Galvanic Corrosion Between Coated Al Alloy Plate and Stainless Steel Fasteners, Part 2: Application of Finite Element Method and Machine Learning to Study Galvanic Current Distributions

Abstract

Aluminum alloy panels joined with stainless steel fasteners have been known to occur in aerospace structures, due to their respective optimized mechanical properties. When connected via a conductive solution, a high-driving force for galvanic corrosion is present. The combination of the dissimilar materials, indicating galvanic corrosion, and complex geometry of the occluded fastener hole, indicating crevice corrosion, leads to the detrimental combined effect of galvanic-induced crevice corrosion, as investigated previously in Part I. The present work extends the validated finite element method (FEM) model to predict the current distribution and magnitude in a variety of geometric and environmental conditions, with the goal of preventing corrosion damage within the highly-susceptible fastener hole. Specifically, water layer thicknesses ranging from bulk full-immersion (800 μm) to atmospheric (89 μm) conditions was investigated, as well as the impact of external scribe dimensions. Two avenues for mitigation were determined, 1) to force the majority of current away from the fastener hole and onto the bulk surface of the panel, and 2) to lower the overall galvanic coupling current. A random forest machine learning algorithm was developed to generalize the FEM predictions and create an open-source applicable prediction tool.

Accelerating FEM-Based Corrosion Predictions Using Machine Learning

“…4 and 5). Another recent study from Marshall et al 58 combined ML through a random forest model with FEM simulations to create a supervised classification mapping and interpolated between the multivariable input database. The random forest model allowed for an optimized approach to conserve the computational cost, preserve the accuracy of the FEM, and increas the accessibility to real-time calculations.…”

“…The random forest model allowed for an optimized approach to conserve the computational cost, preserve the accuracy of the FEM, and increas the accessibility to real-time calculations. 58 Therefore, having the ability to predict corrosion current and corrosion susceptibility with ML allows the creation of ML multifidelity (i.e. experimental and FEM data) surrogate models with new, enhanced predictions, potentially decreasing error in predictions.…”

“…Therefore, there is a clear need for efficient, computational methodologies to quantify corrosion given the vast potential exposure environments 16 or material environment combinations. Most importantly, there is a need for a framework that is capable to distill actionable knowledge from previously performed models and furthermore that the obtained knowledge can be seamlessly used to predict values of interest for new/unseen scenarios in a fast and accurate manner.…”

“…Nevertheless, it is important to point out that these previous successes are focused on implementing/developing models for data that has been obtained experimentally, which presents challenges on its own since data describing corrosion exposures can be incomplete or noisy, as well as the lengthy time scales typically associated with field or laboratory exposures experiments. The application of ML to FEM corrosion simulations was shown by Marshall et al 16 who leveraged random forests to establish a classification model to predict where corrosion will occur in a coated Al plate and stainless steel fasteners based on a training set of FEM simulations.…”

Corrosion Modeling of Aluminum Alloys: A Brief Review

“…Specifically noteworthy are the works that combine finite element models with machine learning. [73,74] This integration symbolizes so far, the most comprehensive approach to address multiscale aluminum corrosion, which has the potential to produce substantial advances in the accuracy and predictive capability of models applied to different industries. Lastly, a potentially promising direction for the field is the application of fractional calculus, which is a powerful tool of widespread use for describing dynamic processes of diverse nature.…”