Parametric CFD models aid cooling-nozzle design

Enginsoft UK has used a novel methodology to design the layout of the tip cooling-nozzles of a high-pressure rotor-blade turbine.

The methodology uses a complete CAE approach, by means of a parametric CFD model that is run many times so an optimiser can explore several different designs.

This is why the design is carried out automatically by parallel computations, with the optimisation algorithms taking the decisions rather than the design engineer.

The engineer can decide upon the physical settings of the CFD model, the number and the extension of the geometrical parameters of the blade tip holes and the optimisation algorithms.

The optimiser generates a database of simulated tip-cooling configurations, allowing the designer to find laws, functions and correlation between input parameters and performance output.

This study is part of an AVIO project concerning the development of high-pressure turbine blades with advanced cooling systems.

Because of the high gas temperatures entering the turbine of the most recent aero-engines, (in general up to 2000K at the turbine inlet at 40bars) an efficient cooling system is needed to maintain the metal temperatures below the allowable limits.

This means using cold air extracted directly from the compressor, with a significant negative impact on the engine performance.

One of the most critical areas from a thermal point of view is the tip region of the un-shrouded rotor blades.

Tip regions are generally cooled using rotor internal-air ejected in the flow path through a series of small holes located in the tip surfaces.

The ejected air must cover all the surfaces to create a cold film between the hot gas and the metal.

As the tip region is characterised by a complex 3D flow field, it is difficult to optimise the cooling system using the standard design methodologies, also considering the other blade tip requirements such as minimising the hot leakage air from pressure to suction side, which has a negative impact on turbine aerodynamic efficiency.

For these reasons, the area of the tip is investigated with a parametric CFD approach: a parametric model is run several times, guided by an optimisation algorithm, such that an optimal solution in terms of performance can be found.

This kind of approach requires linking an optimisation software (modeFRONTIER) to a 3-D CFD code (ICEM-CFX5) with the aim of finding the optimal values of some geometrical parameters of the tip area of the high-pressure rotor blade, to achieve certain performance objectives.

Using interpolators or expert system techniques become compulsory if a 3-D fluid-dynamic optimisation has to be approached.

Several methods are generally available within optimisation software: for example RSM and ANN.

In this case, an ANN method was chosen because of the nonlinearity of the system.

This way, after a preliminary series of CFD analyses and after the estimation of ANN, the 3-D CFD model can be substituted by a series of mathematical functions and the computational time is considerably reduced.

The expert system, represented by an ANN, must be introduced after a fair number of analyses are run, such that the expert system is reliable.

The error of the expert system is a known value and is the parameter that yields the accuracy of the interpolator relative to the database of real experiments so far acquired.

It is up to the designer to choose the threshold error value of his expert system.

The more CFD analyses that are run, the more trained and the more accurate the expert system becomes.

A parametric batch procedure allows the automatic creation of different geometrical models, the mesh generation and the CFD analyses of the blades.

Preliminary CFD simulations are planned and a screening is performed to build an input-output database.

The optimiser calculates ANN co-efficients for the two layers.

An MOGA algorithm investigates runs with further CFD virtual analysis, exploring the space of possible solutions on the ANN.

A virtual optimisation of the cooling system is carried out without further expensive CFD analysis.

The best virtual solutions are selected and the ANN virtual solutions are validated by a 'real' CFD analysis.

More accurate Neural Nets can now be estimated with a larger database.

The virtual optimisation can be executed again and new performing designs can be found.

This procedure is repeated until the desired convergence to the set of optimal solutions is achieved.

Eventually the optimiser finds a layout of tip-cooling nozzles, which are then validated by a CFD analysis.

The final design chosen proved to transfer the same amount of heat but with a reduction of about 16 per cent of the cooling air required.

Back to top