Physics-Informed, Data-Driven Model for Atmospheric Corrosion of Carbon Steel Using Bayesian Network

Abstract

Atmospheric corrosion is a significant challenge faced by the aviation industry as it considerably affects the structural integrity of an aircraft operated for long periods. Therefore, an appropriate corrosion deterioration model is required to predict corrosion problems. However, practical application of the deterioration model is challenging owing to the limited data available for the parameter estimation. Thus, a high uncertainty in prediction is unavoidable. To address these challenges, a method of integrating a physics-based model and the monitoring data on a Bayesian network (BN) is presented herein. Atmospheric corrosion is modeled using the simulation method, and a BN is constructed using GeNie. Moreover, model calibration is performed using the monitoring data collected from aircraft parking areas. The calibration approach is an improvement over existing models as it incorporates actual environmental data, making it more accurate and applicable to real-world scenarios. In conclusion, our research emphasizes the importance of precise corrosion models for predicting and managing atmospheric corrosion on carbon steel. The study results open new avenues for future research, such as the incorporation of additional data sources to further improve the accuracy of corrosion models.