Machine vision for process industries: Monitoring, control, and optimization of visual quality of processes and products

A new paradigm for machine vision for the process industries is proposed, and a framework is illustrated through several industrial applications. The framework is designed for handling the stochastic nature of the visual quality of processes and products in the process industries. In this thesis visual quality means spectral (i.e., color) and/or textural appearance of processes and products where the visual appearance is of great importance and no alternative measurement is available.; The first part of this thesis discusses the challenges of machine vision problems in the process industries and highlights the differences and difficulties compared to those of other machine vision problems. A new machine vision framework is then presented, consisting of several steps: (1) textural and/or spectral feature extraction from images, (2) feature reduction and analysis for the estimation of the visual quality, and (3) predictive modeling, control, and optimization of visual quality. The implementation of the framework in several applications is also discussed.; The second part presents the methodology for extracting textural and spectral features from images. For extracting textural information from images, wavelet texture analysis (WTA) is used throughout the thesis. Theories of WTA are overviewed and illustrated through an application to the classification of steel surface quality. For spectral information extraction, multivariate image analysis (MIA) based on principal component analysis (PCA) of multivariate images is used throughout the thesis.; The third part presents applications of the proposed machine vision framework to industrial problems. For flotation froth monitoring and control, new spectral and textural features based on MR-MIA are proposed for the robust and efficient analysis of froth bubble images. Compared to contemporary approaches for froth image analysis based on texture analysis, the new textural features have a clear morphological meaning, which is closely related to the performance of flotation processes. It is illustrated that monitoring flotation froth can be done easily using multivariate statistical process control (MSPC) charts developed from the MR-MIA features. For froth-based flotation control, a steady-state causal model is built from the MR-MIA features and manipulated process variables, and is used to design an optimization-based controller. (Abstract shortened by UMI.)...