Page 84 - IJAMD-1-3

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

distributions, guiding the design of more erosion-resistant analysis with FEA for structural response, may take several
blades. This coupling allows for iterative analysis, where hours or days on high-performance computing systems. To
structural deformation feedback refines fluid dynamics train an ML surrogate model, CFD and FEA simulations
models, ensuring a more accurate representation of real- are performed on a limited number of turbine blade
world conditions and improving blade performance under configurations under varying conditions (e.g., particle
harsh environments. Table 11 presents and describes some velocity, impact angle, and material properties). The
of the advanced coating systems for turbine blades. outputs, such as erosion rate distributions and stress levels,
are recorded for each scenario. An ML algorithm, such as
3.3.1. FEA application in erosion-induced fatigue a Gaussian Process Regressor (GPR) or a deep NN (DNN),
By applying the impact forces calculated from CFD, FEA is trained on the dataset to approximate the relationship
simulates how erosion-induced material loss affects the between input parameters and outputs. For example, given
stress distribution across the blade surface. For example, inputs such as particle velocity (v), angle of impact (θ), and
Kedir et al. showed that as material is eroded away, the material hardness (H), the model predicts erosion rates (E)
78
remaining blade material experiences higher localized and high-stress zones using Equation IX:
stresses, leading to plastic deformation and crack E = f (v, θ, H) + ϵ (IX)
formation. FEA models simulate cyclic loading that gas ML
turbine blades undergo during operation. By combining where f is the surrogate model function approximated
ML
this with the Paris–Erdogan crack growth law, FEA predicts by the ML algorithm; and ϵ is the residual error between
the remaining useful life of the blade by calculating how simulated and predicted values. The trained model is
quickly cracks will propagate from erosion-initiated validated using a separate test set to ensure its predictive
defects. In high-temperature environments, FEA can also accuracy. Metrics such as Mean Squared Error or R-squared
model creep deformation, which occurs alongside erosion. (R2R^2R2) evaluate the surrogate model’s performance.
32
Prashar et al. demonstrated that FEA can simulate the Example validation results include:
creep strain accumulation in areas where erosion has • R score: 0.98 (indicating high correlation between
2
reduced the blade thickness, predicting when and where predicted and simulated values).
the blade will fail due to a combination of thermal stresses, • Computational time reduction: Surrogate models
erosion, and mechanical fatigue. In our framework, FEA reduced prediction time from 8 h per simulation to <1 s.
simulations are essential for understanding how erosion- The surrogate model is applied to predict erosion
induced damage translates into structural failure. Once patterns for new operational conditions without running
CFD has identified erosion-prone areas, FEA can analyze full CFD or FEA simulations. For instance:
how material loss affects stress distribution and predict the • Input: Particle velocity = 400 m/s; impact angle = 70°;
crack initiation and growth due to cyclic loading, fatigue,
and creep. material hardness = 500 MPa.
• Output: Predicted erosion rate distribution on the
3.3.2. Case example: Predicting erosion patterns using blade surface, with high-risk zones highlighted.
ML surrogate models Surrogate models eliminate the need for repeated high-
High-velocity particle impacts on turbine blades lead fidelity simulations, reducing the computational expense
to complex erosion patterns that require analysis with by several orders of magnitude. Engineers can quickly test
extensive CFD and FEA simulations. A single simulation multiple design modifications (e.g., blade geometries and
cycle, integrating CFD for airflow and particle trajectory materials) to identify configurations with minimal erosion

Table 11. Advanced coating systems for turbine blades

Coating system Composition Application technique Advantages Limitations
TBCs Ceramic topcoat with Plasma spraying or High thermal resistance Degrades under sustained
MCrAlY bond coat electron beam PVD and insulation erosion and spallation
Anti-corrosion coatings Metal oxide or nitride Thermal spraying or dip Protects against oxidation Limited erosion resistance
layers coating and chemical degradation without additional layers
Environmental barrier Silicon-based ceramics Slurry or chemical vapor Protects against oxidation High cost and limited thermal
coatings deposition and hot corrosion fatigue resistance
Wear-resistant coatings Carbides (e.g., tungsten HVOF spraying High hardness and wear Susceptible to thermal cracking
carbide) resistance at high temperatures
Abbreviations: HVOF: High-velocity oxygen fuel; PVD: Physical vapor deposition; TBCs: Thermal barrier coatings.

Volume 1 Issue 3 (2024) 78 doi: 10.36922/ijamd.5188