Composite plate mechanical characterisation through dynamic tests
A reliable and efficient method for characterising the mechanical properties of composite plates has been developed. The effective mechanical properties of the composite plate are estimated from its natural frequencies, which are measured experimentally. Initially, the possibility of applying various combinations of boundary conditions is explored through the use of knife-edge supports in the context of a multi-step characterisation approach. Since the simulation of simple support conditions requires significant time and effort to achieve accuracy and repeatability, the totally free-edge condition is adopted, which is commonly accepted as simple and reliable. The characterisation approach based on free plate vibration experiments is designed for applications involving a large number of specimens with identical geometry. While it is feasible to expand it so that it would be applicable to a more complicated specimen geometry, the method has been demonstrated with plate specimens. The characterisation procedure is based upon iterative minimisation of the sum of the squared differences between the measured natural frequencies and those corresponding to the updated material prediction. Similar approaches reported in the literature use a full-scale dynamic analysis in order to predict frequencies, resulting in a time consuming characterisation process. The application of such an approach to a large number of specimens with a complicated geometry would require finite element analysis, thus lengthening further the characterisation process. This thesis addresses this issue by making use of metamodelling for approximating the relationship between the material properties and the corresponding natural frequencies. This results in a much quicker in-situ characterisation process. The metamodels are based on radial basis functions, with the training data selected using a forward subset selection method. The procedure is applied to characterising four epoxy/glass composite specimens and the results are compared with those from static tests. Inaccuracies in both dynamic and static test measurements as well as in metamodelling are considered as potential sources of error in material properties prediction. A further refinement of the strategy also accounts for mode shape information whereby the error minimisation is carried out over the frequencies belonging to similar modes. The strategy has been demonstrated successfully through simulated data with most of the properties predicted with a high degree of accuracy and in short characterisation time. The discussion on scope for further enhancement of the proposed strategy is primarily focused on possible refinements of metamodelling and on improvements in experimental techniques and apparatus in order to achieve even higher accuracy in frequency predictions and measurements, respectively.