Hello Pablo, 

As already mentioned in your title, attrition is the main mechanism for size reduction, i.e. interaction between ore particles and grinding media gradually reducing the size of the ore. From what you're saying your media size may be a bit on the fine side, though this depends on whether your feed has a coarse tail. 

For predictive models look up signature plots. This is the standard approach for ultrafine grinding. 

Look at this paper:

http://www.isamill.com/EN/Downloads/Downloaded%20Technical%20Papers/Development%20of%20the%20Larson%20Morrison%20IsaMill%20JKSimMet%20Model.pdf

"The IsaMill is a high intensity horizontal stirred mill utilizing small 2-6 mm ceramic grinding media for attrition grinding. Grinding duties range from feeds of up to 300 microns being ground to 40 microns and UFG grinds as fine as 5 microns. Being completely dissimilar to normal ball mill breakage it was desired to produce a model for this process in JKSimMet. The result of this study was the discovery that IsaMill breakage can be reliably predicted on a basis of energy versus the squared value of the percent passing given sizes. This relationship can be used to analyse circuit efficiency for varying feed sizes. By using this new relationship the product size distributions for new feed sizes and energies can be reliably predicted with simple math, something not previously possible with only a signature plot. This paper will provide details on the model development, validation and implementation."

and obtain a copy of  Larson thesis from University of Queensland, "Experimental Study of IsaMill Performance Leading to a Preliminary Model", 2013.

https://espace.library.uq.edu.au/view/UQ:298924

"As the IsaMill has gained greater acceptance and is utilized in a greater range of applications it has become desirable to develop a model to predict the IsaMill performance. Laboratory testing was undertaken with the M4 IsaMill to determine which operating variables affect the grinding energy efficiency and to what extent they impact the grinding energy efficiency. The standard signature plot used for scaling grinding energy requirements ensures that the lab scale results will reliably translate to mine site operations. Variables which have been tested include the mill speed, feed pump volumetric flowrate, grinding media type, grinding media size, media filling and feed pulp density. In addition viscosity was investigated as it relates to the feed pulp density and grinding energy efficiency. Additional tests were done exploring the viability of grinding coarser mill feeds with a range of media sizes and in series. Further work was completed analyzing the ceramic media with the cone beam tomography (CBT). Analysis of IsaMill product ultrafine particles for sizing and assay was also done with the Mineral Liberation Analyser (MLA).

The end result of this work was a better understanding of the performance of the M4 IsaMill. It was found that the mill speed did not affect grinding energy requirements; it simply changed residence time requirements as the energy moved along the signature plot line. The same effect was seen for volumetric feed pump flowrate, as it resulted in duplicating the same signature plot line. For the copper concentrate and other coarse feed samples, it was found that there will be an optimum media size which depends not only on the coarseness/hardness of the feed, but is also dependant on the target product size. A preliminary guide to choosing ceramic media size was developed based on this limited work with coarser feed. For choosing media type, overall stiffness appears to be the most important variable. Media loading was shown to be proportional to grinding energy efficiency, where the power required to grind to a target size reliably decreased as media loading increased. Feed pulp density was shown to have an inconsequential effect on grinding energy efficiency until a point was reached where slurry viscosity began to rise quickly. This appears to correlate with the volume percent solids. General trends were also observed in net mill power when changing different variables that may make it possible to predict the M4 performance in the future and better control signature plot tests.

From these results a simple model was developed to predict product size distributions with a given feed while changing mill operating conditions. A novel approach was developed analyzing fines production energy requirements to account for changing feed sizes. While the fines production could not be validated, the base of the model was successfully proven over the final set of tests. A simple MS Excel model has been developed. It is also suitable for implementation as a model in JKSimMet."