0
Select Articles

Toward Gradient-Based Optimization of Full Gas Turbines PUBLIC ACCESS

Mechanical Engineering 141(03), 54-55 (Mar 01, 2019) (2 pages) Paper No: ME-19-MAR7; doi: 10.1115/1.2019-MAR-7

Abstract

The design of a full gas turbine is a painstaking process, with many interactions between different physics, components, and engineers. Not surprisingly, this is an effort spanning over many years before a compromise can be found that satisfies all involved engineering disciplines. But can the design cycle not be shortened in this modern age, dominated by increasing computational power and emerging artificial intelligence?

Professor Tom Verstraete Von Karman Institute for Fluid Dynamics

The design of a full gas turbine is a painstaking process, with many interactions between different physics, components, and engineers. Not surprisingly, this is an effort spanning over many years before a compromise can be found that satisfies all involved engineering disciplines. But can the design cycle not be shortened in this modern age, dominated by increasing computational power and emerging artificial intelligence?

Engineers and scientists are obsessed with modelling: they create models to better understand complicated physics. For some scientists this can be a goal on itself, as it allows to better understand the world we live in. But for engineers these models are developed with a clear focus to eventually improve products. This is particularly true for turbomachinery: Computational Fluid Dynamics (CFD) plays nowadays an essential role in the study of complex flow phenomena, which allows engineers to remedy efficiency losses by redesigning components. However, being able to model does not necessarily mean being able to improve. Engineers can spend large amounts of time using CFD models to improve their designs, as simulations do not provide clear guidance on how to improve their design. Additionally, the advent of more accurate models allows one to specify more goals and constraints, such that it becomes increasingly harder to meet all imposed targets. Paradoxically, while models have been introduced to improve products, and thus to progress with respect to old practices, they actually lead to a significantly more complex design problem with longer development cycles. There is thus an emerging need for optimization tools which assist the engineer in proposing design alterations to reach the specified targets.

To overcome these difficulties and speed up the design process, more and more engineers make use of optimization techniques. These techniques automate significant parts of the design process, and additionally use search algorithms to explore autonomously the vast design space. Various techniques exist, and arguably the most commonly used today is the Evolutionary Algorithm [1]. This technique is based on Darwin's principle of survival of the fittest, which actually mimics very well a golden rule of the market: products that are best adapted to the current requirements are more successful. Unsuccessful designs will thus rapidly be replaced by more successful ones and gradually the search leads towards the optimum design.

Evolutionary algorithms are however computationally expensive when rich design spaces need to be explored. This is not surprising: the current wealth of natural species required millions of years to evolve and adapt to changes in the environment. When applied to engineering problems, the slow convergence limits the use of these algorithms to only tens [2], or potentially hundreds of degrees of freedom [3], that is largely insufficient to explore interactions between different components. As such, these techniques have been mainly used to optimize the shape of one single blade row at a time, leaving a large potential for improvement unused.

Gradient-based optimization techniques, on the other hand, seem to provide a solution. These techniques rely on gradient information, i.e. on how the performance changes when infinitesimal changes to the design variables are made. Especially when the adjoint technique [4] is used to compute the gradient, the computational advantage is apparent: the cost of this computation is equivalent to that of one additional CFD computation, regardless the number of design variables. This allows for very rich design spaces at a very small computational cost [5].

Figure 1 shows the result of such additional adjoint computation for a turbine blade shape optimization: next to the Mach number distribution, obtained by the CFD computation, arrows on the profile wall indicate in which direction the profile needs to change to increase the total pressure losses. A reduction of the pressure loss is thus obtained by moving the blade wall opposed to the arrows indicated in the figure. Even when not introduced in an automated design optimization framework, the added information is crucial to designers, as it visually shows where to apply design changes. This technique is for instance used by car manufacturers to modify the exterior shape of the car for reduced drag coefficients.

Figure 1.Surface sensitivity with respect to pressure loss. Movement of the blade wall opposed to the arrows will reduce the pressure loss.

Grahic Jump LocationFigure 1.Surface sensitivity with respect to pressure loss. Movement of the blade wall opposed to the arrows will reduce the pressure loss.

Adjoint techniques for CFD have been around for over 3 decades, and although they are now commonly used for exterior aerodynamics, they are yet not commonly applied to turbomachinery, even though the potential is very widely recognized. The optimism of the early days has been replaced by the realism that turbomachinery is a more complex field, where large separation zones occur, where the shape parametrization is more complex, and where the interactions between different disciplines (aerodynamics, thermal, structural) are more pronounced. Many different researchers have however advanced the state-of-the-art in the past decades, such that today we are at the verge of introducing these techniques on an industrial level in the multi-disciplinary design of multistage machines with over 10,000 degrees of freedom. Over the years, issues of stability have been addressed to make the adjoint more stable for mildly unsteady processes such as separations [6]. Additionally, the multi-disciplinary character has been addressed where structural, vibrational and heat transfer sensitivities can now also be computed by the efficient adjoint method [7]. A significant effort has also been put on including CAD parametrizations in the computation of these sensitivities, as turbomachinery components are best parametrized using CAD-based models. Still, there are a number of issues to be addressed, but the progress has been steady and there is a large optimism now that these methods will be deployed in the industrial design process in the next decade. This is also clear from the introduction of adjoint methods in commercial CFD codes, which pick up the methodology as well.

Against common belief, optimizations of large scale problems, such as full multistage compressors or turbines, are within reach of current computational power. Now with the development of suitable methods reaching enough maturity, we will soon see the appearance of these techniques on the work floor. And it will not go unnoticed, as they will highly impact the development cycle of future gas turbine engines, enabling significant performance gains with much faster design processes.

References

Goldberg, D.E., Genetic Algorithms in Search, Optimization, and Machine Learning, Addison-Wesley Publishing Company, 1989
Verstraete, T. Alsalihi, Z. van den Braembussche, R.A. Multidisciplinary Optimization of a Radial Compressor for Micro Gas Turbine Applications ASME Journal of Turbomachinery, 132, 3, 2010
Siller, U. Voss, C. Nicke, E. Automated Multidisciplinary Optimization of a Transonic Axial Compressor 47th AIAA Aerospace Sciences Meeting, January 2009, Orlando Florida
Giles, M.B. Pierce, N.A. An introduction to the adjoint approach to design Flow, Turbulence and Combustion, 65, (3-4), 2000, pp. 393–415 [CrossRef]
Reuther, J. Jameson, A. Alonso, J.J. Rimlinger, M.J. Sauders, D. Constrained Multipoint Aerodynamic Shape Optimization Using an Adjoint Formulation and Parallel Computers Journal of Aircraft, Vol. 36, 1999, pp. 51–74 [CrossRef]
Xu, S. Radford, D. Meyer, M. Mueller, J.D. “Stabilisation of Discrete Steady Adjoint Solvers” Journal of Computational Physics, Vol. 299, 2015, pp. 175–195 [CrossRef]
Verstraete, T., Mueller, L. Mueller, J.-D. Multidisciplinary Adjoint Optimization of Turbomachinery Components Including Aerodynamic and Stress Performance. In 35 th AIAA Applied Aerodynamics Conference, AIAA AVIATION Forum, (AIAA 2017-4083) https://doi.org/10.2514/6.2017-4083
Copyright © 2019 by ASME
View article in PDF format.

References

Goldberg, D.E., Genetic Algorithms in Search, Optimization, and Machine Learning, Addison-Wesley Publishing Company, 1989
Verstraete, T. Alsalihi, Z. van den Braembussche, R.A. Multidisciplinary Optimization of a Radial Compressor for Micro Gas Turbine Applications ASME Journal of Turbomachinery, 132, 3, 2010
Siller, U. Voss, C. Nicke, E. Automated Multidisciplinary Optimization of a Transonic Axial Compressor 47th AIAA Aerospace Sciences Meeting, January 2009, Orlando Florida
Giles, M.B. Pierce, N.A. An introduction to the adjoint approach to design Flow, Turbulence and Combustion, 65, (3-4), 2000, pp. 393–415 [CrossRef]
Reuther, J. Jameson, A. Alonso, J.J. Rimlinger, M.J. Sauders, D. Constrained Multipoint Aerodynamic Shape Optimization Using an Adjoint Formulation and Parallel Computers Journal of Aircraft, Vol. 36, 1999, pp. 51–74 [CrossRef]
Xu, S. Radford, D. Meyer, M. Mueller, J.D. “Stabilisation of Discrete Steady Adjoint Solvers” Journal of Computational Physics, Vol. 299, 2015, pp. 175–195 [CrossRef]
Verstraete, T., Mueller, L. Mueller, J.-D. Multidisciplinary Adjoint Optimization of Turbomachinery Components Including Aerodynamic and Stress Performance. In 35 th AIAA Applied Aerodynamics Conference, AIAA AVIATION Forum, (AIAA 2017-4083) https://doi.org/10.2514/6.2017-4083

Figures

Tables

Errata

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In