조선 네스팅 문제의 부재 페어링을 위한 딥러닝 기반 부재 분류 방법

Abstract

With the rapid development of artificial intelligence technology, deep learning-based classification techniques have had enough reliability to be applied to industrial sites. However, while the study of the object classification on data acquired with 3D scanners or cameras has made remarkable progress, research activity based on geometric data sets is still in its infancy. In particular, in order to improve the classification performance of ship parts based on deep learning in the nesting problem to increase productivity in shipbuilding, the study of the construction of part datasets and data pre-processing is necessary. In this paper, we introduce a method to apply the artificial neural network technology of deep learning to the nesting algorithm for shipbuilding. Labeled with histogram-based shape contexts for constructing a dataset for classifying ship parts using Convolutional Neural Networks (CNNs). In addition, we introduce the preprocessing method of the geometric information of the ship parts for learning and the no-fit polygon (NFP) method for classified parts to pair up. To train the classification model for the 23,201 ship parts, a data set of 842 classes was constructed through the shape matching algorithm. The trained CNN model was able to classify those parts with an accuracy of 85.13%.