I apologize for the delayed response.
You can create this kind of a model using Scio Lab tools.
As you mentioned, for one collected data set, 5 different models should be created – each time the data should be filtered differently to create the required combination.
Our experience shows that due to different grains weight and particles size, it is difficult to preapre the samples so that the mixture will be homogenous.
As you may know, the more homogounes that sample is, the better results will be calculated.
Also note that different grains may have quite similiar spectral signatures so that it might be difficult to classify between some of them.