Abstract：The purpose of the work is to develop a general method, to predict the corrosion resistance of Znbased coatings, expressed as total mass loss in an accelerated salt spray test. The method is to be based on just three analytical parameters; the total coating weights of Zn, Al and Mg. The reason for this restriction is that determination of these three parameters is possible in on-line analysis. The predicted corrosion resistance could then be included in a process/quality control system. Accelerated corrosion tests have been carried out by Swerea KIMAB IC (Institut de Corrosion) in Brest, and CRM in Belgium. Test were run according to the Renault ECC1 test D172028/C (12 weeks), and with an accelerated cyclic corrosion test developed by CRM. The materials were divided into four corrosion classes according to total mass loss. All corrosion experiments show clearly the well documented positive influence of magnesium and aluminium. In relation to the masses of these elements in the coatings, the influence of both elements is considerably higher than the influence of zinc alone. For this reason, a new quantity is introduced, called "equivalent Zn coating weight". This quantity is a linear combination of the coating weights of zinc, aluminium and magnesium. A model for prediction of corrosion resistance was developed with the expert system, based on a combination of regression analysis and a "decision tree" algorithm. The model was able to correctly classify 25 out of 27 materials based on just the three analytical parameters mentioned above: the total coating weights of zinc, aluminium and magnesium. In conclusion, the approach shows that an accurate prediction of the corrosion behaviour is possible even on-line. For purposes of material development, the expert system can also be expanded to include additional analytical parameters.
［1］Hildebrand L, Michalski C, Wickrath M, et al. Strategy implementation for mobile robots using the pipes & filters architecture ［J］. Proceedings "FIRA Robot World Congress 2003", Wien, 2003.
［2］ Hildebrand L, Fathi M. Linguistic color processing for human-like vision systems ［J］. Proceedings "Electronic Imaging 2000, IS&T/SPIE 12th International Symposium", San Jose, USA, 2000.
［3］ Bengtson A, Nelis T. The concept of constant emission yield in GDOES ［J］. Analytical and Bioanalytical Chemistry, Springer, 2006, 385(3): 568585.
［4］ Quinlan R J. Learning with Continuous Classes ［J］. Procedings of 5th Australian Joint Conference on Artificial Intelligence, Singapore, 1992, 343348.
［5］ Quinlan R J. C4.5: Programs for Machine Learning ［M］. Morgan Kaufmann Publishers, San Mateo, CA, 1993.
［6］ Winter J, Hildebrand L. Interactive Decision Trees and Artificial Neural Networks ［J］. Proceedings International Conference on Knowledge Management for Composite Materials; Germany, 2007.
［7］ Ruiz J, González A, Cabalín L M, et al. OnLine LaserInduced Breakdown Spectroscopy Determination of Magnesium Coating Thickness on Electrolytically Galvanized Steel in Motion ［J］. Appl. Spectrosc. 2019, 64 (12):13421349.
［8］ Balzer H, Hoehne M, Sturm V, et al. New approach to online monitoring of the Al depth profile of the hot-dip galvanised sheet steel using LIBS ［J］. Spectrochimica Acta Part B, 2005, 285(2):225233.
