Statistical Methods for Turbine Blade Dynamics
Abstract
A novel statistical methodology has been designed aimed at quantification of uncertainties embedded in stress/strain prediction of structural performance for turbine engine blades. The developed methodology relies on Bayesian network representation of existing model-based stress/strain predictions that use both analytical/numerical modal information and experimental forced-response data. It accounts in a statistically consistent fashion for the evidence provided both at the input and output levels of the model. The need to incorporate the evidence at the output level is not unique to the considered problem, so the results can be applicable to other fields, such as system identification and model calibration. In addition, various sources of uncertainty were analyzed, namely component geometric and material properties, sensor-induced and data-processing errors. Simulation of uncertainty propagation and sensitivity analyses were conducted to identify the main uncertainty contributors and provide the guidelines for simplifying the Bayesian network. Assessment of the proposed framework and identification of its benefits and drawbacks were performed using experimental data for plate structures obtained using the industry-standard testing procedures for turbine engine blade analysis.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 30, 2008
- Accession Number
- ADA636052
Entities
People
- Giorgio M. Calanni-fraccone
- Massimo Ruzzene
- Vitali Volovoi
Organizations
- Georgia Tech