Semester

Spring

Date of Graduation

2024

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Chemical and Biomedical Engineering

Committee Chair

Fernando V. Lima

Committee Member

Debangsu Bhattacharyya

Committee Member

Mario Perhinschi

Committee Member

Gerardo Ruiz-Mercado

Committee Member

Lawrence J. Shadle

Committee Member

Stephen Zitney

Abstract

Abstract

Modeling, Control, and Fault Detection of Energy Systems under Limited High-Confidence Data Scenarios

Selorme K. Agbleze

Utilizing process measurements for fault detection is an established approach for processes with adequate datasets. For systems with limited high-confidence data representing fault cases and some amount of low-confidence data, few quantitative hybrid techniques exist for performing fault detection. In real systems, it is time-consuming, expensive, and sometimes not productive to generate enough high-confidence data with fault characteristics of a specific process. The problem of limited high-confidence data scenarios may also arise due to process novelty, the need for new operating conditions, or after process modifications are made. This problem is explored in this dissertation for the case of a developed subcritical coal-fired power plant model and also validated on the Tennessee Eastman process challenge problem already available in the literature.

A dynamic modeling approach for subcritical coal-fired power plant components is presented. The modeling for the power plant includes the fireside, with the effects of fuel, and air combustion, as well as the dynamics of the entire waterside, and power generation units. The main contributions from this modeling task ensure enabling the simulation of custom input signals that include noise, time constants, and delays and the introduction of additional variables in the power generation section. This extends the prediction capability of the power plant model and enables the simulation and analysis of the important short and long-timescale dynamics of subcritical coal-fired power plant components. The control loops for the furnace, drum, superheater, and turbine/generator section are also developed to enable exploration of the closed-loop process dynamics.

The hybrid fault detection framework is developed that enables augmentation of the limited dataset available with HAZOP data allowing the utilization of both human expert knowledge, pseudo-data, and generation of artificial data in adversarial training to reduce false positives. Criteria for selecting pseudo-data and discarding outliers are also proposed. A comparison of the proposed framework and purely supervised and unsupervised methods are also explored for the power plant and Tennessee Eastman examples. The contributions of this dissertation thus enable simulation and control of subcritical coal-fired powerplants or systems with similar process components, in addition to fault detection in limited high-confidence data scenarios.

Download

Share

COinS