A Modeling and Analysis Framework To Support Monitoring, Assessment, and Control of Manufacturing Systems Using Hybrid Models

Abstract

The manufacturing industry has constantly been challenged to improve productivity, adapt to continuous changes in demand, and reduce cost. The need for a competitive advantage has motivated research for new modeling and control strategies able to support reconfiguration considering the coupling between different aspects of plant floor operations. However, models of manufacturing systems usually capture the process flow and machine capabilities while neglecting the machine dynamics. The disjoint analysis of system-level interactions and machine-level dynamics limits the effectiveness of performance assessment and control strategies.

This dissertation addresses the enhancement of productivity and adaptability of manufacturing systems by monitoring and controlling both the behavior of independent machines and their interactions. A novel control framework is introduced to support performance monitoring and decision making using real-time simulation, anomaly detection, and multi-objective optimization.

The intellectual merit of this dissertation lies in (1) the development a mathematical framework to create hybrid models of both machines and systems capable of running in real-time, (2) the algorithms to improve anomaly detection and diagnosis using context-sensitive adaptive threshold limits combined with context-specific classification models, and (3) the construction of a simulation-based optimization strategy to support decision making considering the inherent trade-offs between productivity, quality, reliability, and energy usage. The result is a framework that transforms the state-of-the-art of manufacturing by enabling real-time performance monitoring, assessment, and control of plant floor operations. The control strategy aims to improve the productivity and sustainability of manufacturing systems using multi-objective optimization. The outcomes of this dissertation were implemented in an experimental testbed. Results demonstrate the potential to support maintenance actions, productivity analysis, and decision making in manufacturing systems. Furthermore, the proposed framework lays the foundation for a seamless integration of real systems and virtual models.

The broader impact of this dissertation is the advancement of manufacturing science that is crucial to support economic growth. The implementation of the framework proposed in this dissertation can result in higher productivity, lower downtime, and energy savings. Although the project focuses on discrete manufacturing with a flow shop configuration, the control framework, modeling strategy, and optimization approach can be translated to job shop configurations or batch processes. Moreover, the algorithms and infrastructure implemented in the testbed at the University of Michigan can be integrated into automation and control products for wide availability.