3D multi-point/multi-objective optimisation tool coupling inverse design, CFD/FEA and multiple optimization schemes
- Using inverse design, optimization becomes dramatically quicker, easier and more reliable
- Less design parameters are needed to define the geometry
- All designs respect the specific work required
- Response surfaces are far smoother than for direct design
- Combined with the easy-to-use user interface of TURBOdesign Optima, full automatic optimization of designs becomes a reality in industrial timescales
- Includes both direct and response surface based optimization schemes
- Input files for coupling with external codes are output automatically
Capabilities of TURBOdesign Optima
TURBOdesign Optima is a flexible and powerful optimisation code for all types of turbomachinery, using the power of 3D inverse blade design from TURBOdesign1 to provide engineers with optimum results. Using inverse design rather than conventional direct geometry parameterisation for optimisation has many benefits which turn automatic optimisation from something to be used on only the most difficult or non-time critical cases to a part of the daily design process.For more details, see 'Benefits of inverse design-based optimisation' below.
TURBOdesign Optima is fully integrated with the TURBOdesign Suite tools and is designed to maintain the speed and ease-of-use users will experience from TURBOdesign Pre and TURBOdesign1. As for TURBOdesign1, Optima provides a drastic reduction in design time required compared to conventionaal codes. Time taken for design scales from:
- 2-3 minutes for a TURBOdesign1 manual iteration
- 2-3 hours for a single-point, multi-objective optimisation using TURBOdesign Optima's direct optimisation capability
- 2-3 days for a multi-point, multi-objective optimisation using TURBOdesign Optima's response surface optimisation capability coupled with an external CFD/FEA analysis code.
All significantly faster than could be achieved using conventional design methods.
Benefits of inverse design-based optimization
Many companies have attempted to or are introducing automatic optimisation to their design processes. Using conventional design methods to set up input parameters based on the blade geometry has however caused significant problems in many instances, with many hundreds of design points being required even for simple cases, and even this does not guarantee good quality optimisation outputs. This has led to most companies employing automatic optimisation just for specific research programs or with a limited remit. The use of 3D inverse design parameterisation via the blade loading distribution removes these problems and allows automatic optimisation to be the default design process in industry.These benefits can be grouped into 2 primary areas:
- Reduced computational cost
- Increased accuracy and applicability of optimisation results
Reduced computational cost:
- Parameterisation of the blade geometry: With direct design design, 30 to 100 parameters are typically required to parameterise a blade. Reducing the number of parameters dramatically reduces the design space to be explored. With inverse design a amximum of 8 parameters are required to define the full 3D blade via the blade loading distribution. This immediately provides a 4 to 12 reduction in the number of design points required.
- Consistency of specific work: With direct geometry parameterisation, all designs will produce varying specific work (pressure rise, head etc.). Therefore to ensure representative results from the optimisation, a constraint on work is required, which can lead to the majority of design points being rejected post-CFD, further increasing the number of points required. With the inverse method the specific work is an input, so all designs automatically meet this constraint.
- Identifying flow features: With direct design methods, CFD analysis must be run on each geometry to extract basic data such as the peak surface velocity or maximum diffusion. With the inverse method, the 3D inviscid flow-field at design point is produced automatically in each 15 second iteration. This means that design-point optimisations can be peformed in a matter of 2-3 hours rather than days or weeks.
Increased accuracy and applicability of optimisation results:
- Smoothness of blade surfaces: With direct geometric paramerisation, many geometries will be non-smooth across the span, meaning that they cannot be manufactured or wil have high stresses. With the inverse method, the output geometry is always smooth as the solver is fully 3D. Therefore all output geometries can be used in downstream processes.
- Shape of the objective function: The relationship between geometry and fthe resulting flow-field is highly non-linear. Therefore in response surface-based optimisations many design points are required to provide enough data points to ensure an accurate fit. With the inverse method, there is a much more direct relationship between input and output parameters, meaning that far fewer design points are required for a good fit. Typically, a 2-3x reduction in the number of designs for the same number of parameters can be used.
- Generality of optimised results: With direct design-based optimisation, the output is a blade geometry. Many studies have shown that this optimum shape will not be generally applicable for different conditions or even different sizes of the same design. With inverse design optimisation, the output is a blade loading distribution. Optimum blade loading distributions have been shown to have good generalisation across a wide range of operating conditions and machine sizes, see <papers>. As such, optimisation studies can also be used as know-how generators for future designs.
TURBOdesign Optima for direct optimisation
TURBOdesign Optima uses the NSGA-II multi-objective genetic algorithm for direct optimisation. The use of genetic algorithms is a very powerful method for finding optimum solutions in non-linear design spaces, but requires typically hundreds if not thousands of iterations to reach a converged optimum / establish a full Pareto front. With conventional design CFD analysis this is impractical, but with each TURBOdesign1 iteration taking 10-15 seconds to solve, it becomes easily achievable to run several optimisations in 1 day!
TURBOdesign Optima for response surface-based optimisation
TURBOdesign1, as a design code, only provides limited information on off-design performance. To allow this to be assessed TURBOdesign Optima also has response surface-based optimisation capability buitl-in, with support for Latin Hypercube sampling and Kriging response surface modelling. Using inverse design provides many benefits for response surface optimisation, as are explained in 'Benefits of inverse design-based optimisation'. TURBOdesign Optima will automatically produce the required input files for external CFD codes (eg. TurboGrid curve files), with the user manually meshing and analysing cases. For a fully integrated solution using ANSYS TurboGrid and CFX, see the pages for TURBOdesign WB.