Estimation of damage process using classification of AE waveforms from a pressurized plate type heat exchanger

Abstract

The final goal of this study is to assess integrity of a corrugated fin type heat exchanger under pressure using acoustic emission (AE) technique. In this study, AE signals from the heat exchanger sample under pressurizing test to burst was monitored. As two type fractures, 1) inclusion fracture in an Al alloy and 2) fracture at fin and fillet of blazing part, are expected to occur in the sample, we tried to classify AE into the two fracture types with principal component analysis and convolutional neural network. The PCA was done with seven traditional AE features representing AE waveforms. The first and second components are mainly related to amplitude and frequency, receptivity. Two fracture types are roughly separated on first and second principal components. The Alex net which has pre-determined weights in the neural network was used in this study, a spectrogram of AE by short time Fourier transform was prepared as input data for CNN. A part of AE classified as inclusion fracture by PCA was assigned to a fin fracture with CNN. The reason of this different classification between two techniques is unclear.