Prediction of Single-Phase Erosion-Corrosion in Mild Steel Pipes Using Artificial Neural Networks and a Deterministic Model

Abstract Erosion-corrosion is flow-assisted corrosion that can cause wall thinning in fluid piping systems. Several key parameters, such as pH, temperature, flow rate, mass transfer coefficient (which is a function of the geometry and pipe configuration), and materials determine the rate at which damage develops. In this study, we generated an experimental data base from the open literature which we used to train and to test an Artificial Neural Network (ANN). We also developed a deterministic model which we used to make predictions. The predictions from the deterministic model, and from the ANN were compared to the experimental data collected and the results are reviewed and discussed. The artificial Neural Network was designed to learn from about 60% of the experimental data collected. The data contained as variables experimental single phase erosion-corrosion rates (mm/yr) (for several configurations of mild steel piping under various environmental and mechanical conditions including: pH, temperature, flow rate, mass transfer coefficient, and oxygen concentration). However, most of the data collected contains no information on the oxygen concentration in the solution, the hydrodynamic numbers characterizing the geometry, or flow velocity. Instead of the hydrodynamic characteristics, the mass transfer coefficient was given (the mass transfer coefficient will account for geometry and flow velocity effects). The experimental information usually does not contain detailed information on the material composition, or on the chemical composition of the solution. Accordingly, the number of variables used to train the ANN was limited. The main difference between the two models concerns the prediction of erosion-corrosion rates at various temperatures at high pH values. The ANN, which was trained on experimental data, predicts that erosion-corrosion will decrease with pH independently of the temperature and flow rate. On the other hand, the deterministic model predicts that the erosion-corrosion rates will remain high even at high pH and high temperature, and is independent of the flow rate (this latest result agrees with the measured observations on the dissolution of magnetite), and only at low temperatures will it decrease with pH. At this time, there is insufficient experimental data of sufficient quality to indicate which of the two models more accurately predicts erosion-corrosion rates versus pH at several temperatures.