Data-Driven Engineering Design: Aerodynamic design insight in the age of data

Broadly, this talk is about the capitalization of data guided by principles of aerodynamics, dimension reduction and uncertainty quantification. This data—acquired from computer models, instrumentation sensors, or both—is likely to be noisy, high-dimensional and sparse. The objective of my research is to identify salient aerodynamic relationships in this data to empower improved aerodynamic design insight, analysis, and manufacturing strategies. In this talk, I will present two case studies that encapsulate these ideas. 
The first case study concerns the way one designs and manufactures blade, where data arises from computational flow simulations. Here ideas from the field of subspace-based dimension reduction can play a powerful role in ensuring more efficient design workflows, while simultaneously cutting down the costs associated with manufacturing to certain tolerances. The identification and exploitation of these subspaces—weighted linear combinations of all the design variables—facilitate detailed quantification of aerodynamic design trade-offs. In other words, instead of simply stating that certain blades are more sensitive to manufacturing variations, or that they lie on a hypothetical “efficiency cliff”, I will demonstrate how we can visualize, quantify and make precise such notions of design.
The second case study revolves around how one can reconstruct an aerodynamic flow-field given sparse (and noisy) measurements from thermocouples. We are concerned here with approximating the temperature field (and its area average) given data from a few circumferentially scattered rakes in an engine. By developing new ideas within uncertainty quantification, specifically Bayesian statistics, I will demonstrate how we can estimate the spatial flow-field (and its uncertainty), permitting a greater understanding of key loss mechanisms and aerothermal harmonics. Overall, this work has broad ramifications on the way we book-keep uncertainties in experiments, and articulate how existing metrics in international uncertainty standards (e.g., ASME, ISO GUM) fall short.
I will close this talk by offering a glimpse of ongoing and future avenues for my research.

Bio of speaker: Pranay Seshadri is a postdoctoral fellow at the Department of Engineering, University of Cambridge. He is also a Group Leader within the Data-Centric Engineering Programme, largely focused on aerospace. He obtained his Ph.D. in turbomachinery and computational engineering at the University of Cambridge.