Corrosion in Wet Gas Piping: Root Cause, Mitigation, and Neural Network Prediction Modeling

Abstract

This paper discusses the root causes and operational mitigations of corrosion anomalies reported for an FPSO wet gas system, and crucially, proposes a neural network (NN) prediction model. The NN model involves ‘back-propagation’ processing of each nodal root cause and mitigation to obtain a value which when combined with a processing weight and then summed, provides an output value. This value is then used to further adjust the weights. Each weight correlates with the magnitude of influence on the overall corrosion rate. The ability to train the model (i.e., weight-adjustment during processing) makes it responsive and adaptable, such that when fresh data inputs are made in a ‘forward-propagation’ mode, into the large modeling database that has been developed (which includes a large number of susceptibility factors), significant increases in the accuracy of predicting corrosion rate and integrity behavior of the wet gas system can be achieved. The identified root causes and mitigations will be useful in further understanding the internal degradation mechanisms operating in wet gas systems in general.