Session: 34-06 Geometry and fluid-structure interaction

Paper Number: 153217

Advancing Turbomachinery Meanline Modeling and Optimization With Automatic Differentiation 

Meanline modelling plays a crucial role in the turbomachinery design process by enabling accurate and computationally efficient performance analysis and design optimization. In gradient-based optimization techniques, accurate derivatives of the objective function and constraints are essential to guide the search for optimal solutions. However, if these gradients are inaccurate, particularly in high-dimensional and complex simulations, the reliability of optimization frameworks can be compromised, leading to convergence issues and instability. Obtaining exact derivatives analytically is often infeasible due to the complexity of meanline models and equations of state, while commonly used numerical differentiation techniques such as finite difference methods introduce inaccuracies in gradient computation due to truncation and round-off errors. To address these challenges, this paper presents the first application of automatic differentiation in turbomachinery meanline modeling. Automatic differentiation was implemented to compute exact gradients within an existing meanline model, and the performance of various gradient-based optimization solvers was evaluated by comparing their convergence when evaluating exact derivatives with automatic differentiation versus approximate derivatives obtained through finite differences. This was done for various test cases, including single-stage and multi-stage turbines. The results suggest that the automatic differentiation technique increases accuracy and computational efficiency by decreasing the number of iterations, particularly in multi-stage turbine problems involving a larger number of design variables. These findings demonstrate that the meanline model using automatic differentiation for gradient calculations leads to faster and more reliable convergence, paving the way for its application to more complex flow problems, such as those encountered in, for example, reversible and two-phase turbomachinery.

Presenting Author: Srinivas Prakash Diwanji Technical University of Denmark

Presenting Author Biography: I am currently pursuing PhD studies in Technical University of Denmark in the field of Turbomachinery focusing on Multiphase flows turbine design and optimization.

Authors:

Srinivas Prakash Diwanji Technical University of Denmark
Lasse Borg Anderson SINTEF
Roberto Agromayor Technical University of Denmark
Fredrik Haglind Technical University of Denmark