Powder bed fusion processes are additive manufacturing technologies, where parts are built up layer by layer in a powder bed. The highly complex geometries manufactured with these technologies are the main advantage and the difference to conventional technologies, where they are hardly machinable. Numerical models can replace the intensive and expensive trial-and-error principle to find suitable process parameters for the production of accurate parts. This work presents a stochastic algorithm to generate a random powder bed for numerical simulations of powder bed fusion processes. The main focus is the efficient coupling of a classical discrete element approach to a grid-based solver used to simulate the melting process. The algorithm is implemented and validated in the range of common relative powder layer densities of powder bed fusion processes. The computational efficiency is demonstrated and finally, the complete coupling to a process simulation of the selective electron beam melting process is presented. (C) 2018 Elsevier B.V. All rights reserved.