Análise de integridade de dutos rígidos sujeitos à corrosão generalizada empregando inteligência artificial

Abstract

The phenomenon of corrosion is a recurrent problem in the oil and gas industries, where unexpected failures can cause damage and failures of the components, affecting the safety and lives of users. In this case, by considering that, the degradation processes in such industries occur by physico-chemical phenomena capable of completely degrading the walls of pipes and ducts, these industries have been spent significant amounts of money in inspections and equipments of it with the aim of monitoring their assets suffering from corrosion. Thus, this work aims to use artificial intelligence (AI) approaches combined with statistics to estimate remaining useful life (RUL) of subsea pipelines subjected to general corrosion processes. The proposed methodology incorporates extreme value methods to generate synthetic data by using bootstrap methods to enrich the initial data for training and test the neural networks. This strategy proposed in this study has demonstrated efficiently due to the poor historical initial data available by performing non destructive testing (NDT) inspection data by the industries. In the quest for the neural network approach, it has been used the so-called Multilayer Perceptron (MLP) taking into account the wall thickness measurements over the equipment’s operational period as parameter for a time series analysis. The proposed methodology showed results with a mean absolute error of around 23,75% when compared with the 2024 inspection. Finally, the RUL estimated by the proposed method has compared with the classical t-student method and semi-probabilistic approach, showing the remaining life prediction in each methodology.