Abstract
This paper presents two efficient reduced-order modelling (ROM) techniques for predicting the forced response statistics of bladed disc assemblies. The first method, a Pass-band based Reduced Order Model (PBROM), uses component modes of a sector as basis vectors, reducing the model size to the number of sectors. The second method, a Stochastic Reduced Basis approach, represents the frequency-domain response using a linear combination of stochastic basis vectors spanning a preconditioned Krylov subspace. Both techniques enable efficient Monte Carlo simulations, providing accurate statistical predictions at a computational cost comparable to simple mass-spring systems.
Introduction
In jet engine design, turbine blades are ideally identical, but manufacturing tolerances and material variations—known as mistuning—are unavoidable. These small differences can lead to "mode localization," where vibration energy is trapped in specific blades, causing unexpected stress and fatigue. To ensure reliability, engineers must perform statistical assessments using thousands of simulations. However, high-fidelity Finite Element Models are too computationally expensive for this task. This paper introduces two ROM techniques to overcome this bottleneck.
Methodology
The researchers developed two strategies to simplify the complex dynamics of mistuned systems:
Pass-band based Reduced Order Model (PBROM)
PBROM utilizes the vibration modes of a single, perfectly tuned sector as a foundation. By projecting the massive system of equations onto these basis vectors, the model size is drastically reduced to match the number of blades. This allows for rapid pass-band calculations and efficient statistical sampling.
Stochastic Reduced Basis Methods (SRBMs)
SRBMs simplify the problem by focusing on the randomness itself. The system response is represented through stochastic basis vectors in a Krylov subspace. A Stochastic Bubnov-Galerkin scheme is then used to solve for unknown coefficients, yielding direct estimates of the vibration mean and variance.
Results and Discussion
The methods were validated using a 10-beam discretized model across weak, moderate, and strong coupling scenarios with a 5% random variation in blade stiffness.
Accuracy and Robustness
The PBROM method proved highly effective, particularly in weakly coupled systems where mode localization is dominant. As shown in the frequency analysis, the PBROM predictions align almost perfectly with the exact mistuned results, capturing the shifts that a standard tuned model ignores.
While PBROM excels in high-density modal regions, SRBMs offer greater robustness for strongly coupled systems where global disc behavior is more prevalent. Both methods provide results nearly indistinguishable from exact Monte Carlo simulations but at a fraction of the cost.
Conclusion
The proposed ROM techniques significantly reduce the computational burden of predicting mistuning effects. By enabling fast and accurate statistical analysis, these methods allow for more robust engine component designs. Future research will explore hybrid approaches, such as Stochastic Component Mode Synthesis, to handle even more complex industrial models.
Summary updated