# A Shop Floor Production Scheduling Case based on RFID-supported Smart Factory

**Authors:** Zhihui Chen, Yize Sun, Yuhao Dong, Zeyu Xiao, Ray Y. Zhong  
**arXiv:** [2608.16626](https://arxiv.org/abs/2608.16626)  
**Date:** 2026-08-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2608.16626v1/

## Key points

- RFID technology is crucial for real-time data collection in smart manufacturing shop floors
- Study conducted on dynamic shop floor scheduling using RFID-collected production data
- Proposed deep reinforcement learning approach for shop floor production scheduling
- Feasibility and practicality of dynamic production scheduling framework demonstrated through simulation studies
- DRL-based framework outperformed existing dispatch methods in minimizing operation makespan

## Summary

, , , , 

In the realm of smart manufacturing, Radio Frequency Identification (RFID) technology plays a crucial role in real-time data collection on manufacturing shop floors. This data is instrumental in supporting dynamic shop floor production planning and scheduling. However, the inherent uncertainty in operation and production processes poses a challenge to achieving optimal utility in scheduling systems. To address this issue, a study was conducted on dynamic shop floor scheduling for a real-life case smart factory equipped with RFID technology. The study involved feasible production sequence mining and real-time processing rate estimation using RFID-collected production data to quantify operation and production uncertainties. A deep reinforcement learning approach based on RFID data analysis was then proposed for shop floor production scheduling. Through simulation studies based on real-life case data, the feasibility and practicality of the dynamic production scheduling framework were demonstrated. Specifically, it was observed that the proposed framework outperformed existing dispatch methods such as first in first out (FIFO), last in first out (LIFO), and deep Q network (DQN) in terms of minimizing operation makespan. The training process of the deep Q network showed convergence after around 140 epochs, with both reward and production makespan reaching stable levels. The DRL-based dynamic shop floor scheduling framework showcased superior performance compared to other deep learning-based and classical scheduling approaches. In conclusion, this paper contributes significantly to smart manufacturing by addressing uncertainties in shop floor scheduling through RFID technology. It explores the feasibility of production sequence mining and delves into the impact of dynamic processing times on production rates. The proposed DRL-based framework presents a promising solution for optimizing shop floor production schedules amidst operational uncertainties, showcasing its potential for enhancing efficiency in manufacturing operations.