Deep Learning Enabling Digital Twin Applications in Production Scheduling:
Case of Flexible Job Shop Manufacturing Environment

This project explores the integration of deep learning (DL) within Digital Twin environments for Flexible Job Shop (FJS) Manufacturing Systems. The goal is to address the computational challenges of running online simulations by using machine learning, particularly DL models, to estimate production metrics efficiently and enable real-time scheduling decisions.

• The project is based on a paper presented and is in the proceedings of the 2023 Winter Simulation Conference (WSC '23).

Key Contributions:

• Deep Learning Meta-Models: DL architectures based on Recurrent Neural Networks (RNNs) and Attention Mechanisms are developed to predict key production metrics, particularly Makespan. These models allow the Digital Twin system to operate efficiently without relying on resource-heavy simulations.

• RNN + Hierarchical Cross-Attention: The combination of RNN and a Hierarchical Cross-Attention architecture helps the models capture sequential dependencies and complex interactions within the production system, making them suitable for dynamic and flexible manufacturing environments.

• Performance: The DL models demonstrated high accuracy, achieving a Mean Absolute Percentage Error (MAPE) of less than 7.4% for Makespan predictions, proving the viability of DL for Digital Twin applications in production scheduling.

Contents

This repository contains the implementation of the DL-based meta-models that utilize example Manufacturing Execution System (MES) data to facilitate online and dynamic scheduling within smart manufacturing systems. The source code and models are available for exploration and further development.

Citation

Ghasemi, A., Yeganeh, Y. T., Matta, A., Kabak, K. E., & Heavey, C. (2023, December). Deep Learning Enabling Digital Twin Applications in Production Scheduling: Case of Flexible Job Shop Manufacturing Environment. In 2023 Winter Simulation Conference (WSC) (pp. 2148-2159). IEEE. https://ieeexplore.ieee.org/abstract/document/10407811

@inproceedings{ghasemi2023deep,
  title={Deep Learning Enabling Digital Twin Applications in Production Scheduling: Case of Flexible Job Shop Manufacturing Environment},
  author={Ghasemi, Amir and Yeganeh, Yavar Taheri and Matta, Andrea and Kabak, Kamil Erkan and Heavey, Cathal},
  booktitle={2023 Winter Simulation Conference (WSC)},
  pages={2148--2159},
  year={2023},
  organization={IEEE}
}

Contact

For inquiries or collaboration, please reach out to yavar.taheri@polimi.it or yavaryeganeh@gmail.com.