Material flow simulation is a powerful tool to realize efficient operation in complicated production systems such as high-mix and low-volume production. However, it takes great efforts and expertise to construct accurate simulation models. On the other hand, in recent years, IoT and machine learning techniques that collect and utilize field data are advancing rapidly. In this research, we propose a data-driven and multi-scale modeling approach which constructs accurate simulation models semi-automatically. The proposed approach aims to optimize the configuration of simulation model by combining deductive models such as queue model and inductive model such as machine learning model to maximize accuracy. In this article, we introduce the concept of the proposed method and experimental results on a simple production system.