Deep learning-based time-series model for predicting high-temperature oxidation behavior in austenitic heat-resistant steels

Austenitic heat-resistant steels (AHSs) are widely used in high-temperature components, and their oxidation resistance at high temperature is closely related to long-term service reliability. However, high-temperature oxidation is an intrinsically complex, nonlinear, strongly time-dependent, and multi-stage process, making accurate prediction of oxidation kinetics challenging. In this work, an attention-enhanced bidirectional long short-term memory (Bi-LSTM) framework is proposed to predict oxidation mass gain of AHSs in dry air. The model integrates static compositional descriptors with dynamic time-series inputs, enabling the extraction of sequence-dependent features, while an attention mechanism adaptively emphasizes critical temporal information. The developed model achieves high predictive accuracy in a representative run ( R ² = 0.9524), outperforming traditional machine learning models and the baseline LSTM model. The framework provides a practical data-driven sequence modeling method for long-term lifetime assessment and design of AHSs, and offers a transferable methodological basis for other alloy systems with complex time-varying processes.