Application of artificial intelligence to predict flow assisted corrosion in nuclear/thermal power plant

Abstract

Flow assisted corrosion (FAC) is a wall-thinning phenomena of carbon steel pipe in nuclear and thermal power plant. Due to FAC, many accidents have taken place in nuclear plants resulting in casualties. In FAC, dissolution of iron from the iron-oxide fluid interface at pipe wall takes place and it is affected by pH, oxygen concentration, flow rate, temperature and chromium content of piping material. Due to complex interaction of these parameters, FAC prediction is difficult using conventional modeling tools and experimental evaluation is time consuming and costly. In this work, artificial neural network (ANN) has been used for FAC prediction using 320 data points collected from published literature. The neural network training was carried out using Lavender-Marquardt back-propagation algorithm in Matlab. The results show that ANN is a powerful tool for predicting FAC rate with regression coefficient above 90% and hence it can be very useful by regular training of the model with actual operational data in safety management and long term planning in nuclear/thermal power plant. A sensitivity analysis with respect to each parameter has been carried out using ANN model. It is observed that FAC rate is lower under alkaline conditions and goes through a maxima in a temperature range of 140 to 150°C.