Use this URL to cite or link to this record in EThOS:
Title: A decision support system for production ramp-up : a reinforcement learning approach
Author: Doltsinis, Stefanos
Awarding Body: University of Nottingham
Current Institution: University of Nottingham
Date of Award: 2013
Availability of Full Text:
Access from EThOS:
New technologies have been developing rapidly in the last decades. Enterprises require incorporating these technologies in the development of new products. That creates a high pace of new product flow with an increasingly small life cycle. In order to support the fast pace, manufacturing lines have to adapt to the new product requirements as fast as possible. Production ramp-up is a phase in the manufacturing line that has a significant role on the required time to market but currently constitutes a bottleneck in the manufacturing process. Studies have focused on analysing ramp-up and defining its requirements to make the process more efficient. Literature shows that the key for improvement is to enhance the awareness and understanding of human operators and carry out the process more efficiently. Research studies are limited on the analysis without providing solutions on how to improve awareness. This thesis proposes an integrated approach to support decision making during production ramp-up. The work is composed of three main parts. First, a formal model is defined in order to capture the process followed on the shop floor. The model is designed as a Markov Decision Process reflecting the sequence of actions and their effect during the process. The model is composed of and defined through three main elements, namely a state space, a list of actions, and a reward formed as measure of performance during ramp-up. The second contribution of this work delves into the requirements and development of a decision support system for a more efficient ramp-up process. A decision support system is designed to operate complementary to the ramp-up process and support human operators. It captures ramp-up experience in a structured manner through a ramp-up model, processes it through a learning mechanism and communicates the extracted knowledge to human operators. The proposed system operates in two modes and supports the two identified ramp-up cases. Finally, a reinforcement learning algorithm is proposed, to extract the most effective policy for ramp-up and with a limited number of episodes. The algorithm is an outcome of a comparison study between model-based and model free algorithms. The proposed algorithm shows efficiency under the limitations of ramp-up and lack of data.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available