Failure strength prediction of unidirectional tensile coupons using acoustic emission peak amplitude and energy parameter with artificial neural networks

Abstract Acoustic emission studies have been carried out on 18 numbers of ASTM-3039 unidirectional carbon/epoxy tensile specimens, while loading to failure with a 100 kN Universal Testing Machine. AE response from each of the specimens was filtered and the data acquired up to 50% of actual failure load was only considered for further analysis. Significant AE parameters like peak amplitude, event duration and energy were utilized for analyzing different failure modes in composites viz, matrix crazing, fiber fracture and delamination. The Back propagation neural network structured as 66-45-1 was able to predict the failure load of tensile specimens within 1.22% error tolerance. Only amplitude frequencies and corresponding failure loads of each of the specimens in the training set was taken as input and output vectors of the network, respectively. Another network structured as 66-38-1, along with cumulative energy of each of the amplitude at 1 dB interval as input and the failure strength as output, was able to predict the failure strength of test specimens within 5.75% error tolerances. The prediction accuracy of earlier network was found better, however both the networks were having good correlation in predicting failure strengths.