Using open source code libraries for robust design analysis

Abstract

The design of systems today often involves computer simulation to assess performance and design margins. Understanding how variability erases design margin is important to assure adequacy of margins, especially in optimization efforts. In this paper, we develop a toolchain using open source code libraries in Python, and encapsulate it in Jupyter notebooks, to provide an open source, interactive uncertainty quantification and sensitivity analysis toolchain. This works generally with simulation tools, where a reference folder is created containing a script that reads an input file of parameter values and runs the simulation. With that easily created, the toolchain executes the necessary uncertainty quantification steps with replicates of that reference folder. This approach fits within a broader workflow outlined that defines the variation modes to study, maps to simulation inputs, and screens the variables for sensitivity before conducting an uncertainty quantification. An example is shown in the simulation analysis of a Stirling engine.