EIS Prediction of degradation of Aluminum beverage can lids using LSTM neural networks

Corrosion prediction in aluminum can lids is important to maintain the longevity and safety of beverage containers, with Electrochemical Impedance Spectroscopy (EIS) serving as a helpful modern tool to assess coating performance. This study aims to develop a Long Short-Term Memory (LSTM) network for predicting EIS spectra, enabling more efficient and real-time corrosion assessments. The impedance data was collected using a custom-designed experimental setup that replicated real beverage can environments, generating comprehensive EIS spectra under controlled conditions. This data was then used to train and validate an LSTM model tailored for time-series predictions. The model uses a three-dimensional input structure, incorporating frequency, real impedance (Z'), and imaginary impedance (Z'') as input features. This approach captured broader changes in the EIS spectra, allowing the model to simultaneously map changes in Nyquist plots, Bode magnitude, and phase angle plots. The results demonstrated good predictive accuracy, with the model capturing the complex dynamics of EIS spectra over conventional analytical methods. This study tries to integrate advanced machine learning with electrochemical analysis to enhance material evaluation. This approach can be expanded to predict corrosion not only in coated aluminum lids but also across broader applications involving aluminum and its alloys. By incorporating recurrent neural networks (RNNs) or advanced LSTM architectures, future work could further refine predictions of EIS spectra, advancing corrosion monitoring techniques in diverse industrial and environmental contexts.