For any industrial actor that is managing complex equipment and processes on a daily basis, maintenance is crucial for running the operations efficiently and safely. Developing timely and efficient maintenance strategies for systems in operation is currently one of the main industrial challenges. This is of particular importance for complex and critical systems used for drilling operations. There are three basic approaches for equipment maintenance: corrective, preventive (planned) and predictive (condition based) maintenance. Predictive maintenance is an advancement in the maintenance philosophy. It is far more than just applying condition monitoring technologies. With predictive maintenance repair decisions are based on the condition of the machine rather than on the calendar. Surveys indicate that predictive maintenance leads to significant reduction in maintenance cost, elimination of equipment failure, reduction in rig downtime and increase in production efficiency when compared to other maintenance approaches. Despite the latest industrial advancements, such as exponential growth of available data combined with technologies like machine learning and artificial intelligence, there remains a low maturity of digitalization across the oil and gas industry. Deep learning emerged lately as a method that can offer promising results for fault diagnosis and prognostics. This is a relatively unexplored area especially when related to oil and gas industry. This thesis will explore the potential for using deep learning to predict drilling equipment problems and estimate its remaining useful life. Specifically, focus will be to develop and apply deep learning techniques to enable autonomous decision-making for improving service performance through condition-monitoring, predictive maintenance and spares management based on the real-time data available from the drilling rigs in operation.
Project leader: Pål Skogerbø
Institution: MHWIRTH AS