Machine Learning Based Data Compression in High-Fidelity Numerical Simulation

NE has invested heavily in the development of a wide range of computational frameworks over the past few decades. This considerable investment in advanced high fidelity codes has allowed the Nuclear engineering community to greatly reduce the non-recurring costs associated with reactor development. However, as the fidelity of these simulations grows, so too do the difficulties associated with using the simulations in industrial settings. Issues associated with writing, storing and analyzing the extreme levels of high-fidelity simulation data are not unique to NE. In fact, NASA, the DOE, and CERN have all expressed the need for software and tools that can marshal high-fidelity simulation data into a format that can be quickly and efficiently used to rapidly inform real-world design decisions. This framework will use our novel DLS data compression algorithm to support best in class accuracy-to-compression ratios while still enabling fast, intuitive data analytics. In Phase I we demonstrated the algorithmic and computational feasibility of DLS data compression in two ways. First, we used DLS data compression to compress an unstructured NE dataset representing flow through a bed of pebbles. With that dataset we obtain 19x compression with ˜2% error. We also implemented a parallel MPI version of the DLS algorithm that significantly reduced DLS compression times. In Phase II we will implement the optimizations needed to commercialize our novel approach to data compression. This will include an investigation into more advanced machine learning algorithms, GPU acceleration, integration into existing HPC workflow tools, demonstrations, and more. For the developers of numerical software, the library will provide a fast, in-situ method for exporting scientific data from advanced numerical simulations. The customers falling into this group include many government agencies who rely on numerical simulation including the DOE, NASA, Air Force, Navy, Army, and Marines. The commercial sector is also a heavy user of mesh based numerical simulation applications. Those applications include the aeronautical, biomechanical, and automotive industries. The underlying algorithms of the proposed library could be used for data compression and/or fusion of large-scale experimental data sets, or multi-physics datasets such as those common in fracture analysis and space vehicle design, among many others.