Development of AI-driven Prediction Models for Alloy Corrosion Resistance

Thesis Abstract

Corrosion of alloys presents a significant challenge in the materials engineering domain, leading to economic losses and safety concerns in sectors such as aerospace, marine, and infrastructure. Traditional methods for predicting alloy corrosion resistance rely heavily on empirical testing and conventional statistical models, which are often time-consuming, resource-intensive, and limited in predictive accuracy across diverse environmental conditions. The primary aim of this study is to develop robust and accurate artificial intelligence (AI)-driven prediction models that can reliably forecast alloy corrosion resistance based on compositional, microstructural, and environmental parameters. To achieve this, the study formulates specific objectives (1) to analyze the influence of compositional and environmental factors on alloy corrosion through empirical data; (2) to design and train various machine learning algorithms, including random forest, support vector machines, and neural networks, for corrosion prediction; (3) to evaluate and compare the predictive performance of these models; and (4) to establish an optimized AI-based framework for corrosion susceptibility assessment. The research adopts a quantitative, exploratory research design involving the collection of experimental corrosion data from 150 alloy samples subjected to controlled laboratory environments mimicking real-world conditions such as salt spray, humidity, and temperature variations. The population consists of commercially significant alloys with diverse alloying elements, microstructures, and surface treatments. Stratified random sampling ensures representative selection of alloy types and environmental conditions. Data collection is facilitated through standardized corrosion testing protocols, including electrochemical impedance spectroscopy, potentiodynamic polarization, and gravimetric analysis, complemented by detailed compositional analysis via energy-dispersive X-ray spectroscopy (EDS) and microstructural characterization using scanning electron microscopy (SEM). The validity and reliability of the collected data are ensured through calibration of instruments, repeat testing, and adherence to ASTM corrosion testing standards. Analytical techniques employed include feature engineering to preprocess input variables, with exploratory data analysis applying correlation matrices and principal component analysis (PCA) to discern key influencing factors. The core modeling employs supervised machine learning techniques, with model training and validation performed via k-fold cross-validation to prevent overfitting. Model performance is assessed using metrics such as R-squared, root mean square error (RMSE), mean absolute error (MAE), and receiver operating characteristic (ROC) curves. Further, hyperparameter tuning is conducted through grid search optimization to enhance model accuracy. The study also explores the integration of ensemble learning methods to improve predictive robustness. Expected findings include the identification of critical alloying elements and environmental factors influencing corrosion resistance, as well as the development of AI models that outperform traditional statistical approaches in predictive accuracy and generalizability. It is anticipated that neural network models will demonstrate superior performance due to their ability to capture complex nonlinear relationships in data. The research aims to produce an adaptable, user-friendly AI framework capable of providing rapid corrosion resistance evaluations, significantly reducing reliance on lengthy experimental procedures. This study contributes new knowledge by bridging materials science and artificial intelligence, offering a novel predictive tool that enhances decision-making in alloy design and maintenance planning. It extends current understanding of corrosion modeling by applying advanced machine learning algorithms to diverse alloy-environment combinations, addressing existing gaps in predictive capability and scalability. The main conclusion emphasizes that AI-driven models can revolutionize corrosion prediction protocols, leading to cost-effective, accurate, and sustainable corrosion management strategies. Based on these findings, recommendations include the integration of the developed models into industrial corrosion monitoring systems and the expansion of datasets to encompass other alloy systems and environmental conditions. Future studies are suggested to explore real-time adaptive learning systems and the incorporation of additional variables such as mechanical stress and coating integrity for comprehensive corrosion prediction frameworks.