Page 87 - IJAMD-1-3

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

performance by creating a laminar boundary layer that have been shown to reduce drag coefficients by 10 – 12%,
helps reduce particle deposition and subsequent erosion. indirectly lowering erosion rates by altering particle
The elliptical shape provides a balance between lowering trajectories. 43,49,51
drag and maintaining airflow smoothness. In summary, geometrical modifications to turbine
The trailing edge, another critical zone, is prone to blades, including adjustments to the leading and trailing
localized flow separation caused by the abrupt deceleration edges, surface contouring, and blade profile optimization,
of airflow. Streamlining the trailing edge can mitigate this are effective strategies for mitigating erosion. These
issue. A tapered trailing edge design ensures a smoother changes not only enhance the blade’s resistance to erosion
exit flow by reducing the abrupt changes in velocity but also improve aerodynamic performance by optimizing
that cause vortices to form. These vortices are often fluid flow. By implementing such modifications, engineers
laden with particles that can repeatedly collide with the can design more durable blades, ultimately extending their
blade surface, exacerbating erosion. Another innovative operational lifespan and reducing maintenance costs. CFD
approach involves the use of chevron patterns or serrations and FEA simulations are critical for identifying erosion-
at the trailing edge. These features disrupt large vortices prone areas and predicting structural failures. These tools
into smaller, less erosive structures, improving pressure help engineers design more durable turbine blades by
recovery, and minimizing turbulence. The result is reduced optimizing geometries and selecting materials that can
particle impingement in this region, extending the blade’s better withstand erosive forces.
operational life.
3.4.6. ML integration
Surface contouring can also play a key role in managing
boundary layer behavior and controlling the interaction ML plays a critical role in enhancing the prediction of
between fluid flow and solid particles. For example, a erosion rates, optimizing material selection, and enabling
concave adjustment along the blade’s midsection can proactive maintenance strategies in gas turbine systems.
stabilize the boundary layer, reducing turbulence and This section elaborates on the methodology, mathematical
particle collisions. This stabilization encourages smoother frameworks, and applications of ML in the study.
streamlines, preventing the reattachment of turbulent flows The first step in ML integration involves preparing a
that can cause concentrated erosion. Similarly, micro- robust dataset. Input parameters such as particle velocity
grooved surfaces – small, precisely engineered grooves (v), impact angle (θ), particle size (d), material hardness
along the blade – help channel airflow in a controlled (H), and operational temperature (T) are identified as
manner. These grooves not only reduce drag but also divert key features influencing erosion. These features are used
particles away from high-stress zones, further minimizing to predict output labels, which represent erosion severity
erosion. metrics such as material loss per unit area (E) or erosion
Other geometrical changes, such as twisting the blade rate. The dataset is generated from a combination of CFD
profile or adjusting the lean of the blade tips, help optimize and FEA simulations, historical operational data, and
particle trajectories and airflow distribution. A twisted blade experimental results. Splitting the dataset into training,
profile ensures that the flow remains more uniform along validation, and testing sets ensures the model’s reliability
the blade’s length, preventing the formation of localized and generalizability. The study employs supervised
zones of high velocity where particles can concentrate. This learning algorithms to predict erosion rates. Models such
even distribution of flow reduces the severity of erosion as SVMs or RF classify erosion severity (E , E moderate ,
low
by spreading it more evenly across the surface. Similarly, E ) or predict continuous values of erosion rate (E). For
high
backward-leaning blade tips are designed to align the flow instance, regression models map relationships between
more effectively with the blade geometry, reducing the input features and erosion rate using Equation X:
formation of eddies at the tips, which are known to cause E = f (v, θ, d, H, T) + ϵ (X)
erosion-prone turbulence.
where f is the function approximated by the ML model;
CFD simulations have been instrumental in validating and ϵ is the error term. In addition, feed forward NNs
the benefits of these geometrical modifications. For with activation functions like ReLU (g(x) = max(0,x)) are
instance, studies have shown that a smoother leading edge used to capture non-linear relationships between features
can reduce turbulence intensity by up to 20%, significantly and erosion severity. For spatially dependent data such
decreasing the velocity of particle impacts. Similarly, as particle impact distribution maps, CNNs are utilized.
serrated trailing edges have been found to reduce wake CNNs analyze input images or grid data to predict erosion-
turbulence by 15%, which helps maintain the integrity prone zones. The loss function for CNN training, typically
of erosion-resistant coatings. Micro-grooved surfaces the Mean Squared Error, is calculated using Equation XI:

Volume 1 Issue 3 (2024) 81 doi: 10.36922/ijamd.5188