Author : Alexandra Aulova
Publisher :
Page : 125 pages
File Size : 10,74 MB
Release : 2016
Category :
ISBN :
Polymeric materials, which are typical representatives of dissipative systems, are entering demanding engineering applications. Therefore, a precise and reliable monitoring of durability and health of polymeric structures is required. Such health monitoring system should be able to detect geometrical changes, as the existing systems do and detect changes in time-dependent mechanical properties caused by the external static and dynamic loading and environmental conditions. The thesis addresses a problem of obtaining segments of relaxation modulus curves from the experimental data, obtained in uniaxial constant strain rate experiments, which represents an inverse problem. The inverse problem was solved with the Multilayer Perceptron and the Radial Basis Function neural networks. In addition, the problem of direct determination of relaxation mechanical spectrum from the same experimental data was addressed. In this case the so-called pre-structured neural networks were applied. The Radial Basis Function networks demonstrated promising generalization and robustness abilities compared to the commonly used "classical" numerical method of exponential fitting. Hence, this approach may be utilized for development of an on-line monitoring systems of polymeric structures exposed to complex mechanical loadings. On the other hand, the proposed pre-structured neural networks approach requires further investigation and improvements.