MIMICKING ARTIFICIAL NEURAL NETWORKS IN PREDICTING THE RESIDUAL STRENGTH OF PRE-FATIGUED GFRP LAMINATES THROUGH ACOUSTIC EMISSION ANALYSIS

The acoustic emission activity generated in monotonic four-point bending tests by pre-fatigued glass fibre-reinforced plastic specimens was treated by a conventional interpolation method, with the aim to predict the material residual strength. From the numerical analysis, a three-constant empirical model was derived, by which the residual strength was predicted as a function of the applied stress and the associated cumulative number of acoustic emission events. The results obtained were compared with those discussed in a previous work, where the same scope was pursued by the application of artificial neural networks. The response of the two methods was practically the same, in terms of both mean values and scatter. It was concluded that the classical interpolation may be preferable in the case examined, because the artificial intelligence approach required considerably longer times in the selection of the best network configuration.