Date of Graduation

2017

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Mechanical and Aerospace Engineering

Committee Chair

Mario G Perhinschi

Committee Co-Chair

Debangsu Bhattacharyya

Committee Member

Fernando V Lima

Committee Member

Hever Moncayo

Committee Member

Victor H Mucino

Abstract

This dissertation presents the design, development, and simulation testing of a monitoring and control framework for dynamic systems using artificial intelligence techniques. A comprehensive monitoring and control system capable of detecting, identifying, evaluating, and accommodating various subsystem failures and upset conditions is presented. The system is developed by synergistically merging concepts inspired from the biological immune system with evolutionary optimization algorithms and adaptive control techniques.;The proposed methodology provides the tools for addressing the complexity and multi-dimensionality of the modern power plants in a comprehensive and integrated manner that classical approaches cannot achieve. Current approaches typically address abnormal condition (AC) detection of isolated subsystems of low complexity, affected by specific AC involving few features with limited identification capability. They do not attempt AC evaluation and mostly rely on control system robustness for accommodation. Addressing the problem of power plant monitoring and control under AC at this level of completeness has not yet been attempted.;Within the proposed framework, a novel algorithm, namely the partition of the universe, was developed for building the artificial immune system self. As compared to the clustering approach, the proposed approach is less computationally intensive and facilitates the use of full-dimensional self for system AC detection, identification, and evaluation. The approach is implemented in conjunction with a modified and improved dendritic cell algorithm. It allows for identifying the failed subsystems without previous training and is extended to address the AC evaluation using a novel approach.;The adaptive control laws are designed to augment the performance and robustness of baseline control laws under normal and abnormal operating conditions. Artificial neural network-based and artificial immune system-based approaches are developed and investigated for an advanced power plant through numerical simulation.;This dissertation also presents the development of an interactive computational environment for the optimization of power plant control system using evolutionary techniques with immunity-inspired enhancements. Several algorithms mimicking mechanisms of the immune system of superior organisms, such as cloning, affinity-based selection, seeding, and vaccination are used. These algorithms are expected to enhance the computational effectiveness, improve convergence, and be more efficient in handling multiple local extrema, through an adequate balance between exploration and exploitation.;The monitoring and control framework formulated in this dissertation applies to a wide range of technical problems. The proposed methodology is demonstrated with promising results using a high validity DynsimRTM model of the acid gas removal unit that is part of the integrated gasification combined cycle power plant available at West Virginia University AVESTAR Center. The obtained results show that the proposed system is an efficient and valuable technique to be applied to a real world application. The implementation of this methodology can potentially have significant impacts on the operational safety of many complex systems.

Share

COinS