Multivariate projection and analysis of microstructural characteristics of mechanically activated hematite in different grinding mills

A statistical analysis was done to investigate the relationship between grinding variables and structural changes during mechanical activation of hematite concentrate. Experiments were carried out according to a statistical design by varying the grinding time, media surface and mill type. Several multivariate techniques are applied to interpret the grinding processes.The variance analysis revealed that the media surface and grinding time significantly influence the five main response variables at 95% confidence level. The use of multivariate analysis allows the projection of high-dimensional data to a low subdimensional subspace. An overview of principal component analysis (PCA) on 27 variables yielded a three component model explaining 89% and predicting 76% of the total variance. It was found that the observations belonging to low and high levels of media surfaces fall into two groups. Most of the microstructural characteristics such as microstrain, dislocation and amorphization and granulometric surface area, BET specific surface area, specific energy input, stored energy, portion of smaller particles and stress energy coincide with high level of media surface group. The variables crystallite size, peak intensity and mean particle size appear with lower media surface. The PLS-DA (partial least squares discrimination analysis) made it possible to discriminate the three types of mills. From the projection of dummy variables, it was concluded that the vibratory mill caused comparatively less structural changes in hematite than the other mills in spite of releasing higher stress energy. The planetary mill introduced relatively higher dislocation defects and generated higher lattice strain. The hematite ground in the tumbling and planetary mills had comparatively higher X-ray amorphization degree and subsequently higher excess energy than the hematite ground in the vibratory mill. The tumbling mill produced relatively lower specific surface than the others. It was concluded that the products of the tumbling mill represented higher defect concentration (amorphization) per unit surface area despite releasing lower stress energy level. From the PLS modeling of the five main response variables, it was found that the X-variables specific energy input and stress energy are the most influencing factors.