AI-Enhanced Tools for Performance Enhancement in Energy and Water Systems

Abstract

Abstract Whether used for transportation of gas, oil, produced water, or wastewater, pipeline infrastructure is critical to many industries and is the most effective way to transport large quantities of fluids and gases. Considering that that large portions of pipelines are increasingly reaching or exceeding their service lifetimes, technologies that help with maintenance and extension of service life are a critical future need. One increasingly popular method of improving performance and reducing maintenance requirements are utilizing internal surface application of coatings and surface treatments to reduce the chance of deposits and the formation of corrosion. Proper coating and application require careful surface preparation to minimize the risk of coating adhesion failure and catastrophic delamination. The natural variation in training and experience of applicators can often lead to quality control issues for coating, and on system performance and expected lifetime. Utilizing a disruptive application of artificial intelligence to reduce subjectivity and reliance on operator interpretation can thus enable a more systemically uniform method of achieving quality control goals, while additionally making the process of coating and surface treatment more efficient. This work describes a novel approach to combine a digital inspection system with artificial intelligence (AI) to identify defects and areas of risk in existing pipelines to assist with coating application quality control. Video captured from borescopes and pipeline inspection gauges (PIGs) were supplemented with sensor data for an AI to utilize for evaluation of coating quality and coverage. A two-step approach was followed, whereby transforms and training took place on small diameter pipes and using the algorithm with a pig-mounted capture tool. The inspection method transforms videos into a static two-dimensional (2D) image map that is representative of the inspected tubular being longitudinally cut and unwrapped into a flat surface, which can then be processed with the AI component to identify defects and features of interest. Applications described in this work include estimation of corrosion, cleanliness after intervention, and coating coverage after application. Additionally, the 2D generated maps were backmapped onto 3D plots to preserve the relative geometry and dimensions of the pipeline in a fully interactive virtual space which could be manipulated by a human user for further defect analysis and inspection. As a proof of concept, the AI system was trained to distinguish the level of discoloration present on steel substrates designed to imitate the interior diameter of transport pipelines. Classification accuracy was 91.2% overall when predicting for cleanliness using the AMPP standard classification system.