Corrosion prediction of FPSOs hull using machine learning

Abstract

Corrosion is considered an important aspect in assessing the integrity of offshore marine structures. It is a process that involves the risk of keeping floating production storage and offloading (FPSO) tanks out of operation for a long time, incurring undue costs for the operator. Additionally, repairs inside tanks take a long time, especially when material purchases, such as certified steel plates, are required. Therefore, operators are interested in being able to accurately predict when structural elements must be repaired. Despite recent efforts to address this problem, accurate modeling of corrosion growth remains a challenge, mainly due to its complexity and inherent uncertainties. This work proposes the use of a regression tree model, which is a well-known machine learning technique, with the purpose of predicting when and what structural elements of FPSO tanks should be repaired. A prediction model was created by learning and testing from a real data set to estimate corrosion loss as a function of the type of structural element, age, and the fluids surrounding it. The Classification and Regression Trees (CART) algorithm was employed. The results show potential application in the material purchase planning process, minimizing the critical inspection and repair path of the FPSO cargo tank, and preventing loss of storage capacity during operation.