This work continues the recently intensified research aimed to propose efficient methods for the optimization of industrial turbine components under the presence of uncertainties. Improvement and automation of the multi-disciplinary simulation process combined with optimization and optimization under uncertainty (OUU) techniques should bring considerable reduction of development time, while improving the design. The first deliverable of this work is a new multi-disciplinary analysis (MDA) chain, which includes aerodynamic, thermal and structural analysis followed by fatigue life prediction. Accurate three-dimensional transient thermal analysis is introduced, to the knowledge of the author, for the first time into the automatic MDA chain within this work. The transient thermo-mechanical analysis results into a significant increase of computational costs, which limits the total number of system evaluations for further deterministic optimization and OUU applications. This motivated the development and implementation of efficient optimization and OUU methods, capable of solving expensive engineering problems within a very limited amount of system evaluations.
In this work, robustness of an objective function together with reliability with respect to design constraints are requested simultaneously in order to consider the design as robust, leading to the concept of Reliability-Based Robust Design Optimization (RBRDO). Special attention is paid to efficient adaptive surrogate-based methods, aimed at solving computationally demanding engineering problems. A novel Efficient Global Optimization with Performance Measure Analysis (EGOPMA) method for inverse reliability analysis with the help of Gaussian process surrogates is proposed to be used within the RDO framework. The method is capable to obtain the first order Probabilistic Performance Measures (PPM) for non-linear functions within a very limited number of system evaluations. This allowed formulating the novel RBRDEGO method, which combined efficiency and global properties of the EGO method for optimization with the proposed EGOPMA for reliability assessment and sampling-based robustness quantification. Special attention is paid to handling noisy simulation-based results and missing data from failed simulations as well as to the ability of the method to exploit parallelism of an industrial computational environment. Comparison with the latest state-of-the-art methods provided validation of the RBRDEGO method and indicated its high efficiency. The method was successfully applied to realistic high-dimensional simulation-based problems, such as aerodynamic 3D shape of a single vane.
Finally, a synergy of the main deliverables of this research enabled a multi-disciplinary reliability-based robust design optimization of the LPT vane cluster with the help of the developed MDA process chain and proposed RBRDEGO. The optimization of the vane design led to a significant reduction of the stage losses, while respecting all the multi-disciplinary constraints. RBRDO allowed to include effects of shape and loads uncertainty into the optimization, so that an obtained optimal design also satisfies robustness requirements. The optimum was validated using Monte-Carlo random sampling, which confirmed the robustness of the design and the accuracy of the method.
«
This work continues the recently intensified research aimed to propose efficient methods for the optimization of industrial turbine components under the presence of uncertainties. Improvement and automation of the multi-disciplinary simulation process combined with optimization and optimization under uncertainty (OUU) techniques should bring considerable reduction of development time, while improving the design. The first deliverable of this work is a new multi-disciplinary analysis (MDA) chain,...
»