| Abstract Scope |
Artificial Neural Networks (ANNs) and random forest regression were used to develop a predictive damage function for galvanic corrosion of 7075-T6 Al alloy panel with different coating systems in various environmental factors. In this research, lost volume has been modeled based on different parameters such as pretreatment, primer coating, topcoat, chloride concentration, RH, galvanic current, impressed current and environment. The best model was based on lost volume as output and the same factors except environment descriptors as inputs. The room mean square error (RMSE) for this function was 0.2 mm3. Although the RMSE was higher than for some other models, it is more realistic not to use standard environments as an input. To predict lost volume in this formula, the ANN model involved three nodes in one hidden layer with hyperbolic tangent functions. The ANN was able to get a good fit for training and validation (RMSE=0.2 and R2=0.7). |