Mistuning Identification of Bladed Disks Utilizing Neural Networks

Date

2010-6-14

Author

Özgüven, Hasan Nevzat
Ciğeroğlu, Ender
Yümer, M. Ersin

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

63
views

11
downloads

Cite This

Mistuning affects forced response of bladed disks drastically; therefore, its identification plays an essential role in the forced response analysis of realistic bladed disk assemblies. Forced response analysis of mistuned bladed disk assemblies has drawn wide attention of researchers but there are a very limited number of studies dealing with identification of mistuning, especially if the component under consideration is a blisk (integrally bladed disk). This paper presents two new methods to identify mistuning of a rotor from the assembly modes via utilizing neural networks. It is assumed that a tuned mathematical model of the rotor under consideration is readily available, which is always the case for today’s realistic bladed disk assemblies. In the first method, a data set of selected mode shapes and natural frequencies is created by a number of simulations performed by mistuning the tuned mathematical model randomly. A neural network created by considering the number of modes, is then trained with this data set. Upon training the network, it is used to identify mistuning of the rotor from measured data. The second method further improves the first one by using it as starting point of an optimization routine and carries out an optimization to identify mistuning. To carry out identification analysis by means of the proposed methods, there are no limitations on the number of modes or natural frequencies to be used. Thus, they are suitable for incomplete data as well. Moreover, since system modes are used rather than blade alone counterparts, the techniques are ready to be used for analysis of blisks. Case studies are performed to demonstrate the capabilities of the new methods, using two different mathematical models to create training data sets; a lumped-parameter model and a relatively realistic reduced order model. Throughout the case studies, the effects of using incomplete mode families and random errors in assembly modes are investigated

URI

https://hdl.handle.net/11511/105197

Collections

Department of Mechanical Engineering, Conference / Seminar