Journal
IEEE TRANSACTIONS ON HUMAN-MACHINE SYSTEMS
Volume 48, Issue 3, Pages 229-240Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/THMS.2017.2717885
Keywords
Decision support; machine learning; ramp-up; symbiotic human-machine systems
Funding
- European Commission, the FP7 NMP FRAME project [CP-FP 229208-2]
- EPSRC Centre for Innovated Manufacturing in Intelligent Automation [EP/IO33467/1]
Ask authors/readers for more resources
Constantly shorter product lifecycles and the high number of product variants necessitate frequent production system reconfigurations and changeovers. Shortening ramp-up and changeover times is essential to achieve the agility required to respond to these challenges. This work investigates a symbiotic human-machine environment, which combines a formal framework for capturing structured ramp-up experiences from expert production engineers with a reinforcement learning method to formulate effective ramp-up policies. Such learned policies have been shown to reduce unnecessary iterations in human decision-making processes by suggesting the most appropriate actions for different ramp-up states. One of the key challenges for machine learning-based methods, particularly for episodic problems with complex state-spaces, such as ramp-up, is the exploration strategy that can maximize the information gain while minimizing the number of exploration steps required to find good policies. This paper proposes an exploration strategy for reinforcement learning, guided by a human expert. The proposed approach combines human intelligence with machine's capability for processing data quickly, accurately, and reliably. The efficiency of the proposed human exploration guided machine learning strategy is assessed by comparing it with three machine-based exploration strategies. To test and compare the four strategies, a ramp-up emulator was built, based on system experimentation and user experience. The results of the experiments show that human-guided exploration can achieve close to optimal behavior, with far less data than what is needed for traditional machine-based strategies.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available