Non-Smooth Optimisation Applied to Improving Steam Turbine Designs
- Researcher: Jonathan Grant-Peters
- Academic Supervisor: Raphael Hauser
- Industrial Supervisor: Tobias Ruehle, Peter Gerkens
Background
Siemens, being a designer and manufacturer of steam turbines, wish to design their turbines to be as efficient as possible. Having already designed a computational model to predict the efficiency of a steam turbine given its design, all that remains to identify the design features which result in the turbine being as efficient as possible. This is a problem for mathematical optimisation, where the objective function is the efficiency of the steam turbine as computed by the previously mentioned computational model.
Outcomes
The main obstacle we face when trying to solve this problem comes from the fact that the objective function is not continuously differentiable. Because of this, most existing software is not applicable to solving this problem. On the other hand, any existing software which is suitable in theory takes too long to run.
Figure 1: Non-smooth optimisation.
Therefore, in this project, we have developed a new optimisation algorithm, suitable for non-smooth functions, which is designed to take advantage of the unique features observed within Siemens’ objective function. Part of this method involves a new exact line-search algorithm, again of our design, which performs notably faster than existing robust methods and is more robust than existing fast methods for the 1-D non-smooth optimisation problem.