A Framework for Corrosion Resistance Prediction in Industrial Alloys Using Machine Learning

Thesis Abstract

Corrosion-induced degradation in industrial alloys remains a critical challenge, leading to significant economic losses and safety concerns across various sectors such as oil and gas, aerospace, and manufacturing. Despite advancements in alloy engineering and protective coatings, predicting corrosion resistance accurately remains complex due to the multifaceted interactions between environmental factors, alloy composition, and operational conditions. This study aims to develop a comprehensive machine learning-based framework for predicting corrosion resistance in industrial alloys, thereby enhancing preventative maintenance strategies and material selection processes. The specific objectives include identifying key predictors of corrosion resistance through feature selection, developing and validating predictive models using diverse machine learning algorithms, and establishing a practical, user-friendly framework for industry application. The research adopts a quantitative, cross-sectional design to analyze experimental and field data on corrosion behavior. The study population comprises 120 industrial alloy samples obtained from certified corrosion testing laboratories and manufacturing plants, representing various alloy classes such as stainless steels, nickel-based alloys, and titanium alloys. A stratified random sampling technique was employed to ensure representative sampling across different alloy categories and environmental conditions. Data collection involved standardized corrosion testing methods including electrochemical impedance spectroscopy (EIS), potentiodynamic polarization, and salt spray testing, complemented by detailed recording of alloy composition, surface treatments, operational environments, and climatic factors. To ensure data quality, instruments were calibrated regularly, and measurement reproducibility was verified through repeated tests. Data analysis integrates descriptive statistics, correlation analysis, and advanced machine learning techniques. Feature engineering and selection leverage techniques such as recursive feature elimination and principal component analysis (PCA) to identify the most significant predictors influencing corrosion resistance. Supervised learning algorithms—including Random Forests, Support Vector Machines (SVM), Gradient Boosting Machines (GBM), and Artificial Neural Networks (ANN)—are evaluated based on accuracy, precision, recall, and Receiver Operating Characteristic (ROC) curve metrics. The models are trained using 80% of the dataset, with the remaining 20% reserved for validation. Cross-validation and hyperparameter tuning are performed to optimize model performance, while explainability is ensured through SHAP (SHapley Additive exPlanations) analysis. The framework incorporates a decision tree-based model for interpretability and an ensemble approach for improved accuracy. Expected findings suggest that alloy composition, environmental salinity, pH levels, exposure time, and surface finish significantly influence corrosion resistance, with machine learning models achieving an accuracy rate exceeding 90% in resistance classification. The framework is anticipated to outperform traditional empirical models by capturing complex nonlinear interactions among predictors. This novel integration of machine learning techniques into corrosion prediction contributes to the theoretical understanding of corrosion mechanisms and offers a practical, scalable tool for industry stakeholders. The study's contribution to knowledge includes establishing a validated, data-driven framework that enhances predictive accuracy over conventional models, facilitating proactive material management and design optimization. It advances the application of artificial intelligence in corrosion science, aligning with the theoretical underpinning of the Theory of Material Degradation and the Model of Predictive Analytics. The research emphasizes the importance of integrating domain expertise with machine learning, fostering interdisciplinary approaches to corrosion challenges. In conclusion, the developed framework presents a reliable, efficient tool for early prediction of corrosion resistance, supporting strategic decisions in materials engineering and maintenance planning. Recommendations include integrating this framework into industrial corrosion monitoring systems, expanding the dataset to include long-term field data, and exploring deep learning approaches for further performance improvements. Future studies should also investigate the applicability of the framework to other failure modes, such as fatigue and stress corrosion cracking, to broaden its utility across metallurgical and structural health monitoring domains.