This study applied statistical methodology and machine learning to predict the electrochemical and mechanical characteristics in relation to localized corrosion of stainless steel under marine environment. In machine learning, multiple linear regression (MLR) and artificial neural network (ANN) were used. The selected target material was austenitic stainless steel (316L, 904L, AL-6XN), and, based on data derived from a series of engineering experiments, the relationship between variables (input) and response (output) was modeled through MLP and ANN.
First, significant factors were selected from various environmental variables that affect localized corrosion, such as pitting resistance equivalent number, temperature, and pH, as well as chloride, sulfate, and nitrate concentration, and their effects on localized corrosion were evaluated using machine learning methods.
Second, the optimal solution to test the sensitization of AL-6XN was clarified by the Taguchi method, and localized corrosion characteristics were predicted using sensitivity and environmental variables.
Finally, an MLP model was designed to predict mechanical strength with pit parameters, and its performance was verified.
The statistical methodology and machine learning achieved acceptable performance for the training, validation, and test processes. In particular, the ANN showed excellent performance using the nonlinear data of the complex-coupling corrosive environment.
This paper presents evidence that these statistical machine learning approaches can be used as a good corrosion predictor.