Burst Pressure Prediction in Graphite/Epoxy Pressure Vessels Using Neural Networks and Acoustic Emission Amplitude Data

摘要

A burst pressure prediction model was generated from the acoustic emission amplitude distribution data taken during hydroproof for three sets of ASTM standard 145 mm (5.75 in.) diameter filament wound graphite/epoxy bottles. The three sets of bottles featured the same design parameters and were wound from the same graphite fiber, the only difference being in the epoxies used. Therefore, the three resin types were categorized using dummy variables, which allowed the prediction of burst pressures in all three sets of bottles using a single back-propagation neural network. Three bottles from each set were used to train the network. The resin category and the acoustic emission amplitude distribution data taken up to 25 percent of the expect burst pressure were used as network inputs. The actual burst pressures were supplied as target values for the supervised training phase. Architecturally, the network consisted of a 48 neuron input layer (a categorical variable defining the resin type, plus 47 integer variables for the acoustic emission amplitude distribution frequencies), a 15 neural hidden layer for mapping, and a single output neuron for burst pressure prediction. The network, trained on three bottles from each resin type, was able to predict burst pressures in the remaining bottles with a worst case error of -3.89 percent, well within the desired goal of ±5 percent.