CoA. PhD, EngD, MPhil & MSc by research theses
Permanent URI for this collection
Browse
Browsing CoA. PhD, EngD, MPhil & MSc by research theses by Course name "MPhil"
Now showing 1 - 4 of 4
Results Per Page
Sort Options
Item Open Access The application of neural networks to spacecraft control(1994-08) Cooper, A.; Lewis, D. J. G.This thesis investigates how two neural network-based control techniques can be applied to a specific spacecraft control problem. The neural networks used are simple backpropagation networks, consisting of one or more tansigmoidal neurons (neurons with tanh transfer functions) in a hidden layer, and a linear neuron in the output layer. The neural network control techniques investigated here are Direct Model Inversion and Indirect Model Inversion. The spacecraft control problem is that of reducing the vibrations of a spacecraft payload. The source of the vibrations is a mass imbalance in one of the reaction wheels of the spacecraft. Four components are represented in the spacecraft model. These are rigid body inertia, solar array flexure, fuel slosh and payload vibration. A simple sinusoidal signal is used to model the disturbance torque produced by the reaction wheel mass imbalance. The complete model is broadly based on the Solar Heliospheric Observatory (SOHO) that is due for launch in 1995. Each of the neural network control techniques used is shown to be successful in reducing the effects of the disturbance torques on the spacecraft payload. However, in each case, a simple positional feedback gain term provides more effective and reliable control.Item Open Access Application of nonlinear control theory in weapon guidance and control(2001-03) Brundle, D.; Thomasson, P. G.This thesis considers the application of nonlinear control theory in two subjects pertinent to weapon applications. Initially, Section 2 considers the development of a simple nonlinear autopilot for a Laser Guided Bomb (LGB). Later a nonlinear autopilot design is developed using a Pulse-Width Modulated (PWM) controller derived from the method developed by Bemelli-Zazzera et al4. This is applied to an LGB utilising a “bang-bang” actuator, enabling the control surfaces to achieve a pseudo-proportional response. The PWM design stems from an equivalent Pulse Amplitude Modulated controller, which required a design technique to be developed for a linear autopilot and, in addition, simulation of an electro-mechanical actuator. Simulation demonstrated that the PWM controller can achieve the desired response but the design must incorporate actuator dynamics. Section 3 considers the use of nonlinear control theory to examine the nonlinear intercept equations using a Proportional Navigation (PN) guidance law. Using a simple heuristic example, PN is introduced and vector algebra used to develop a simple model of the intercept. The model is then used to illustrate the importance of the kinematic gain. Using the method pioneered by Ha et al16, Lyapunov theory is used to demonstrate that PN is a robust guidance law. Although generally derived assuming the target maintains rectilinear flight, Lyapunov theory is used to demonstrate interception is always possible provided the pursuer has sufficient manoeuvre advantage over the target. Noting that many missiles incorporate a 1 directional warhead, Lyapunov theory is used to design a time-varying rate bias that controls the direction of approach to the target. Simulation demonstrates that the guidance requirements are indeed achieved by this law but additional effort is required by the control system. In Section 3 it is demonstrated that the PN guidance law will always ensure an intercept, i.e. it does not by itself generate miss-distance. In the final part of Section 3, using adjoint software designed by Zarchan42, it is demonstrated that miss-distance develops in practical systems as the result of sub-system dynamicsItem Open Access Modelling naturalistic decision making using neural networks(2000-01) Duggan, S.; Harris, DonThis thesis describes two studies conducted within a naturalistic decision making paradigm. Study One examines the choice of university for master level education. This decision is presented as a consequential choice decision task. Students, who had been offered placements at Cranfield University for the 1998/99 term, participated in this research. Factors influencing the participant’s decision to attend or not to attend Cranfield were collected with a questionnaire specifically designed for this purpose. The final data set contained 267 questionnaires. Study two describes a decision where a disruptive passenger threatens a hypothetical flight. Sixty-five professional members of flight crew participated in a series of semi-structured telephone interviews during which they described their decision-making process for dealing with this situation. This decision process is presented as a pattern-matching task. Artificial neural networks were used to model the decision on the basis of the input variables (questionnaire items in study one and interview variables in study two) undertaken to produce an empirically verifiable model of the participants decision making process. Cross-validation of the results showed that decision outcomes could be predicted on the basis of the models. The cross-validation results, in terms of classifications are compared with discriminant function analysis classification results, to determine if neural networks or discriminant function analysis is a more appropriate form of analysis for modelling a naturalistic decision. Both studies show that neural networks outperformed the discriminant function analysis results in terms of classification. Press’s Q analyses also support this finding. It is suggested that neural networks may be a viable way of modelling naturalistic decisions.Item Open Access Structural design for passenger safety(1996-05) Roots, Mark R.; Brown, J. C.This thesis covers the design and analysis of a roll cage structure for use on a sports racing car. The method used to design and verify the roll structure was novel as small automotive companies tend to use evolution as a design tool. Evolutionary design works well for certain problems, however, is not well suited to major structural modifications. The method used in this report integrates the existing structure with the roll cage to improve the torsion stiffness and hence the handling of the vehicle. Careful integration of the roll cage with the rest of the chassis enabled the torsion stiffness to be increased by over 400 %. In addition the weight efficiency of the final chassis was increased over that of the original chassis by over 200 % . The investigation of the torsion stiffness was carried out using linear finite element analysis using the NASTRAN suite of programs. The second stage of the investigation was to develop this design into a crashworthy roll cage. The resulting model and design are presented in this report. The design of the crashworthy roll cage was carried out using non linear finite element analysis with N AS TRAN. The N AS TRAN results were then verified with a full structural test on the chassis. The results of the tests are presented and compared to the NASTRAN analysis results. Good correlation was achieved and the method showed promise for applications in the small automotive industries. The use of finite element analysis for the design of an integrated structure represented a novel application of a well established technique to an industry where experience is the main design tool. The results of the investigation were encouraging and a close correlation was achieved between the analysis and test results. Finite element analysis represents a relatively cheap and quick method of investigating the effect of structural changes. This method could be used for the design and development of new structures and would give a good indication of the effect of these changes. Small automotive companies, such as TVR, should find the technique particularly useful for both the design of new structures and for the modification of those already in use.