Bayesian latent degradation performance modeling and quantification of corroding aluminum alloys

Corrosion-induced degradation of aluminum alloys is a key causing factor of structural failure for many safety-critical and mission-critical engineering components/systems, such as aircraft wing and nuclear battery. Successful modeling and quantification of their degradation performance is important yet challenging since the degradation performance is not often directly observed. Existing data-driven degradation models often assume the directly observable degradation performance. Although existing physical models allow evaluating latent degradation performance, they are not flexible enough to capture the degradation complexity and fail to quantify (i) the model parameters uncertainty due to the limited experimental data; and (ii) the degradation performance variability among multiple test units. In this paper, we propose a physical-statistical modeling approach to evaluating and quantifying the latent degradation performance of corroding aluminum alloys at both individual and population levels. Based on the non-destructive testing data, the fractional order system dynamics is first considered to capture the complex electrochemical process of corroding aluminum alloys. Bayesian hierarchical models are further established to take into account both the parameters uncertainty of individual test unit and the population variability of multiple units. Real case studies are also provided to illustrate the proposed work and demonstrate its effectiveness in the experimental testing applications of different corroding metallic units.