Session: 26-02 Applications using Probabilistic and Machine Learning Methods

Paper Number: 122477

122477 - Possibilities of Applying Machine Learning Techniques for Probabilistic Analysis in a Turbine Blade-Disk Interface 

The design of a blade-disk interface in a jet engine is crucial, as the stress on both disk and blade can affect the engine's service life. Any variation in design parameters due to manufacturing tolerances can have a significant impact on the life of the components. While finite element (FE) analysis is commonly used for this purpose, surrogate models can be a much more effective alternative in such cases, as long as they maintain the desired quality.

This paper addresses the challenge of modeling a high-dimensional, parameterized finite element (FE) model of a blade-disk interface and uses deep neural networks (DNNs) as surrogate models to approximate the system behavior. Probabilistic methods, especially Monte Carlo simulations, are used for training, and the main goal is to evaluate the efficiency of using DNNs for this purpose.

In applications such as FE analysis, where interactions between neighboring nodes are essential, DNNs are advantageous due to their ability to process multiple nodes simultaneously. They can adapt to different mesh topologies, eliminating the need for topologically identical positions, in contrast to conventional approaches such as polynomial regression. This advantage of DNNs is demonstrated in this work by a comparative analysis of geometry variation due to manufacturing tolerances and its impact on the stress values using an internally developed probabilistic tool. With this tool, components can be predesigned by means of surrogate models. In addition, the ability of using DNNs to perform sensitivity analysis is evaluated. Furthermore, the conditions for the effective use of this method for robust optimization and the feasibility to determine failure probabilities are discussed.

Presenting Author: Elmira Emmrich Technische Universität Dresden

Presenting Author Biography: 2007-2011 - Bachelor of Engineering in Mechanical Engineering from Guilan University, Iran
2013-2017 - Master of Science in Mechanical Engineering at TU Berlin, Germany
2017-2020 - Development Engineer at IMK engineering GmbH, Germany
since 2020 - Research Associate at the Chair of Turbomachinery and Flight Propulsion, TU Dresden, Germany

Authors:

Elmira Emmrich Technische Universität Dresden
Md Tarekul Islam Technische Universität Dresden
Matthias Voigt Technische Universität Dresden
Lukas Bruder MTU Aero Engines AG
Julian Von Lautz MTU Aero Engines AG
Ronald Mailach Technische Universität Dresden