| Summary: | The combination of the Finite Element Method (FEM) with Convolutional Neural Networks (CNNs) presents a key breakthrough in the assessment of the structural integrity of offshore pipelines. The advantage of the standard FEM is in stress visualization, but it is time-consuming due to high computational analysis. This research aims to quickly and accurately determine the severity of pipeline corrosion categorized as high, intermediate, or low through stress images generated from FEM. A transfer-learning algorithm was applied to refine and validate the model using a diverse image dataset of uniformly corroded pipelines (200x200 mm, 100x100 mm, 75x75 mm, 50x50 mm, and 10x10 mm), annotated with corresponding severity levels. Moreover, the model was validated for prediction with irregular-sized corroded pipelines (50x100 mm and 10x100 mm). Both samples are modeled for degrees of corrosion, 30%, 50%, and 70% of the corrosion depth, with API 5L X42 specifications. An exceptional predictive accuracy was observed, attaining average confidence levels between 97% and 100%. This work substantially augments the effectiveness of structural analyses that provide a better safety feature for critical infrastructural assets within the oil and gas industry and has great advantages to engineers, researchers, and academicians working on pipeline integrity management. © The Authors.
|
|---|