This article is devoted to a detailed description of the developed physical model of the pneumatic classification process for detecting the rotating suspended layer and ensuring the frequency of loading and unloading of a pneumatic classifier. The related mathematical model is developed for estimating the non-stationary concentration of fine particles in a gas-dispersed flow with respect to time and height of the working space of the apparatus. The research is aimed at developing a pneumatic classification method for granular materials using a rhomb-shaped apparatus and ensuring the reliability of the operating process based on the influence of the flow on the granular material concentrations. The obtained experimental results allow evaluating the rational geometrical parameters of the working space in pneumatic granulators, as well as verifying the proposed mathematical model based on the implementation of the quasi-linear regression procedure. It is shown that the rhomb-shaped pneumatic classifier provides effective separation of granular material, reaching up to 95% of the target fraction. As a result, the proposed methodology can be implemented for optimizing geometrical profiles of pneumatic classifiers in terms of defining the required technological parameters of the pneumatic classification process.