Date of Completion


Embargo Period



Optimal Experimental Design, Active Fault Diagnosis, Fault Identification, Classification, Distance Measures

Major Advisor

George Bollas

Associate Advisor

Matthew Stuber

Associate Advisor

Ranjan Srivastava

Associate Advisor

Douglas Cooper

Associate Advisor

Peter Luh

Field of Study

Chemical Engineering


Doctor of Philosophy

Open Access

Open Access


Fault detection and isolation (FDI) is crucial to identifying problems that can occur in complex systems. Many industrial systems require the implementation of model-based FDI to reduce the risk of system failure and increase good performance while minimizing the use of additional hardware. Active FDI has become increasingly popular as it uses auxiliary tests to reduce the impact of uncertainty and improve fault diagnosis. A methodical approach is needed to consistently generate feasible test designs for the optimal detectability and isolability of faults in targeted systems. The focus of this dissertation is to design a general framework that improves the identifiability of faults based on methods of active fault diagnosis and optimal sensor selection and placement. The proposed standard work and corresponding methods treat fault detection and isolation as a series of constrained optimization problems. The issues of uncertainty caused by system operations and modeling error are addressed through the formulation of mathematical problems to minimize their impact on fault detection and isolation procedures. This method is based on optimal experimental design (OED) techniques that reduce correlations between targeted model parameters.

The objective was to select optimal test designs that improve the detection and isolation of faults by maximizing available information with respect to faults, with the information represented as sensitivities of selected system outputs. FDI was treated in this dissertation as a series of constrained optimization problems. After the optimal test design was determined, the system of interest was evaluated at the optimal operating conditions to assess the success rate of the proposed fault diagnosis. Simultaneous test design and sensor selection is presented as a mixed-integer nonlinear optimization problem, and novel approaches were implemented to select the most effective FDI test from available sensors and inputs. The methodology presented in this dissertation is intended to refine system design and maintenance schedules to mitigate costs associated with anticipated faults. The resulting FDI test designs are shown to consistently reduce false alarms and nondetections with implementation. Verification of the improved fault diagnosis through the proposed techniques was done with benchmark virtual systems of various levels of scale and complexity.