Time-Embedding Transformer-Based Remaining Lifetime Prediction of Organic Coatings: From Short-Term EIS Monitoring to Long-Term Performance Evolution

The long-term degradation of organic coatings is governed by coupled effects of material and environmental factors, which poses great difficulties for accurate service performance prediction. Electrochemical impedance spectroscopy (EIS) is a common technique for evaluating coating aging and failure. However, conventional EIS empirical models have poor generalization and limited capacity to describe dynamic degradation processes, while existing data-driven neural networks cannot effectively capture temporal dependencies or support reliable long-term extrapolation. To solve these problems, this study develops a modified time-embedding transformer (TET) model to predict coating degradation and remaining service life using only EIS data available up to the prediction time. The embedded time-embedding module enhances temporal feature extraction, and the self-attention structure captures long-range dependencies in impedance time series. The low-frequency impedance modulus (|Z|0.01 Hz) was adopted as the primary model input, and the model was built and checked separately for nine epoxy coating–environment combinations. The results demonstrate that, when each coating system was modeled and checked separately, the TET model presents high prediction accuracy and stability across the studied systems. Compared with typical neural networks, the proposed method reduces prediction errors and effectively overcomes the limitations in dynamic degradation description and long-term trend extrapolation. This study provides an efficient and accurate method for coating durability evaluation to support the maintenance decision-making of anticorrosive organic coatings.