Dynamic Estimation Of Unmeasured Variables In An Industrial Grinding Circuit Using A Kalman Filter

A PDP 11/34 computer has been inter- faced to existing instrumentation in a 400 tph rod-mill/ball-mill grinding circuit and programed with a Kalman filter to estimate several unmeasured variables. A dynamic two-size- fraction model of the circuit is used to predict the complete state of the circuit including particle size distributions around the circuit, mass holdups and flowrates, the ball-mill grinding rate, water additions and cyclone parameters. Online measurements of sump level, slurry densities and product particle size are used to update the model values sequentially on a 20-second interval. This procedure results in optimal estimates of circuit variables given the relative reliability of measurements and the accuracy of the model.