A Neural Network for Predicting Ultimate Strengths of Aluminum-Lithium Welds from Acoustic Emission Amplitude Data

Abstract

Acoustic emission (AE) amplitude data have been shown to contain information concerning failure mechanisms and their correlation to ultimate strengths in both metallic and composite materials. As such, AE flaw growth activity was monitored in a set of eleven aluminum-lithium weld specimens from the onset of tensile loading to failure. The amplitude data from the beginning of loading up to 25% of the expected ultimate strength for five of the specimens were used along with the actual measured ultimate strengths to train a backpropagation neural network to predict ultimate strengths. Architecturally, the fully interconnected network consisted of an input layer for the AE amplitude data, two hidden layers for mapping, and an output layer for ultimate strength. The trained network was then applied to the prediction of ultimate strengths in the remaining six specimens where the worst case prediction error was found to be 4.3%.