Gas hold-up estimation in bubble columns using passive acoustic waveforms with neural networks

Abstract

Abstract Passive acoustic waveforms produced experimentally from a bench-scale two-phase bubble column were recorded using a miniature hydrophone at three axial positions. The generated acoustic waveforms were processed and trained using artificial intelligence against global gas hold-up measurements. Two neural network architectures, the radial basis function (RBF) neural network and the recurrent Elman neural network, were employed. Both neural network techniques achieved accurate gas hold-up estimation, characterised by low mean square errors of 2.70 and 1.68% for the RBF and recurrent Elman networks respectively. The designed and trained neural networks were found to be a powerful tool for learning and replicating complex two-phase patterns. Passive acoustic waveforms were found to be a useful measuring technique for gas hold-up estimation in bubble columns under moderate operating conditions.