Development and Applications of Self-learning Simulation in Finite Element Analysis
Date: 28 May 2019
University of Exeter
Doctor of Philosophy in Engineering
Numerical analysis such as the finite element analysis (FEA) have been widely used to solve many engineering problems. Constitutive modelling is an important component of any numerical analysis and is used to describe the material behaviour. The accuracy and reliability of numerical analysis is greatly reliant on the constitutive model ...
Numerical analysis such as the finite element analysis (FEA) have been widely used to solve many engineering problems. Constitutive modelling is an important component of any numerical analysis and is used to describe the material behaviour. The accuracy and reliability of numerical analysis is greatly reliant on the constitutive model that is integrated in the finite element code. In recent years, data mining techniques such as artificial neural network (ANN), genetic programming (GP) and evolutionary polynomial regression (EPR) have been employed as alternative approach to the conventional constitutive modelling. In particular, EPR offers great advantages over other data mining techniques. However, these techniques require a large database to learn and extract the material behaviour. On the other hand, the link between laboratory or field tests and numerical analysis is still weak and more investigation is needed to improve the way that they matched each other. Training a data mining technique within the self-learning simulation framework is currently considered as one of the solutions that can be utilised to accurately represent the actual material behaviour. In this thesis an EPR based machine learning technique is utilised in the heart of the self-learning framework with an automation process which is coded in MATLAB environment. The methodology is applied to simulate different material behaviour in a number of structural and geotechnical applications. Two training strategies are used to train the EPR in the developed framework, total stress-strain and incremental stress-strain strategies. The results show that integrating EPR based models in the framework allows to learn the material response during the self-learning process and provide accurate predictions to the actual behaviour. Moreover, for the first time, the behaviour of a complex material, frozen soil, is modelled based on the EPR approach. The results of the EPR model predictions are compared with the actual data and it is shown that the proposed model can capture and reproduce the behaviour of the frozen soil with a very high accuracy. The developed EPR based self-learning methodology presents a unified approach to material modelling that can also help the user to gain a deeper insight into the behaviour of the materials. The methodology is generic and can be extended to modelling different engineering materials.
Item views 0
Full item downloads 0