Date of Graduation
Statler College of Engineering and Mineral Resources
Chemical and Biomedical Engineering
Mario G. Perhinschi.
The development of an evolutionary algorithm and accompanying software for the generation and optimization of artificial immune system-based failure detectors is presented in this thesis. These detectors use the Artificial Immune System-based negative selection strategy. The utility is a part of an integrated set of methodologies for the detection, identification, and evaluation of a wide variety of aircraft sub-system abnormal conditions. The evolutionary algorithm and accompanying software discussed in this document is concerned with the creation, optimization, and testing of failure detectors based on the negative selection strategy. A preliminary phase consists of processing data from flight tests for "self" definition including normalization, duplicate removal, and clustering. A first phase of the evolutionary algorithm produces, through an iterative process, a set of detectors that do not overlap with the "self" and achieve a prescribed level of coverage of the "non-self." A second phase consists of a classic evolutionary algorithm that attempts to optimize the number of detectors, overlapping between detectors, and coverage of the "non-self" while maintaining no overlapping with the "self." For this second phase, the initial population is composed of sets of detectors, called individuals, obtained in the first phase. Specific genetic operators have been defined to accommodate different detector shapes, such as hyper-rectangles, hyper-spheres, hyper-ellipsoids and hyper-rotational-ellipsoids. The output of this evolutionary algorithm consists of an optimized set of detectors which is intended for later use as a part of a detection, identification, and evaluation scheme for aircraft sub-system failure.;An interactive design environment has been developed in MATLAB that relies on an advanced user-friendly graphical interface and on a substantial library of alternative algorithms to allow maximum flexibility and effectiveness in the design of detector sets for artificial immune system-based abnormal condition detection. This user interface is designed for use with Windows and MATLAB 7.6.0, although measures have been taken to maintain compatibility with MATLAB version 7.0.4 and higher, with limited interface compatibility. This interface may also be used with UNIX versions of MATLAB, version 7.0.4 or higher.;The results obtained show the feasibility of optimizing the various shapes in 2, 3, and 6 dimensions. Hyper-spheres are generally faster than the other three shapes, though they do not necessarily exhibit the best detection results. Hyper-ellipsoids and hyper-rotational-ellipsoids generally show somewhat better detection performance than hyper-spheres, but at a higher calculation cost. Calculation time for optimization of hyper-rectangles seems to be highly susceptible to dimensionality, taking increasingly long in higher dimensions. In addition, hyper-rectangles tend to need a higher number of detectors to achieve adequate coverage of the solution space, though they exhibit very little overlapping among detectors. However, hyper-rectangles are consistently and considerably quicker to calculate detection for than the other shapes, which may make them a promising candidate for online detection schemes.
Davis, Jennifer N., "The design of an evolutionary algorithm for artificial immune system based failure detector generation and optimization" (2010). Graduate Theses, Dissertations, and Problem Reports. 2163.