| Summary: | Chemical attack is one of the most significant issues affecting porous ceramic systems employed as membranes for separation technologies, which necessitate frequent system reliability testing. In this work, the non-linear predictive power of a hybridized machine learning prediction model, specifically Jaya-XGBoost to predict the corrosion-induced mass loss of monolithic and nickel-reinforced porous alumina ceramics has been examined. This study demonstrates the mass loss of monolithic and Ni-reinforced porous alumina developed using rice husk and sugarcane bagasse in acidic and alkaline corrosive media. Based on empirical findings, the formation of a very stable Ni3Al2SiO8 spinelloid phase in the RH-graded composites increased their chemical stability in the corrosive environments compared to their monolithic and corresponding SCB-graded counterparts. Corrosion testing data of these specimens were collected and fitted into both XGBoost and Jaya-XGBoost machine learning algorithms. The results showed that the Jaya-XGBoost model performed better in predicting the corrosion-induced mass loss of both the monolithic and the nickel-reinforced porous alumina than the regular XGBoost model in terms of statistical accuracy measures. The Jaya-XGBoost model developed in this study effectively predicted the mass loss in NaOH (R2 = 0.9984; MAE = 0.0168) and mass loss in H2SO4 (R2 = 0.9824; MAE = 0.0217) of the monolithic and nickel-reinforced porous alumina. The precision that can be obtained by modifying hyper-parameters with the Jaya method, combined with the well-known accuracy of XGBoost, renders the proposed model novel. © 2024 The Authors
|
|---|