Heating Line Prediction for Forming of Plate based on AI Machine Learning

Abstract

Despite the recent boom in global shipbuilding orders, Korean shipbuilders continue to face significant challenges stemming from a chronic shortage of skilled labor. In this context, the implementation of smart yards shipyards that integrate advanced automation, robotics, digital technologies, and data-driven production management offers a promising solution. Smart yard technologies not only alleviate workforce constraints but also enhance productivity, production accuracy, and workplace safety by reducing dependence on manual labor and improving process stability. In shipbuilding, hull blocks are broadly categorized according to whether the outer shell plating exhibits curvature. Blocks constructed with flat shell plates are typically assembled in the mid- ship region, where the hull geometry remains relatively straight and uniform. In contrast, curved blocks are located in the bow and stern areas, where the hull form transitions into complex three-dimensional curvature to satisfy hydrodynamic and structural requirements. Most curved plates to fabricate the curved blocks are fabricated through a sequential combination of cold bending to generate the primary curvature, followed by manual flame heating to achieve the final three-dimensional shape. Therefore, in this dissertation, an automated line heating system was established to enhance the productivity of hull curved plate fabrication, and a deformation prediction method was developed for the automatic generation of heating lines. Furthermore, a deep-learning–based model capable of automatically generating heating paths was proposed. In chapter 1, the current status of the Korean shipbuilding industry and the procedures involved in constructing commercial vessels were described, and the forming processes used to fabricate doubly curved hull plates were additionally analyzed. In chapter 2, an automated line heating system based on high frequency induction heating was developed to automate the thermal forming of doubly curved hull plates, and a predictive method for angular distortion was established to enable the automatic generation of heating lines required for the system. The effects of key variables such as plate size, material strength, and heating length were taken into consideration, and the predicted results showed excellent agreement with the experimental measurements. In chapter 3, a deep learning model was proposed to compute the out-of-plane mechanical deformation of a deck plate in the erection stage. For image-to-image translation (three deck conditions to distortion), the AI model was built on a GAN architecture, and CNN-based encoder and decoder were used in the GAN generator and discriminator for feature recognition and reconstruction. It was shown that the proposed model structure is adequate for a precise prediction of the out-of-plane deck plate deformation.[1] In chapter 4, we investigated the feasibility of an AI based methodology for automated heating line generation by assessing whether the deformation shape of a plate can be accurately predicted when key process variables such as namely the heat input condition, plate thickness, and the heating path along which the heat is applied are varied. To do this, the deep learning model developed in Chapter 3 was employed to evaluate the predictive capability under these different parameter combinations. The amplitude and overall shape of the deformation curves exhibit an almost perfect match between the FEM and AI predictions.