Monitoring of froth systems using principal component analysis

[摘要] ENGLISH ABSTRACT:Flotation is notorious for its susceptibility to process upsets and consequently its poorperformance, making successful flotation control systems an elusive goal. The control ofindustrial flotation plants is often based en the visual appearance of the froth phase, anddepends to a large extent on the experience and ability of a human operator. Machinevision systems provide a novel solution to several of the problems encountered inconventional flotation systems for monitoring and control.The rapid development in computer VISIon, computational resources and artificialintelligence and the integration of these technologies are creating new possibilities in thedesign and implementation of commercial machine vision systems for the monitoring andcontrol of flotation plants. Current machine vision systems are available but not withouttheir shortcomings. These systems cannot deal with fine froths where the bubbles arevery small due to the segmentation techniques employed by them. These segmentationtechniques are cumbersome and computationally expensive making them slow in realtime operation.The approach followed in this work uses neural networks to solve the problemsmentioned above. Neural networks are able to extract information from images of thefroth phase without regard to the type and structure of the froth. The parallel processingcapability of neural networks, ease of implementation and the advantages of supervisedor unsupervised training of neural networks make them potentially suited for real-timeindustrial machine vision systems. In principle, neural network models can beimplemented in an adaptive manner, so that changes in the characteristics of processesare taken into account.This work documents the development of linear and non-linear principal componentmodels, which can be used in a real-time machine vision system for the monitoring, andcontrol of froth flotation systems. Features from froth images of flotation processes were extracted via linear and non-linearprincipal component analysis. Conventional linear principal component analysis andthree layer autoassociative neural networks were used in the extraction of linear principalcomponents from froth images. Non-linear principal components were extracted fromfroth images by a three and five layer autoassociative neural network, as well as localisedprincipal component analysis based on k-means clustering. Three principal componentswere extracted for each image. The correlation coefficient was used as a measure of theamount of variance captured by each principal component.The principal components were used to classify the froth images. A probabilistic neuralnetwork and a feedforward neural network classifier were developed for the classificationof the froth images. Multivariate statistical process control models were developed usingthe linear and non-linear principal component models. Hotellings T2 statistic and thesquared prediction error based on linear and non-linear principal component models wereused in the development of multivariate control charts.It was found that the first three features extracted with autoassociative neural networkswere able to capture more variance in froth images than conventional linear principalcomponents, the features extracted by the five layer autoassociative neural networks wereable to classify froth images more accurately than features extracted by conventionallinear principal component analysis and three layer autoassociative neural networks. Asapplied, localised principal component analysis proved to be ineffective, owing todifficulties with the clustering of the high dimensional image data. Finally the use ofmultivariate statistical process control models to detect deviations from normal plantoperations are discussed and it is shown that Hotellings T2 and squared prediction errorcontrol charts are able to clearly identify non-conforming plant behaviour.

[发布机构] Stellenbosch University