Multi-job production is a class of manufacturing systems that produce different products within the same production system. These systems are widely used in production assembly, and becoming a trend with smart factories. In this paper, we propose a Deep-Q reinforcement learning driven scheduling algorithm for multi-job production. In particular, we take into account machine breakdown and production plan change as the inputs of the learning process, which are typically considered as unexpected situations in previous studies. We validate the proposed scheme with real data collected for 6 months between May and October 2019 from a tier-1 vendor of a world top-4 motor company. Our case study shows that the proposed scheme improves the throughput of the production line by 37% compared to the conventional rigid method.
Table Of Contents
Abstract List of Figures List of Tables Notation, Symbols, and Acronyms 1 Introduction 1 2 Related Work 3 3 Target Manufacturing System 5 3.1 Data Processing 5 3.1.1 Work Time 8 3.1.2 Machine Parameters 8 3.1.3 Buffers 10 3.2 Factory Description 11 3.3 Distinctive Feature of the Line 12 3.4 Scheduling Problem 12 4 Deep-Q Scheduling 14 4.1 Agent and Environment of RL 14 4.2 States, Actions, and Rewards 15 4.3 Training Method 17 5 Experimental Results 20 5.1 Data Sets and Training Details 20 5.2 Performance Evaluation 21 6 Conclusions 25 Bibliography 26 국문초록 29
