Simplicit: Implicit geological simulation using high performance computing
This project is based on a technical proposal requested by BHP for an applied research project for acceleration and optimization of geostatistical simulation algorithms, using High Performance Computing (HPC) capabilities. The proposed method simulates distances to geological contacts by using a novel multivariate spectral turning bands simulation algorithm. The simulation is augmented by incorporating other kinds of distances as secondary variables, obtained from qualitative and measured sources such as the geological model and drillhole data for the studied domain. The accompanying graphical interface, dubbed Simplicit, allows controlling the input and output formats and fine-tuning of the composing algorithms.
The distance computing and simulation algorithms were both implemented in CUDA for running on Graphics Processing Units (GPU), making full use of the HPC capabilities provided by consumer-level cards and beyond. These algorithms were designed for GPU performance in mind, allowing their CUDA implementations to largely surpass naive parallelizations of methods such as sequential gaussian simulation and typical distance computation algorithms. This allows generating simulations of dozens of realizations in minutes, which could not be previously done for industry-scale block models and was considered infeasible due to the required time with existing alternatives.
The project was developed over the course of three years of collaboration, feedback and validation by BHP personnel, resulting in the successful transfer of the software and the ongoing writing of a research publication.