Achieving The In-Situ Rapid Detection Of Marine Corrosion By The Combined Ann Method

Marine corrosion is a problem which must be solved and faced in ocean exploitation. The determination and comparison of marine corrosiveness in different region not only play an important role in the selection and designation of ocean engineering material, but also provide more reference value while considering the life of steel structure in ocean environment. Out of this mind, a method of combined artificial neural networks (ANN) is employed to establish the prediction model of"seawater environmental factors—steel corrosion rates"for six steels (3C,A3,16Mn,10MnPNbRe,10CrMoAl,D36) and then predict their corrosion rate under 10200 groups of seawater environmental factors. So, the corrosiveness of seawater in different region can be characterized by the corrosion rates, which are input into the ACESS database. And the in-situ rapid query of marine corrosiveness in different region is achieved by the database query interface, which is programmed by VB(Visual Basic) language. The main contents of this paper include the four aspects:1. Constructing the model of combined ANN. In this new method, the training and predicting were divided into two steps: clustering the samples at first, and then training and predicting them at the clustered areas respectively by SOM and RBF neural network.2. The corrosion rates of six steels(3C,A3,16Mn,10MnPNbRe,10CrMoAl,D36) under 10200 groups of seawater environmental factors are predicted by the combined model. And then, the prediction data is input into ACESS database with a friendly VB query interface, which helps to achieve the rapid query of"seawater environmental factors—steel corrosion rates"3. The prediction results gained by the combined ANN method is compared with that by BP model , which shows that the prediction error of BP model is improved by 5.36% mostly ,and convergence rate is improved by 2-3 times.4. On the one hand, the trend of saturated seawater's corrosiveness on temperature is predicted by combined model, on the other hand, the predicted results is compared with those of the actual measurement in Maidao. All the results show that the predicted trend and error can meet the practical requirements very well.