Duyar, Ahmet

This thesis presents the design of a fault detection and diagnosis system for the T700 helicopter engine. Fault refers to a malfunction that degrades the performance of a system. Fault parameter estimation approach has been followed to accomplish this task along with the hypothesis testing for isolation purposes. Fault detection and diagnosis modules have been written in FORTRAN and run on-line with the engine simulation. Fault detection and diagnosis system has been tested for sensor bias, sensor multiplicative fault and bias in the actuation signal. The results of these experiments are displayed at the end.

Accurate representation of the dynamic behaviour of the Space Shuttle Main Engine (SSME) is required for various diagnostic, design and evaluation purposes. A complete nonlinear dynamic simulation of the SSME developed by the Rocketdyne Division of the Rockwell International Corporation is difficult to use due to its size and complexity, hence the need to obtain a simplified model of the SSME to generate data describing the normal modes of operation in real time is relevant and also the objective of this thesis. The process involved in obtaining such a model are: (1) To determine the point models at various operating points and (2) To obtain a comprehensive piece-wise linear model by regressing the parameters of the point models with variables describing the operating points. The comprehensive model obtained by using such a scheme is presented along with the point models. The nonlinearities inherent in the engine's operation are modelled as separate blocks and are added to linearized model. Finally the validity of the linked model is demonstrated with suitable test signals.

In this thesis, a comparative evaluation of some residue generation methods used for model based failure detection techniques is made. Residues are differences between the actual plant outputs and the outputs of a mathematical model describing the plant. Under a no-fail condition and in the absence of noise, the residues are zero. When a failure occurs, the residues will be non zero and the nature of the residues depend on the residue generation method and the system input. The resulting residues are statistically analyzed to isolate the nature of the failure(s). Three methods for residue generation namely, parity relations, parameter estimation and estimation by the Kalman filter are discussed. A mathematical analysis for the basis of choosing a statistical test which depends on the system input is also presented.

An identification scheme which can be used for discrete time multi-input multi-output time invariant systems is presented. The identification scheme involves two steps; (1) The identification of a set of invariant indices (Structure identification) and (2) The estimation of the parameters of the system (Parametric identification). The technique utilizes a canonical representation of a system which is based on the notion of output injection. This canonical form is dependent on a chosen real number alpha and is therefore called the alpha-canonical form. Least square estimation technique is used for parameter estimation. The off-line version of this identification scheme is presented here. This scheme is then used to generate a linear model of the Space Shuttle Main Engine at the operating point corresponding to the 100% power level from the nonlinear dynamic engine simulation.

This study investigates the technical feasibility of solar
pond power generation in the state of Florida. A onedimensional
model is used to analyze three different modes
of operation of the pond as follows: The steady state
analysis determines the maximum overall efficiency of the
pond, optimum load and optimum brine temperature to be
extracted. The steady periodic analysis determines the
performances of the pond subjected to seasonal variation.
The transient analysis determines the time necessary for the
pond to reach the optimum temperature.
The study determines that for the Miami, Orlando, and
Jacksonville locations, for a pond with 1.5 m NCZ depth
using data for the state of Florida, the maximum overall
efficiency is between 3.30 and 3.32 percent with an optimum
load between 33 and 35 W/m2 and brine temperature to be
extracted between 85 to 92°C. The above results assume that
the power plant efficiency would be equal to that of a
Carnot cycle efficiency. Since, in practice Carnot
efficiency cannot be achieved the actual efficiency will be
lower than the above values.

Until recently, control design techniques for multivariable
systems, such as pole placement or optimal control design,
have been either too complex to be usable, or yielded
designs which were liable to instability if the parameters
of the plant varied from those used in the design. A
technique which uses the singular values of the system
transfer function matrix is now available. This technique
yields control designs which are guaranteed to be robust
with respect to plant parameter variations. This technique,
combined with a novel technique for shaping the frequency
responses of the singular values is used to design a control
system for a gas turbine jet engine. It is shown that
adjusting the crossover frequency of the open loop singular
values affects the closed loop time and frequency response
in the same manner that adjusting the open loop gain affects
the response of a single-input/single-output control system.

A two-dimensional mathematical model is developed to analyze the
thermal behavior of salt gradient solar ponds. This model can be
used to evaluate the suitability of solar ponds for space heating
and power generation as well as the thermal storage capability of
the ground. The solar pond model adopted consists of an upper
nonconvective zone and a lower convective zone. This model
incorporates detailed representation of surface and ground heat
losses. Solutions to the energy equation are obtained for both the
pond and the ground. The energy equation for the lower convective
zone is used to determine the energy that can be extracted from the
pond.

System identification methods are frequently used to obtain appropriate models for the purpose of control, fault detection, pattern recognition, prediction, adaptive filtering and other purposes. A number of techniques exist for the identification of linear systems. However, real-world and complex systems are often nonlinear and there exists no generic methodology for the identification of nonlinear systems with unknown structure. A recent approach makes use of highly interconnected networks of simple processing elements, which can be programmed to approximate nonlinear functions to identify nonlinear dynamic systems. This thesis takes a detailed look at identification of nonlinear systems with neural networks. Important questions in the application of neural networks for nonlinear systems are identified; concerning the excitation properties of input signals, selection of an appropriate neural network structure, estimation of the neural network weights, and the validation of the identified model. These questions are subsequently answered. This investigation leads to a systematic procedure for identification using neural networks and this procedure is clearly illustrated by modeling a complex nonlinear system; the components of the space shuttle main engine. Additionally, the neural network weights are determined by using a general purpose optimization technique known as evolutionary programming which is based on the concept of simulated evolution. The evolutionary programming algorithm is modified to include self-adapting step sizes. The effectiveness of the evolutionary programming algorithm as a general purpose optimization algorithm is illustrated on a test suite of problems including function optimization, neural network weight optimization, optimal control system synthesis and reinforcement learning control.