A Data Driven Modeling Approach for Store Distributed Load and Trajectory Prediction

Author

Nicholas Peters, Embry-Riddle Aeronautical UniversityFollow

Date of Award

Fall 2022

Access Type

Dissertation - Open Access

Degree Name

Doctor of Philosophy in Aerospace Engineering

Department

Aerospace Engineering

Committee Chair

John A. Ekaterinaris

First Committee Member

Reda Mankbadi

Second Committee Member

Hever Moncayo

Third Committee Member

Hong Liu

Fourth Committee Member

Andrew Wissink

Abstract

The task of achieving successful store separation from aircraft and spacecraft has historically been and continues to be, a critical issue for the aerospace industry. Whether it be from store-on-store wake interactions, store-parent body interactions or free stream turbulence, a failed case of store separation poses a serious risk to aircraft operators. Cases of failed store separation do not simply imply missing an intended target, but also bring the risk of collision with, and destruction of, the parent body vehicle. Given this risk, numerous well-tested procedures have been developed to help analyze store separation within the safe confines of wind tunnels. However, due to increased complexity in store separation configurations, such as rotorcraft and cavity-based separation, there is a growing desire to incorporate computational fluid dynamics (CFD) into the early stages of the store separation analysis. A viable method for achieving this objective is available through data-driven surrogate modeling of store distributed loads. This dissertation investigates the practicality of applying various data-driven modeling techniques to the field of store separation. These modeling methods will be applied to four demonstration scenarios: reduced order modeling of a moving store, design optimization, supersonic store separation, and rotorcraft store separation.

For the first demonstration scenario, results are presented for three sub-tasks. In the first sub-task proper orthogonal decomposition (POD), dynamic mode decomposition (DMD), and convolutional neural networks (CNN) were compared for their capability to replicate distributed pressure loads of a pitching up prolate spheroid. Results indicated that POD was the most efficient approach for surrogate model generation. For the second sub-task, a POD-based surrogate model was derived from CFD simulations of an oscillating prolate spheroid subject to varying reduced frequency and amplitude of oscillation. The obtained surrogate model was shown to provide high-fidelity predictions for new combinations of reduced frequency and amplitude with a maximum percent error of integrated loads of less than 3\%. Therefore, it was demonstrated that the surrogate model was capable of predicting accurately at intermediate states. Further analysis showed a similar surrogate model could be generated to provide accurate store trajectory modeling under subsonic, transonic, and supersonic conditions.

In the second demonstration scenario, a POD-based surrogate model is derived from a series of CFD simulations of isolated rotors in hover and forward flight. The derived surrogate models for hover and forward flight were shown to provide integrated load predictions within 1% of direct CFD simulation. Additionally, results indicated that computational expense could be reduced from 20 hours on 440 CPUs to less than a second on a single CPU. Given the reduction of cost and high fidelity of the surrogate model, the derived model was leveraged to optimize the twist and taper ratio of the rotor such that the efficiency of the rotor was maximized.

For the third demonstration scenario, a POD and CNN surrogate model was derived for fixed-wing based supersonic store separation. Results demonstrated that both models were capable of providing high-fidelity predictions of the store's distributed loads and subsequent trajectory. For the final demonstration scenario, a POD-based surrogate model was derived for the case of a store launching from a rotorcraft. The surrogate model was derived from three CFD simulations while varying ejection force. This surrogate model was then validated against CFD simulation of a new store ejection force. Results indicated that while the surrogate model struggled to provide detailed predictions of store distributed loads, mean load variations could be modeled well at a massively reduced computational cost. For each rotorcraft store separation CFD simulation, the computational cost required 10 days of simulation time across 880. While using the surrogate model, comparable predictions could be produced in under a minute on a single core. Overall findings from this study indicate that massive CFD generated data-sets can be efficiently leveraged to create meaningful surrogate models capable of being deployed to highly iterative design tasks relevant to store separation. Through further improvements, similar surrogate models can be combined with a control strategy to achieve trajectory optimization and control.

Scholarly Commons Citation

Peters, Nicholas, "A Data Driven Modeling Approach for Store Distributed Load and Trajectory Prediction" (2022). Doctoral Dissertations and Master's Theses. 697.
https://commons.erau.edu/edt/697