Use this URL to cite or link to this record in EThOS:
Title: Failure detection of composite joints using data mining
Author: Kia, Shima Shirazi
ISNI:       0000 0001 3598 9072
Awarding Body: University of the West of England, Bristol
Current Institution: University of the West of England, Bristol
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
In structural applications such as in aircraft and spacecraft, composite components are often fastened to other structural members (composite or metals) by bolted joints. In bolted composite structures, stress concentration is developed around the holes and severely reduces the strength of the structure. Uncertainties regarding the strength and failure of critical joints in composite lami!1ates may, on the one hand, lead to conservatively designed joints resulting in significant weight penalties, while, on the other hand, lead to non-conservatively designed joints resulting in catastrophic inservice failure ofthe structure. It is therefore important that the strength and failure of joints in composite laminates be fully understood and that appropriate methods for stress and failure analysis be developed. Current approaches for solving this problem are analytical and numerical methods which are evaluated by experimental tests. In these approaches, the structure is simulated regarding the parameters influencing its behaviour. However, due to the enormous range of influencing factors, no robust formulation has been obtained for this problem. ' In the absence of any comprehensive formulation in predicting the failure of composite pin joints, in this study, a novel application of Computational Intelligence (CI) methods is introduced. One of the main advantages of this approach is its capability of indirectly counting for all influencing factors. The objective of this research is to develop a model (or formula) for predicting the behaviour of pin joints. More precisely, the main question this study answers is: 'Can the behaviour of a pin joint can be modelled, using CI methods, so that its behaviour could be predicted when the design parameters change?' The novel application of CI in predicting the behaviour of pin joints has been implemented on two types of materials, aluminium joints and composite joints, in order to prove its robustness (generality). The problem is categorized as a classification problem and two different methods of classification, decision tree and adaptive neuro-fuzzy system, are applied. In order to improve the accuracy, the hybridization of these two methods is also considered. The analysis presented in this work shows that these techniques can help to solve such an engineering problem through data generated from a limited number of experimental/ analytical tests. The key contributions ofthis research are: 1. Introduction and a thorough examination of the applicability of computational intelligence methods in predicting the behaviour of lightweight (composite and aluminium) pin joints are performed. 2. A set of key parameters is established that define the design space for a generalised lightweight pin joint (and it could be expanded for other structures, as well). Furthermore, based on CI methods, the relationship between the key parameters ofdesign space is developed. 3. The behaviour of lightweight pin joints with variable edge distances are predicted by means of three different types of predictive models. The results are assessed against experimental data. 4. To generalise the application of CI methods, pin joints with variable widths are also modelled using three predictive models. The results are validated against FEA data. 5. The study confirms that based on CI methods, the elastic behaviour of pin joints can be predicted efficiently, using appropriate methods. Specifically, it indicates that the performance of decision tree models in evaluating the potential behaviour of composite joints is superior compared to other methods. However, for aluminium joints, an adaptive neuro-fuzzy system proves to be more efficientand accurate. This is due to the fact that aluminium yields while for composites the material either resists or fails. Therefore, it can be concluded that the material's behaviours should be considered when selecting the data mining modelling techniques. Generally, the performance of adaptive neuron-fuzzy models initiated by C4.5 decision tree algorithms is the most reliable technique for both types ofjoints. 6. Based on CI methods, the possibility of modelling the combinatorial effects of two (or more) different design parameters and their influence on failure of the structure is also confirmed.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available