Research On Deep-sea Spherical Structure Design Based On Machine Learning

In recent years,with the deployment of a strategy of powerful marine country,the development of the deep sea has received more and more attention,and the exploration of the deep sea has also promoted the development of marine scientific and technological equipment.Due to the special high-pressure seawater environment of the deep sea,humans can only explore the deep sea through deep-sea submersibles.The deep-sea manned submersible is the only tool for human beings to reach the deep sea.The maximum pressure load and structural safety it can withstand in the deep sea have become the key assessment indicators for the design of the submersible.Deep-sea submersibles rely on pressure-resistant structures to withstand the huge pressure loads generated by deep-sea seawater.Therefore,the strength and reliability of the pressure-resistant structures of deep-sea manned submersibles are particularly important.Due to the high cost of the deep-sea submersible and its pressure-resistant structure model pressure test,the experimental test data of the pressure-resistant structure is very limited,which is very unfavorable for the research on the safety and reliability of the pressure-resistant structure.Therefore,this paper takes the safety and reliability of the spherical pressure-resistant structure of the deep-sea manned submersible as the research goal,and builds a method for optimizing neural network training and cross-validation based on the reliability index.Based on the verification of beam and frame model,the ultimate bearing pressure and structural parameter design of spherical shell structure are studied in this paper.The main contents of the thesis are as follows:(1)This paper summarizes the problems of large amount of input data,complex calculation process and high nonlinearity of limit state function of general structural reliability analysis method through investigation and summary.The training function based on BP neural network model replaces the explicit limit state equation,to achieve the function of predicting the limit state function,thus constructing a reliability analysis method based on the BP neural network method,laying the foundation for establishing the limit state implicit function rule of the deep-sea spherical shell structure,and achieving the goal of building a machine learning optimization project based on the neural network algorithm in this paper.The purpose of structural model calculation and forecasting methods.(2)Aiming at the defects of traditional neural network models such as many uncertain parameters and complex and variable parameter data,this paper uses the cross-validation method to optimize the neural network training process,and seeks scientific and accurate neural network models,reasonable network structures,and fast and effective neural network models.The network parameter training algorithm is used to implicitly simulate the limit state function function.Starting from the mean square error of the iterative process of improving the neural network algorithm,the algorithm optimization of the complex multi-parameter network structure is realized and the prediction accuracy is improved.(3)Based on the basic theory of the structural reliability calculation method,this paper aims to calculate the data reliability index by the Monte Carlo method.By combining the methods of improving the BP neural network method and the machine analysis and cross-training data source method,a coupling improvement method is constructed.In this paper,the structural reliability calculation method of BP neural network and Monte Carlo method is constructed.Finally,through the example of engineering structure model such as typical beam model and frame structure,this paper calculate the reliability and compare the results of the literature method to verify the accuracy of the prediction of the improved neural network model,which provides a method basis for solving the structural reliability index under the limit state of the deep-sea spherical shell model.(4)Based on the reliability calculation method of the improved BP neural network and the Monte Carlo method,a two-parameter limit state prediction method suitable for deep-sea spherical structures is constructed in this paper,and a machine learning method with experimental data as the characteristic parameter index is established.The method takes the critical load parameter and the aspect ratio as the input variable parameter data,and takes the critical buckling pressure as the output target variable parameter.This paper introduces the improved BP neural network model and the cross-training method,and calculates the data reliability by the precise Monte Carlo method,so as to study its convergence to predict the size parameters of the deep-sea spherical shell structure.Finally,the optimization program platform is built on the MATLAB platform,and the calculation accuracy of the method in this paper is verified by the example of the deep-sea spherical shell structure model.(5)Through the two-parameter limit state estimation method of deep-sea spherical structure,this paper optimizes the design of the spherical structure parameters of the deep-sea deep manned cabin based on the limited test parameters and the method of machine learning automatic design parameters.In this paper,the design of 10000 meter spherical structure model is successfully realized through machine learning,the pressure test model is designed,and the model test is completed.The comparative test results show that the machine learning based on neural network and reliability analysis can optimize the design method of spherical shell model.Therefore,the research method in this paper can provide a theoretical reference for the future pressure-resistant spherical structure design and structural safety and reliability analysis of deep-sea submersibles.