Artificial Neural Network Prediction of Ultimate S

Authors: T Sasikumar S Rajendraboopathy K M Usha E S Vasudev

Publish Date: 2008/10/30

Volume: 27, Issue: 4, Pages: 127-133

PDF Link

Abstract

Acoustic Emission AE Monitoring was used to evaluate unidirectional carbon epoxy specimens when tensile loaded with a 100 kN Universal Testing Machine A series of eighteen samples were loaded to failure to generate AE data for this analysis After data acquisition AE response from each test was filtered to include only data collected up to 50 of the actual failure load for further analysis Amplitude Duration and Energy are effective parameters utilized to differentiate various failure modes in composites viz matrix crazing fiber cut and delamination with several sub categories such as matrix splitting fiber/matrix debonding fiber pullout etcThe ultimate strength prediction was performed with an Artificial Neural Network Back propagation algorithm Peak Amplitude values varying from 35–100 dB were taken as the input to the network The impact of signal amplitudes due to different failure mechanism to the ultimate strength was mapped using a supervised network having a middle layer with 45 neurons and actual failure loads were supplied as target values during training phase The network finally trained with twelve specimens was able to predict failure loads of remaining six specimens with in the acceptable error tolerance

Keywords: