Citation: Joo, Y.H.; Park, H.; Kim, H.; Choe, R.; Kang, Y.; Jung, J.-Y. Traffic Flow Speed Prediction in Overhead Transport Systems for Semiconductor Fabrication Using Dense-UNet. Processes 2022, 10, 1580. https:// doi.org/10.3390/pr10081580 Academic Editors: Ming-Jong Tsai and Ricky Min-Fan Lee Received: 10 July 2022 Accepted: 9 August 2022 Published: 11 August 2022 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affil- iations. Copyright: © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). processes Article Traffic Flow Speed Prediction in Overhead Transport Systems for Semiconductor Fabrication Using Dense-UNet Young Ha Joo 1 , Hoonseok Park 1 , Haejoong Kim 2 , Ri Choe 3 , Younkook Kang 3 and Jae-Yoon Jung 1, * 1 Department of Big Data Analytics, Kyung Hee University, Yongin-si 17104, Korea 2 Department of Industrial and Management Engineering, Korea National University of Transportation, Chungju 27469, Korea 3 Material Handling Automation Group, Samsung Electronics, Hwaseong-si 18448, Korea * Correspondence: jyjung@khu.ac.kr Abstract: To improve semiconductor productivity, efficient operation of the overhead hoist transport (OHT) system, which is an automatic wafer transfer device in a semiconductor fabrication plant (“fab”), is very important. A large amount of data is being generated in real time on the production line through the recent production plan of a smart factory. This data can be used to increase productivity, which in turn enables companies to increase their production efficiency. In this study, for the efficient operation of the OHT, the problem of OHT congestion prediction in the fab is addressed. In particular, the prediction of the OHT transport time was performed by training the deep convolutional neural network (CNN) using the layout image. The data obtained from the simulation of the fab and the actual logistics schedule data of a Korean semiconductor factory were used. The data obtained for each time unit included statistics on volume and speed. In the experiment, a layout image was created and used based on the statistics. The experiment was conducted using only the layout image without any other feature extraction, and it was shown that congestion prediction in the fab is effective. Keywords: deep convolutional network; UNet; semiconductor; overhead hoist transport; semiconductor fabrication plant 1. Introduction Efficient operation of overhead hoist transport (OHT) systems is important for the productivity of semiconductor processes [1]. In particular, it is important to predict traffic flow and congestion over time because OHT operations, such as dispatching [2–4] and routing [5–8], are highly dependent on traffic conditions. In this study, the p OHT con- gestion prediction issue is addressed based on volume data. In the past, abnormal flow was detected through an agent-based system; however, this approach requires a schema that considers several factors for accurate prediction. Additionally, there is the possibility of a requirement for a new schema if the condition of the target factory or line under consideration changes. The semiconductor process is complex and forms an environment in which hundreds of processes overlap. Efficient handling of the process in such a complex environment is a direct productivity issue. The scheduling method has traditionally been used for efficient deployment of OHT [1,9,10]. Recently, data-driven methods using machine learning and deep learning have been used to improve the efficiency of semiconductor processes [11,12]. Production planning and scheduling issues caused by complex environmental factors are solved by applying machine learning to past production data. Wang dealt with cycle time forecasting (CTF), which is an important issue in production planning [12]. In this study, a method for dealing with big data with parallel computing is introduced, and a deep neural network methodology for CTF is presented. A convolutional neural network (CNN) was trained using a layout image, in which traffic information was input. A study was conducted to predict the OHT congestion in Processes 2022, 10, 1580. https://doi.org/10.3390/pr10081580 https://www.mdpi.com/journal/processes