Separating Valid from Invalid Inputs for a Digital Aircraft Design Tool

Abstract

Novel aircraft are increasingly designed by combining numerous discipline-specific simulation and modeling tools. To determine whether existing tools can be used to design a given aircraft, manufacturers require in-depth knowledge about their tools' capabilities. The research software used in digital design workflows, however, often lacks formal capability specifications. In this work we develop an oracle for DELiS, one such tool which generates finite element models of aircraft wings and which is a critical component in a real-world engineering toolchain. For this, we identify the six-dimensional input space of the tool and determine the boundary between successful and failing inputs via computing the complex hull of successful inputs. We detail the generation of test cases in this real-world example through uniform random sampling, scaling existing test cases, and binary search along the functionality boundary. Our evaluation shows that a one-dimensional correlation approach proves inadequate for reliable prediction. With a static convex hull oracle we correctly predict crashes in 65% of cases. Finally, with dynamic enhancements through scalar parameter vector multiplication and iterative genetic test case generations, we correctly predict crashes in 76.6% of cases. This work enables project planners to estimate in advance whether proposed aircraft configurations require software modifications, significantly improving resource allocation and reducing project risks in digital aircraft design.