Simulation-based study of the influence of particle physical properties on fertilizer spreading ability

• Computer simulations are a valuable method to study fertilizer spreading ability. • The maximum working width is expressed as a function of five physical properties. • The relative contribution of each fertilizer property contribution is quantified. • Drag coefficient, specific density and modal diameter are key parameters. • The working width is correlated to the mean radius of the spread pattern. Despite spreader performance, different fertilizers are known to have inherent limitations that prevent them from achieving the same working width at a targeted spreading quality. Nevertheless, there is currently no mathematical relationship to express the maximum working width achievable by a fertilizer as a function of its physical properties. The aim of this paper is to develop a numerical method to provide such a mathematical relationship. Since the maximum working width W WCV10 ensuring a satisfying uniformity (CV≤10%) is a complex parameter (resulting from different stages of the spreading process), the issue has been decomposed into two steps. First, a dataset is computed by numerical simulations. 3800 virtual fertilizers are designed and feed into numerical simulations based on a Monte Carlo process. For each fertilizer, twin disk spread patterns are computed for a set of virtual spreader settings. Then, W WCV10 is deduced from simulations of adjacent passes and overlaps. The mean radius R m of the spread pattern is also computed. Second, based on this consistent dataset, multivariate polynomial regressions are established to express W WCV10 and R m as mathematical functions of 5 independent physical characteristics of the fertilizer: the drag coefficient, the specific density and three particle size distribution parameters. Thus, for the first time, the contribution of each variable (to the ability of the fertilizer to achieve a working width) is quantified and the sensitivity of W WCV10 and R m is analyzed by using partial derivatives. The correlation between W WCV10 and R m is also demonstrated.