We propose a novel computational method for generating data needed to create decision strategies for condition-based monitoring algorithms that can effectively differentiate between a healthy system and different types of defects in a damaged system. Currently, the only means available to generate this data are physical testing which is time consuming and expensive, and simplified computer models- either lumped parameter models or 2D models. The most advanced current computational model of drive systems with surface and crack damage can only be deployed on stand-alone computers. The existing contact algorithm relies on shared memory between CPUs, and quickly saturates memory bandwidth. We propose innovative modifications to the algorithm so that models may be efficiently deployed on very large clusters of computers connected by high speed networks. These changes will make possible realistic time-domain 3D modeling of drive systems with surface and crack damage.