Statistical monitoring of a grinding circuit: An industrial case study

Abstract

With the increasing availability of large amounts of real-time process data and a better fundamental understanding of the operation of mineral processing units, statistical monitoring of mineral processing plants is becoming increasingly widespread. Process plants are typically too complex to model from first principles and therefore models based on historical process data are used instead. Multivariate methods such as principal component analysis are indispensable in these analyses and in this paper, it is shown how the statistical analysis of process data from a grinding circuit and a sound fundamental knowledge of the operation of mineral processing plants complement one another. For this purpose a philosophy for the statistical monitoring and cause and effect analysis of a process was outlined. It was shown how a well defined process hierarchy with complementing performance measures can effectively be used to detect a shift in the operation of a mineral processing plant and find the root cause of the shift. Visualisation of the results was found fundamental in communicating the findings of the statistical analysis to the processing plant. This resulted in the requirement for multidimensional visualisation of the process for which principal component analysis plots and process performance graphs in the form of two-dimensional histogram plots and parallel plots were found to be the most effective. Data availability, process variable selection, process hierarchy definition and performance measure selection were also found to be critical factors directly impacting on the success of statistically monitoring a process.