Multivariate statistical analysis of mineral processing plant data

CIM Bulletin, Vol. 86, No. 975, 1993

D. Hodouin, Departement de Mines et Metallurgie, Universite Laval, Quebec, Quebec, J.F. MacGregor, M. Hou, Department of Chemical Engineering, McMaster University, Hamilton, Ontario, and M. Franklin, Noranda Technology Centre, Pointe-Claire, Quebec

Because of the availability of powerful industrial computers which collect huge amounts of real-time data in mineral processing plants, there is a need for efficient methods to extract relevant information from them. Multivariate statistical techniques, such as principal components analysis (PCA) and projection to latent structures (PLS), are well suited to analyze these large sets of noisy and ill-conditioned data. The power of PCA and PLS is illustrated on historical data from a grinding and flotation plant. Three hundred and fifty observations of forty-four process variables are used to show the capacity of these techniques for preliminary data analysis, classification of operating regimes, process monitoring, and process empirical modelling.