Machine learning prediction of erosion resistance of laser-clad coatings on martensitic stainless steel for steam turbine blades

Abstract

Steam turbine blades made of martensitic stainless steel (SS 420) are susceptible to erosion due to the impact from the water droplets. Laser-directed energy deposition (L-DED) process can enhance erosion resistance of steam turbine blades by laser-clad protective coatings, but the process intricacies arising from multiple parameters require some predictive modelling. This study developed machine learning models to predict surface roughness (SR), cumulative mean depth of erosion (CMDE) and hardness of SS 420 substrates coated by L-DED process with Ni-Al, Ni-Al-TiC and Stellite 6 powders. The input parameters were laser power, scanning speed, powder flow rate, energy density, time, initial temperature, feeding rate and distance from the surface. Seven algorithms, namely, Linear Regression, Decision Tree, Random Forest, Gradient Boosting (GB), K-Nearest Neighbours, AdaBoost and XGBoost, were trained on the experimental data. Cross-validation assessed the model

accuracy. XGBoost model showed the lowest mean absolute error and the highest R ² for predicting SR. GB was the most accurate for CMDE and hardness. By demonstrating accurate prediction of the properties of L-DED coating using machine learning, this study provides a framework to optimize the L-DED process for enhanced erosion resistance of steam turbine blades. Further research can assess model performance for new parameter combinations and materials.