Aeroelastic System Development Using Proper Orthogonal Decomposition and Volterra Theory
read more
Citations
Future of Airplane Aeroelasticity
Identification of Nonlinear Aeroelastic Systems Based on the Volterra Theory: Progress and Opportunities
Proper orthogonal decomposition in wind engineering - Part 1: A state-of-the-art and some prospects
Fast prediction of transonic aeroelastic stability and limit cycles
Data-driven modeling for unsteady aerodynamics and aeroelasticity
References
The Proper Orthogonal Decomposition in the Analysis of Turbulent Flows
Computational Fluid Mechanics and Heat Transfer
Turbulence, Coherent Structures, Dynamical Systems and Symmetry
An eigensystem realization algorithm for modal parameter identification and model reduction
Related Papers (5)
Reduced-order modeling: new approaches for computational physics
Three-Dimensional Transonic Aeroelasticity Using Proper Orthogonal Decomposition-Based Reduced-Order Models
Proper Orthogonal Decomposition Technique for Transonic Unsteady Aerodynamic Flows
Frequently Asked Questions (18)
Q2. What was the cost of computing the state-space realization using ERA?
The flow field was projected onto the aeroelastic modes to determine the modal amplitudes, and a linear, state-space realization for the fluid dynamics was synthesized from the modal impulse responses.
Q3. What is the modal solution for the deflection w(x, t)?
Two pinned boundary conditions are enforced at the panel’s endpoints: w = 0 and ∂ 2w ∂x2 = 0.A modal solution for the deflection w(x, t) is assumed:w(x, t) =ms∑i=1ai(t) sin(iπx), (3)where ms is the number of structural modes retained, and the modal amplitudes ai vary in time and are collocated in the array a.
Q4. How many snapshots were taken of the structural response?
Aeroelastic structural modes were generated using 100 snapshots of the structural response obtained during the training of the fluid ROM.
Q5. How many data points were sampled to obtain the full-system response to the impulse?
The full-system response to this impulse was sampled over 30 non-dimensional time units at a rate of dt = 0.015432 for a total of K = 1944 discrete data points.
Q6. What was the ROM for the slug-flow base case?
5. The steady-state base case ROM was better suited to large LCO amplitudes where the larger panel amplitudes excited a panel nonlinearity that corrupted the results of the slug-flow base case.
Q7. What is the reason for the improvement in performance of the ERA?
The authors suspect the improvement in performance associated with the steady-state base flow was most likely due to the choice of data windowing parameters used in the ERA realization.
Q8. How many additional runs of the full order solver were required to provide the impulse response data?
Since there were eight forcing terms, eight additional runs of the full order solver were required to provide the impulse response data.
Q9. What is the effect of impulsing the force term of a truly linear system?
Impulsing the forcing term (or terms) of a truly linear system produces a response that is the building block necessary to recreate the system output from any arbitrary forcing function.
Q10. What was the highest value for stability of the second-order method?
The time-step used for the full system was based on a Courant-Friedrichs-Lewy (CFL) condition of 0.5, which was the highest value allowing for stability of the second-order method.
Q11. What was the phase and frequency error for the steady-state base case?
The phase and frequency error were negligible for the steady-state base case, but the larger amplitudes of the slug-flow base case introduced a small increase in LCO frequency (resulting in an accumulating phase error).
Q12. How many times did the computer process the aeroelastic modes?
An additional 25 time unit run was required for snapshot collection to generate the aeroelastic modes, bringing the total computer processing time to 12546 seconds, or roughly half of the computational cost associated with a single full-system time integration.
Q13. What is the effect of a sudden appearance of a velocity profile on the boundary?
The sudden appearance of a velocity profile on theboundary (and its sudden removal one time step later) produced a shock wave of varying strength running the length of the panel.
Q14. How many time units did the impulse response take to become fully developed?
At Mach 1.2, LCO required about 300 time units to become fully developed, and the small, 25 time unit training window was shown to be adequate in previous work.
Q15. What was the reason for the POD-Volterra ROM?
The POD-Volterra ROM reduced compute time by fourorders-of-magnitude, and realized an improvement in performance consistent with the DOF reduction.
Q16. How many subiterations are necessary to drive R to near machine zero?
A suitable number of subiterations are computed at each time step to obtain a good approximation to Y n+1; typically, 1-2 subiterations are generally sufficient to drive R to near machine zero.
Q17. What was the accuracy of the ROM at representing the full system response?
The accuracy of the ROM at representing the full system response was encouraging, and the frequency of unstable oscillations did match the LCO frequency.
Q18. What was the effect of the modal amplitudes on the aeroelastic flow?
The modal amplitudes were examined and modes 6, 7 and 8 in energy content had much greater contribution to the aeroelastic flow field than did the first 4 modes (note that the modes were normalized).