Detection of abnormal process behavior in copper solvent extraction by Hotelling T2 and squared prediction error control chart

Once a multivariate model is developed, it can be combined with tools and techniques from univariate statistical process control to form multivariate statistical process control tools. It allows development of advanced process monitoring strategies. In the current study, copper plant history data with multiple variables was successfully treated by principal component analysis to detect abnormal process behavior, particularly, in copper solvent extraction. The multivariate model was based on the concentration levels of main process metals recorded by the industrial on-stream x-ray fluorescence analyzer. Normal operating conditions were defined through control limits that were assigned to Hotelling T² values on x-axis and to squared prediction error values on y-axis. Samples that were beyond the limits were classified as either systematic or random errors, or outliers. Model testing showed successful application of control limits to detect abnormal behavior of copper solvent extraction process as early warnings. Compared to the conventional univariate techniques of analyzing one variable at a time, the proposed model allows to detect on-line a process failure summarizing information from all process variables simultaneously. The proposed methodology was combined with on-line quality monitoring tool developed by VTT, Technical Research Center of Finland, to visualize the results. Thus, the proposed approach has a potential in on-line industrial instrumentation providing fast, robust and cheap application with automation abilities.