PhD Proposal by Zhenyu Gao

Primary tabs

Zhenyu Gao
(Advisor: Prof. Dimitri N. Mavris)

will propose a doctoral thesis entitled

Representative Data and Models for Complex Aerospace Systems Analysis


Friday, May 14 at 10:00 a.m. (EDT)
Online via Bluejeans




Data-driven analysis of complex aerospace systems often involves two kernel elements: large amounts of real-world operations data, and a diverse ensemble of models. Nevertheless, in reality due to constraints in computational cost or resources, practitioners can face the inability to process the entire large data set or build a complete model portfolio when performing simulation and systems analysis. In this dissertation we propose the use of “representatives”, which is the opposite of the entire population, to conduct efficient and accurate systems analysis. The proposed methods utilize data mining and high-dimensional data analysis to select a small proportion of representative data and models from the population and apply them to tackle challenges in several application cases in aviation environmental impact modeling.

The first part of this dissertation addresses the challenge of representative data. Specifically, we consider the scenario of an extreme numerosity reduction on large data sets while still maintaining the same data distribution. We propose Probabilistic REpresentatives Mining (PREM), an efficient data mining approach to obtain probabilistically representative small data sets. PREM employs a balanced clustering set-up which avoids over-sampling and under-sampling phenomena produced by traditional clustering algorithms and a multi-stage computing strategy which enables the method to be scalable on massive data sets.

The second part proposes the concept of representative models, which tackles the challenge of insufficient resources in building Aircraft Noise and Performance (ANP) models. In the first scenario, we consider the problem of selecting a representative model portfolio at each permitted size level to sufficiently cover the entire population space with the desirable maximum distortion guarantee. In the second scenario, we introduce the use of mixture models to represent and model ``unconventional groups'' in the population, where the model substitution is compromised by a lack of coverage. Mixture model relies on the identification of complementary candidate models and uses a multi-model approach to outperform any candidate model alone.

Overall the dissertation is expected to make contributions to both analytical methodologies for certain scenes and the solutions to specific challenges in aviation environmental impact modeling.



  • Prof. Dimitri N. Mavris – School of Aerospace Engineering (advisor)
  • Prof. Koki Ho – School of Aerospace Engineering
  • Prof. Graeme J. Kennedy – School of Aerospace Engineering
  • Prof. Yao Xie – School of Industrial and Systems Engineering
  • Dr. Tejas G. Puranik – School of Aerospace Engineering


  • Workflow Status:Published
  • Created By:Tatianna Richardson
  • Created:05/04/2021
  • Modified By:Tatianna Richardson
  • Modified:05/04/2021