Automatic Classification of Thermal Phases During Steel Oxidation via PCA and Infrared Thermography Time Series

High-temperature oxidation affects the surface condition and quality of steel products, yet automated identification of the thermal stages governing oxide-scale formation remains limited. This study evaluated whether infrared thermography time series combined with dimensionality reduction and supervised learning could classify the heating, isothermal, and cooling phases of AISI 1045 steel oxidation. Five controlled Joule-heating experiments generated 3356 thermographic records. Each thermogram was flattened, and principal component analysis was applied independently to each experiment; the first two principal components were then used to train Random Forest, Multi-Layer Perceptron, Support Vector Machine, k-Nearest Neighbors, and XGBoost classifiers. Performance was assessed using leave-one-group-out cross-validation to preserve specimen-level independence. The Multi-Layer Perceptron achieved the best performance, with an accuracy of 0.8934, Macro-precision of 0.8740, Macro-recall of 0.8938, and Macro-F1-score of 0.8562. The highest class-specific area under the receiver operating characteristic curve was 0.95 for cooling. Most errors occurred between temporally adjacent phases, whereas confusion between heating and cooling was negligible, indicating physically consistent classification. These findings establish a proof of concept for automated thermal-phase recognition from raw thermographic sequences and support further validation under broader materials, processing conditions, and industrial environments.