Dynamic decision making under uncertainty for semiconductor manufacturing and healthcare

Date

2019-05-09

Authors

Gupta, Shreya

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

This dissertation proposes multiple methods to improve processes and make better decisions in manufacturing and healthcare. First, it investigates algorithms for controlling the automated material handling system (AMHS) in a wafer fab. In particular, this research examines algorithms that route vehicles for both the pickup and delivery of lots, with the goal of improving vehicle flow, cycle time, and avoiding congested segments in the AMHS. The proposed methods are simulated using both a stylized simulation model and a more detailed Automod model. These simulations demonstrate that algorithms designed specifically to anticipate congestion can significantly improve some fab metrics. Secondly, this research develops several algorithms for ranking tools in a manufacturing facility so that routes can be categorized and the best routes can be used for recipe probing. Ranking is performed using three different metrics: score-based metrics where higher implies better, target-based metrics where a balance has to be struck by the decision maker between accuracy and precision of a tool based on a target value, and count based metrics such as defect data where a lower number is better (e.g., zero defects is the best scenario). In this part of the dissertation, the ranking algorithms designed for count based metrics are the main contribution to the tool-ranking literature for the manufacturing industry. Finally, the dissertation addresses the problem of medical decision making under uncertainty during the treatment of epilepsy. Here the sequential decision making problem is modeled as an average cost Markov decision process (MDP) to maximize a patient's remaining quality of life. A crucial issue is the uncertainty in transition probabilities extracted from medical studies in epilepsy due to attrition of patients from studies, lack of data and lack of proper experimental design owing to the complexity in treatment procedure. This is addressed by formulating a robust MDP that suggests the best course of treatment for a patient.

Description

LCSH Subject Headings

Citation