Enhanced predictive corrosion modeling with implicit corrosion products

Abstract

An advanced mathematical approach to describe the influence of corrosion products on the corrosion rate is presented here. The related model can be used as input equation for numerical predictive corrosion simulations or simply as an empirical model, to extrapolate experimental data of corrosion tests to longer times and to interpret the physical parameters behind. This semiempirical model assumes that a constant share of the dissolved metal precipitates on the surface and hinders the diffusion processes. Hence, the effective corrosion rate decreases exponentially with increasing dissolution. The explicit corrosion progress over time is derived by time integration on a newly developed, time dependent corrosion rate equation. The derived expression can be effortlessly implemented in existing for example finite element method, which is demonstrated for the uniform corrosion of a zinc surface. Furthermore, this approach is qualitatively compared with other empirical models for corrosion products and the validity is demonstrated by fitting of experimental data. A very good agreement between experiment and theory can be achieved for various materials and environments considering no change of the driving corrosion mechanism.K E Y W O R D S corrosion products, corrosion simulation, finite element method, mixed potential theory, uniform corrosion

Semantic segmentation of corrosive critical designs in body‐in‐white structures for corrosion simulation

“…The naive scaling of the corrosion simulations [2][3][4][5] to a whole body-in-white is impossible due to enormous memory requirements. By restricting the application of the corrosion simulation to corrosion-prone areas such as edges and flanges, the quality of the simulation is not affected and the processed data is reduced to 0.88% and 3.10% of the surface area, respectively.…”

“…This neural network detects 95.6% of the edges with 94.7% precision. The parameters p = (5,4,4,10) e and the opening angle…”

“…For example, with the so-called mixed potential theory for electrochemical systems [1] a stable simulation of the corrosion process for long time scales is possible. The corrosion simulations [2][3][4][5] are based on this theory and solve Laplace equations under different boundary conditions by a finite element method (FEM).…”

“…A network with n l 1 neurons in the first, n l 2 in the second and n l k n in the k n -th hidden layer is denoted by [n , n , …, n ] . As a rule of thumb, F I G U R E 5 Number of triangles per car line (dark gray) and the number of edges (gray)T A B L E 1 Best results for the different deterministic edge segmentation methods with p =(5,4,4,10)…”

“…Results of the best deterministic flange segmentation method versus Patternnets on different feature sets for P =(5,4,4,10) with optimized feature normalization β opt and network architecture Net opt on the parameter space (48) broken down to the different car lines and the mean MCC , TPR σ σ , and TPA σ performance over all car lines together with the respective TT in seconds…”