Research On The Recommended Method Of Ship Component Adaptation Parameters

In the design stage of ship models,the quality and efficiency of modeling mainly depend on the performance of industrial modeling software.Therefore,the development of modern shipbuilding industry puts forward higher requirements for ship modeling software.In this paper,aiming at the problem of constantly updating the preset files during the development and maintenance of the ship modeling software and repeatedly specifying the number of parameters during the creation of the ship 3D model,the deep learning method is introduced into the ship modeling software.This paper has carried out the research of parameter recommendation method,and realized the recommendation of model parameters according to the design rules.The main tasks completed are as follows:Firstly,the characteristics of ship structural plate members are analyzed.Taking end cut as an example,according to different values of cutting object,connection method,boundary type,flange direction,inclination angle,etc.,the design rules for applying end cut functions to components are studied.By analogy,the design rules for other functions of the component such as board interleaving and collar creation are formed.Secondly,according to the design rules,a data set that can support the training and verification of the learning algorithm is created.A one-dimensional convolutional neural network algorithm model is established,and the network model is optimized through factors such as learning rate,activation function,and total number of training periods.Compared with the fully connected deep neural network algorithm model,the algorithm model based on one-dimensional convolutional neural network is selected to realize the parameter recommendation of the end cut function.Then,based on the transfer learning algorithm model,according to the isomorphic transfer learning method,the top-level unit of the one-dimensional convolutional neural network model is modified,and the weight and bias are fine-tuned.The pre-trained network model is extended to the recommendation process of other types of component parameters,and the parameter recommendation of the flange type is realized.Finally,a ship component adaptation parameter recommendation module is built,and the role of the configuration file in the modeling process is replaced by a deep learning algorithm to realize the adaptation recommendation of the parameter recommendation module in the modeling software.