Predictive Principal Component Analysis as a Data Compression Core in a Simulation Data Management System
Stefan Mertler, Stefan P. Muller, Clemens-August Thole · 2015
Crash test simulation is a key component of automotive research and development. Managing simulation output has become a huge task, due to the number of simulations and the size of their generated output. Consequently, simulation data management systems (SDMS) were deployed. SDMS provide access to large numbers of simulation results, and therefore, using compression methods to exploit relations between similar simulations can be advantageous. We introduce the predictive principal component analysis (PPCA) as an asymmetric compression method that perfectly fits into SDMS. Since precision requirements in engineering applications prohibits the direct usage of a principal component analysis, we use it as a prediction method. The method divides the compressed data into a database valid for several simulation results and a data set specific to a single simulation. We synchronize the database over night, so that, at the time of request, the data transfer for a single simulation result only involves the result specific compressed file.