Machine Learning-Based Corrosion Prediction Model for Steel Structures

Corrosion is a significant issue in industries, causing structural deterioration, equipment failure, and economic losses. Accurate corrosion prediction and proactively managed maintenance can minimize its impact. The study uses advanced machine learning techniques like Linear Regression, Random Forests, Support Vector Machines (SVM), and Artificial Neural Networks (ANN) to analyse corrosion-related data. Supervised learning techniques are utilised to enhance the corrosion prediction model’s accuracy and generalization capability, used to train the model on labelled data, including experimental Tafel polarization corrosion data and corresponding environmental conditions. The performance of the corrosion prediction model is evaluated using a few datasets obtained in a 3.4% NaCl solution corrosion-prone seawater environment in the temperature range of 30 ℃ to 50 ℃. Evaluation of the predictive ability of the model is performed at 33 ℃ and 55 ℃ temperatures based on performance metrics, including RMSE (Root Mean Square Error), MAE (Mean Absolute Error), MSE (Mean Square Error), and R2 (R-squared). With a minimum of obtained values of MAE (0.023528), MSE (0.000699) and RMSE (0.026446) and, close to the unity value for R2 (0.974750), the Random Forest model has proven to be better in prediction over other models in this study. Comparative analysis is conducted with existing corrosion prediction approaches to highlight the superiority of the machine learning-based model in terms of accuracy, % efficiency, and applicability across various industry sectors, keeping the goal of this project to create a machine learning (ML) based corrosion prediction model that combines cutting edge methods to predict the presence and severity of corrosion in various situations.