Acoustic emission based prediction of local stress exposure

An acoustic emission based method to predict the local stress exposure of a fiber-reinforced material is proposed. This approach uses a step-wise increased load profile as common in testing procedures of fiber-reinforced pressure vessels. An artificial neural network is applied to establish a relationship between acoustic emission criteria evaluated for each load cycle and the globally applied load ratio. The acoustic emission training database for this neural network is obtained from lab-scale experiments, such as tensile tests, bending tests, double-cantilever-beam tests, end-notched-flexure tests and lap-shear tests. The application of this neural network to burst tests of three small-scale pressure vessels (560 mm length) and two large-scale pressure vessels (2600 mm length) for burst pressure prediction is presented. In all cases, the deviation between predicted and measured burst pressure is less than 3.0%, with a maximum prediction uncertainty of 9.8%. In addition, a segmentation technique that allows evaluating the acoustic emission criteria for sub-volumes of the full structure is discussed. This is based on precise source localization of the acoustic emission signals using artificial neural networks. The local prediction of load ratios allows evaluating the stress exposure of the structure as function of the applied load cycles. The resulting concentrations of stress exposure are compared to in situ camera analysis during the tests, strain gage measurements and post mortem analysis of the vessels. In all cases, good agreement of the local load ratio prediction is found.