ABSTRACT
The vulnerability of aluminum alloys to localized corrosion, particularly near fastener holes, is known to directly affect crack initiation. Gaining an improved understanding of environmental assisted cracking mechanisms necessitates an understanding of static and dynamic corrosion morphology associated with cracking. Traditional methods have fallen short in addressing the complexities of characterizing damage observed under diverse testing conditions. This study introduces a new image analysis algorithm, capable of quantifying specific damage features. To validate the algorithm, panels of AA7075-T6 with a SS 316 fastener inserted were exposed to realistic relative humidity cycles that mimic outdoor environmental conditions. Then, robust statistical analyses were used to validate the algorithm's performance by comparison with an analysis performed by a subject matter expert. Multiple analyses, including a significant Cohen's κ statistic, Kruskal γ test, and F1 score were employed. Low Mean Absolute and Mean Square error values were observed, indicating the precision of the algorithm. High R-squared values ensured the algorithm's explanatory capacity, while the Bland-Altman plot reveals overall alignment between algorithmic predictions and expert measurements. Lin’s concordance correlation coefficient value further accentuates the outstanding agreement between algorithmic predictions and expert assessments. Together, these comprehensive statistical analyses affirm the algorithm's accuracy, reliability, and precision in categorizing instances of IGC from digital images, highlighting its potential as a powerful tool in corrosion analysis.
