Characterisation of Dynamic Process Systems by Use of Recurrence Texture Analysis

Abstract

This thesis proposes a method to analyse the dynamic behaviour of process systems using sets of textural features extracted from distance matrices obtained from time series data. Algorithms based on the use of grey level co-occurrence matrices, wavelet transforms, local binary patterns, textons, and the pretrained convolutional neural networks (AlexNet and VGG16) were used to extract features. The method was demonstrated to effectively capture the dynamics of mineral process systems and could outperform competing approaches.