Qualitative statistical analysis of simulated data from a pilot scale mill

Grinding is the process of reducing a particle size distribution of an extracted ore and is commonly performed in a tumbling mill. It is a complex procedure and there is a lack of knowledge of what really happens inside the mill. A number of pilot-scale experiments were done at LKAB's pilot plant at Malmberget, Sweden [1]. In this particular pilot mill, a continuous charge measurement system is installed in one of the lifters and it gives a deflection signal produced by the mill charge. From this signal it is possible to detect features correlated to the settings of the mill. Large, real experiments are very difficult to control and are of course, very costly and time consuming. A 10 cm slice of the mill was simulated with discrete element method (DEM) for different mill operating conditions. From the simulations a deflection signal was extracted and validated against real data. There is a difference in the signal, mainly due to the lack of slurry in the simulations, but the behaviour when the mills operating conditions changes seems to be the same in both the simulated and the measured signals. To analyse the data from the simulation a statistical analysis on a full factorial design was done. Two levels of degree of filling of the mill, two different rotational speeds, two levels of friction and different types of particles were selected as factors. The response data are two angles: toe and shoulder angle. The toe angle is when the lifter hits the charge and the shoulder angle is when the lifter leaves the charge. The analysis show that the toe angle increases when the degree of filling is low and the rotational speed is high. It is also clear that the particle shape influences the charge behaviour. The simulated changes correspond to changes detected in pilot mill runs. This is important since it validates the DEM model. In essence, mill simulations are easily done and the changes of factor levels cause the simulated mill to react in similar manner as in real cases. One advantage is that in simulations one factor can be isolated and changed while the others are kept at constant values, which in turn creates the possibility to investigate one factor at a time. In real experiments, the factors are more dependent on each other and there is a very high disturbance from noise.Godkänd; 2011; 20111229 (palle