A robust and automated design optimization framework is developed in this work. This wrapper program automates the design process by coupling the in-house UNstructured PArallel Compressible (UNPAC) flow solver with a novel toolbox for sensitivity analysis based on the discrete adjoint method. The Fast automatic Differentiation using Operator-overloading Technique (FDOT) toolbox utilizes an advanced recording technique to store the expression tree which can significantly reduce the memory footprint and the computational cost of the adjoint calculations. Additionally, this novel toolbox uses an iterative process to evaluate the sensitivities of the cost function with respect to the entire design space and requires only minimal modifications to the available solver. The design optimization framework, UNPAC-DOF, is then employed for aerodynamic design applications based on a gradient-based optimization algorithm. This framework is used to improve airfoil and wing designs for minimized drag or maximized efficiency.
more »
« less
Memory Efficient Adjoint Sensitivity Analysis for Aerodynamic Shape Optimization
An improved version of the Fast automatic Differentiation using Operator-overloading Technique (FDOT) toolbox is developed in this work. The enhanced sensitivity analysis toolbox utilizes an expression-based tape approach -- a first-of-its-kind implementation in Fortran programming language -- that can significantly reduce the memory footprint while improving the computational efficiency of the adjoint-based automatic differentiation (AD). In the proposed approach, the partial derivatives are calculated for each expression using the reverse adjoint accumulation for the active variables involved on the right-hand-side of that expression. The recorded partial derivative information is then used in a very efficient adjoint evaluation process to calculate the entire Jacobian information. The enhanced toolbox is coupled with the in-house UNstructured PArallel Compressible (UNPAC) flow solver for a robust design optimization framework, called UNPAC-DOF. The efficiency and robustness of the proposed technique and the resulting framework are tested for aerodynamic shape optimization problems applied to airfoil and wing geometries.
more »
« less
- Award ID(s):
- 1803760
- NSF-PAR ID:
- 10176294
- Date Published:
- Journal Name:
- AIAA Scitech 2020 Forum
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)A memory efficient framework is developed for the aerodynamic design optimization of helicopter rotor blades in hover. This framework is based on a fully-automated discrete-adjoint toolbox called FDOT. The in-house toolbox is capable of computing sensitivity or gradient information very accurately, and uses an operator-overloading technique that takes advantage of a unique expression-template-based approach for memory and computational efficiency while still being fully-automated with minimal user interventions. The main goal of the present work is to "design" helicopter rotor blades with increased figure-of-merit. Therefore, the flow around the Caradonna-Tung rotor in non-lifting and lifting hover conditions is studied in order to validate the primal and adjoint solvers based on a rotating frame of reference formulation. The efficacy of the optimization framework is first demonstrated for drag minimization of a rotating NACA 0012 airfoil, which resembles a Vertical-Axis Wind Turbine (VAWT) configuration. Finally, the single- and multi-point design optimization results for the Caradonna-Tung rotor are presented. It is important to note that the current approach (FDOT) can be directly coupled -- in a "black-box" manner -- to other existing codes in the Helios computational platform, which is part of CREATE-AV.more » « less
-
An adjoint sensitivity-based approach to determine the gradient and Hessian of cost functions for system identification of dynamical systems is presented. The motivation is the development of a computationally efficient approach relative to the direct differentiation (DD) technique and which overcomes the challenges of the step-size selection in finite difference (FD) approaches. An optimization framework is used to determine the parameters of a dynamical system which minimizes a summation of a scalar cost function evaluated at the discrete measurement instants. The discrete time measurements result in discontinuities in the Lagrange multipliers. Two approaches labeled as the Adjoint and the Hybrid are developed for the calculation of the gradient and Hessian for gradient-based optimization algorithms. The proposed approach is illustrated on the Lorenz 63 model where part of the initial conditions and model parameters are estimated using synthetic data. Examples of identifying model parameters of light curves of type 1a supernovae and a two-tank dynamic model using publicly available data are also included.more » « less
-
more » « less
Abstract This article investigates the ‐optimal estimation problem of a class of linear system with delays in states, disturbance input, and outputs. The estimator uses an extended Luenberger estimator format which estimates both the present and history states. The estimator is designed using an equivalent Partial Integral Equation (PIE) representation of the coupled nominal system. The advantage of the resulting PIE representation is compact and delay free—obviating the need for commonly used bounding technique such as integral inequalities which typically introduces conservatism into the resulting optimization problem. The ‐optimal estimator synthesis problem is then reformulated as a Linear Partial Inequality (LPI)—a form of convex optimization using operator variables and inequlities. Such LPI‐based optimization problems can be solved using semidefinite programming via the PIETOOLS toolbox in Matlab. Compared with previous work, the proposed method simplifies the analysis and computation process and resulting in observers which are non‐conservtism to 4 decimal places when compared with Pad‐based ODE observer design methodologies. Numerical examples and simulation results are given to illustrate the effectiveness and scalability of the proposed approach.
-
more » « less
Abstract We have developed a differentiable programming framework for truncated hierarchical B-splines (THB-splines), which can be used for several applications in geometry modeling, such as surface fitting and deformable image registration, and can be easily integrated with geometric deep learning frameworks. Differentiable programming is a novel paradigm that enables an algorithm to be differentiated via automatic differentiation, i.e., using automatic differentiation to compute the derivatives of its outputs with respect to its inputs or parameters. Differentiable programming has been used extensively in machine learning for obtaining gradients required in optimization algorithms such as stochastic gradient descent (SGD). While incorporating differentiable programming with traditional functions is straightforward, it is challenging when the functions are complex, such as splines. In this work, we extend the differentiable programming paradigm to THB-splines. THB-splines offer an efficient approach for complex surface fitting by utilizing a hierarchical tensor structure of B-splines, enabling local adaptive refinement. However, this approach brings challenges, such as a larger computational overhead and the non-trivial implementation of automatic differentiation and parallel evaluation algorithms. We use custom kernel functions for GPU acceleration in forward and backward evaluation that are necessary for differentiable programming of THB-splines. Our approach not only improves computational efficiency but also significantly enhances the speed of surface evaluation compared to previous methods. Our differentiable THB-splines framework facilitates faster and more accurate surface modeling with local refinement, with several applications in CAD and isogeometric analysis.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.2514/6.2020-0885
