. The method for predicting the dynamic response of a deep water explosion container according to claim 4, wherein the specific process of Latin hypercube sampling in step 3 comprises the following steps: Dividing the value interval of each parameter into N equal probability intervals, randomly selecting a sample point in each interval, randomly scrambling the sample sequence of each parameter dimension, combining the sample sequence into N groups of non-repeated multidimensional samples, and mapping the layered uniform samples of the Gaussian distribution parameters into Gaussian distribution samples through an inverse normal distribution function.
6. The method for predicting dynamic response of deepwater explosion container according to claim 5, wherein step 3 further comprises performing statistical test on the generated sample set, comparing the mean value, standard deviation and probability density function, verifying distribution uniformity and edge distribution rationality of the sample in each parameter dimension, and forming a parameter combination sample set after verification.
7. The method for predicting dynamic response of deepwater explosion container according to claim 6, wherein the automated script in the step 4 is a Python script, and the specific process of the batch simulation calculation flow comprises the following steps: The method comprises the steps of establishing a basic finite element model in an LS-DYNA and storing the basic finite element model as a keyword template file, marking parameters to be replaced in the template file in a placeholder mode, reading parameter combination sample data through a Python script, replacing placeholders in the template file to generate a plurality of groups of input files, automatically calling an LS-DYNA solver to carry out simulation solving and storing a d3plot file, extracting stress, strain and displacement time courses of key measuring points through the Python script, calculating maximum strain, peak stress and peak response time indexes, and finishing the indexes into a structured table.
8. The method for predicting the dynamic response of the deepwater explosion container according to claim 7, wherein the data preprocessing in the step 5 comprises normalization and normalization, the preset proportion is a training set, a verification set and a test set, and the ratio is 8:1:1.
9. The method for predicting the dynamic response of the deepwater explosion container according to claim 8, wherein the deep learning prediction model in the step 6 is a regression prediction model based on a multi-layer perceptron, the model structure comprises an input layer, a plurality of hidden layers and an output layer, the cross entropy loss or the mean square error is used as a loss function, the parameter updating is carried out through an Adam optimization algorithm, and the cross verification method is adopted to carry out optimization on the network layer number, the learning rate and the batch size super parameters.
10. The method for predicting the dynamic response of the deepwater explosion container according to claim 9, wherein the model prediction precision is evaluated by adopting average absolute percentage error and decision coefficients in the step 6, and the model and the weight parameters are saved as standard format files for subsequent real-time prediction after training is completed.

Description

Dynamic response prediction method for deepwater explosion container Technical Field The invention relates to the technical field of deepwater structure safety protection, in particular to a deepwater explosion container dynamic response prediction method. Background With the increasing frequency of deep sea exploration and development activities, the deep water explosion container is used as core equipment for carrying underwater explosion tests, and the structural safety of the deep water explosion container is of great importance. Under the deepwater environment, the explosion load has the characteristics of instantaneity, high strength, nonlinearity and the like, the container can generate complex dynamic response (such as stress strain distribution, displacement change, vibration characteristics and the like), and once the dynamic response exceeds the structural bearing limit, serious safety accidents such as container rupture, medium leakage and the like are caused, so that the accurate prediction of the dynamic response of the deepwater explosion container is a core requirement for guaranteeing the safety of related engineering. At present, prediction methods for dynamic response of deep water explosion containers are mainly divided into two types: The numerical simulation method is that a finite element model, a fluid-structure coupling model and the like of the deepwater explosion container are established, and commercial software such as ANSYS, ABAQUS, LS-DYNA is utilized to simulate the dynamic response process of the container under the action of explosion load. The method can reflect the physical process more comprehensively, but has the problems of long modeling period, high calculation cost (particularly, a large amount of calculation force and time are consumed when aiming at complex structures or multi-task simulation), high requirements on professional skills of operators (skills such as model simplification, grid division, boundary condition setting and the like are required to be mastered), and the like, and is difficult to meet the requirements of rapid prediction and real-time evaluation in engineering. The traditional empirical formula and theoretical analysis method is to fit the empirical formula based on classical mechanics theory or a large amount of test data to estimate the dynamic response of the container. The method has high calculation speed, but has a narrow application range (only aiming at specific structures and specific load conditions), ignores complex factors such as fluid damping, shock wave reflection and the like in a deepwater environment, has low prediction precision, and cannot cope with the dynamic response prediction requirement under complex working conditions. The numerical simulation method has the problems of low calculation efficiency and high cost, and is difficult to realize rapid prediction and multi-working condition analysis in engineering; The traditional empirical formula and theoretical analysis method has low prediction precision and limited applicable scene, and cannot meet the accurate prediction requirement in the complex deepwater explosion environment; The existing method does not effectively integrate the physical integrity of numerical simulation and the high efficiency of machine learning, cannot form a dynamic response prediction scheme with both precision and efficiency, and cannot provide reliable technical support for real-time safety early warning and structure optimization design of the deepwater explosion container. Disclosure of Invention In order to solve the technical problems, the invention provides a method for predicting the dynamic response of a deepwater explosion container. The following technical scheme is adopted: A method for predicting dynamic response of a deepwater explosion container comprises the following steps: Step 1, acquiring relevant experimental data of a deepwater explosion container, wherein the experimental data comprise container structure parameters, explosive loading quantity, environmental hydrostatic pressure, strain measuring point positions and corresponding actually measured strain values, and preprocessing the experimental data; Step 2, based on the preprocessed experimental data, constructing a three-dimensional dynamics simulation model of the deepwater explosion container by adopting finite element analysis software, and verifying the reliability of the model by comparing a numerical calculation result with measured strain data; Step 3, randomly sampling material characteristic parameters and external load parameters which influence the dynamic response of the container by using a Latin hypercube sampling method to generate a plurality of groups of parameter combination samples; Step 4, based on a plurality of groups of parameter combination samples, establishing a batch simulation calculation flow by utilizing an automatic script and finite element analysis software, and acquiring a dynamic r