Multi-factor mining and corrosion rate prediction model construction of carbon steel under dynamic atmospheric corrosion environment

Abstract

For vehicles and vessels like cars and ships, the practical environmental conditions encountered are consequences of dynamic and static operation, include moving off and stopping. Therefore, it is necessary to perform dynamic atmospheric exposure corrosion tests on metal materials under vehicle operating conditions. In this paper, a carbon steel dynamic atmospheric corrosion test was designed considering vehicle operating conditions. After completion of the corrosion test, the corrosion data was collected, processed and predicted by a series of methods and algorithms. Firstly, multiple dedicated test vehicles operating in Tianjin were chosen as carriers for carbon steel dynamic atmospheric corrosion test. Secondly, the driving characteristics data was collected using an on-board positioning data collection and transmission system. Then, the dynamic fuzzy clustering algorithm was introduced to identify the key influencing factors in the parameters of vehicle driving characteristics and to optimize data dimension, the two key index parameters of the vehicle driving characteristics identified being the average vehicle speed and dynamic/static ratio. During the atmospheric corrosion test, a time-weighted weighting algorithm model was constructed and applied to the previously collected meteorological and pollutant data in order to fit the sampling interval of corrosion data. Finally, four models including GA-SVR, unoptimized SVR, GA-BPNN and unoptimized BPNN were constructed to predict the corrosion rate of carbon steel in the dynamic environment. The final results show that the root mean square error and average relative error reached their minimum for GA-SVR with the prediction result R2 = 0.9771, all indicators suggest that GA-SVR holds an advantage over the three models including SVR, GA-BPNN and BPNN.