Using model discrimination to select a mathematical function for generating separation curves

A statistical technique, known as model discrimination, is presented for selecting a mathematical function to fit separation data in order to estimate the characteristic parameters — the relative density of separation and the probable error. Some of the more commonly used functions are compared by their goodness of fit and their parameters evaluated statistically using a sensitivity analysis approach in order to determine the function with the best fit and the maximum parameter sensitivity. Application of this technique is demonstrated using sets of separation data from actual heavy media cyclone operations. For separator data exhibiting apparent bypassing, i.e., a fraction of feed apparently exiting with the clean coal and/or a fraction of feed apparently exiting with the refuse, a modelling technique is given whereby an existing function can be expanded to incorporate the bypass parameters.