Page 88 - IJAMD-1-3

International Journal of AI for
Materials and Design
Review of gas turbine blade failures by erosion

1 N i () i () 2 • FEA simulation: The FEA model calculates the stress
L()θ = E true ( ∑ − E pred) (XI) distribution, deformation, and failure points based
N i=1 on the applied particle forces and material properties.
where E and E represent the actual and predicted Use the impact forces from CFD to simulate the
pred
true
erosion values, respectively. CNNs efficiently process structural response of the blade, solving for stresses
complex geometries and predict erosion distribution with and deformations.
high accuracy. The ML models are trained by minimizing • Iterative coupling: In more advanced simulations, the
the loss function L(θ) using optimization techniques such deformed geometry from FEA is fed back into CFD to
as stochastic gradient descent. The training data are used update the flow field and particle trajectories, creating an
to adjust the model parameters, while the validation set iterative loop that provides a more accurate representation
monitors overfitting. After training, the test set evaluates of erosion progression and structural failure. Apply von
the model’s performance, ensuring its accuracy in Mises stress criteria and fatigue models (Paris Law) to
predicting erosion rates under unseen conditions. The predict crack growth and blade failure.
trained ML models enable real-time prediction of erosion Molinari and Ortiz applied multi-scale models to
44
rates and severity. These predictions help to identify high- simulate how micro-scale erosion affects the macro-scale
risk zones on turbine blades, allowing engineers to develop performance of gas turbine blades. By combining CFD and
targeted interventions. For example, ML can predict areas FEA, they were able to track how surface roughness from
requiring enhanced coatings or geometric optimization to erosion influences the fluid dynamics (e.g., increased drag)
minimize erosion effects. In addition to erosion prediction, and how this further exacerbates structural deformation.
decision-tree-based models and reinforcement learning Their approach also captured the cascading effects of erosion,
are applied to optimize material selection. These models where small-scale material removal leads to altered flow
analyze material properties such as hardness (H), tensile patterns, increasing turbulence and intensifying localized
strength (TS), and thermal resistance (T ) to identify wear. By coupling CFD and FEA, the theoretical framework
max
materials with the highest erosion resistance. By narrowing provides a comprehensive approach to predicting erosion
down material options through ML, researchers can focus damage and structural failure. This integration allows for
on testing and deploying the most promising candidates. a detailed analysis of the erosion-fatigue interaction and

This structured integration of ML provides an efficient, the long-term degradation of turbine blades, offering
data-driven approach to understanding and mitigating actionable insights for improving blade design, material
erosion in gas turbine blades, paving the way for enhanced selection, and maintenance schedules. Furthermore, the
durability and performance. study highlighted how localized stress concentrations from
erosion propagate cracks, emphasizing the need for more
3.5. Coupling CFD and FEA for fluid-structure robust materials and coatings in high-impact zones. The
interaction findings also showed that modifications to blade geometry,
The most powerful approach in gas turbine analysis is such as smoothing sharp edges or optimizing surface
the coupling of CFD and FEA to simulate fluid-structure profiles, can significantly reduce stress amplification
interaction. This combined method allows engineers to and delay failure. These findings underline the critical
model both the fluid flow and particle dynamics and the importance of coupling high-fidelity simulations with
structural response of the blade. 12,16,18 This coupled CFD- experimental validation to ensure practical applicability in
FEA approach offers a comprehensive way to predict real-world turbine operations. Moreover, this multi-scale
erosion and failure points, optimizing blade geometry and approach bridges the gap between theoretical models and
materials. The following are the key steps in CFD-FEA industrial applications, enabling engineers to proactively
coupling: address failure risks. The study further demonstrated that
• CFD simulation: The Navier–Stokes equations are integrating these models into predictive maintenance
solved to obtain the airflow and particle impact systems can enhance operational reliability by identifying
forces on the blade surface. Solve the Navier-Stokes critical wear zones before catastrophic failures occur. The
equations to predict airflow around the turbine blade entire CFD-FEA for fluid-structure interaction is depicted
and use the Eulerian–Lagrangian approach to track using a flow diagram (Figure 1).
particle trajectories.
• Transfer of impact forces: The particle impact forces 3.6. Surrogate models and ML for real-time
from CFD are transferred to the FEA model as prediction
external loads. Determine particle velocities, angles, As CFD and FEA simulations can be computationally
and impact locations. expensive, ML is increasingly used to create surrogate

Volume 1 Issue 3 (2024) 82 doi: 10.36922/ijamd.5188