Achieving optimal dense suspension rheology is key to meeting product quality requirements in a wide range of industries, but the traditional optimization approach, which is heavily reliant on physical trial-and-error, is prohibitively time consuming and expensive. Virtual optimisation can lead to significant time and costs savings in this context. This work demonstrates an efficient virtual optimisation methodology that combines Discrete Element Method (DEM) simulation and machine learning to rapidly identify the optimal particle scale properties for a target suspension viscosity. The methodology consists of parametrizing the particle size distribution, morphology, volume fraction and surface frictional properties, automatically generating and running DEM simulations of dense granular suspensions subjected to simple shear for a well distributed quasi-random sample of the parameter space and training a Reduced Order Model (ROM) on the resulting synthetic data using machine learning. A multi-objective genetic algorithm is then utilized to rapidly estimate the globally optimal parameter set from the ROM. This results in several orders of magnitude reduction in computational expense relative to the equivalent purely simulation-based approach and makes the virtual optimisation of dense suspension rheology from particle scale properties practical. The advantages and limitations of the proposed methodology are further discussed in the talk.