Smart Maintenance: Enhancing Maintenance Effectiveness at Pace with Artificial Intelligence

Abstract Objectives/Scope The integration of Artificial Intelligence (AI) in production operations within the oil and gas industry has demonstrated significant potential in optimizing maintenance strategies and improving the value gained from maintenance operations. This study explores the application of AI and large language models (LLMs) to enhance Failure Mode and Effects Analysis (FMEA), gaining greater understanding of asset performance, optimizing maintenance effort, and improving asset reliability. The methodology involves classifying maintenance actions, identifying failure modes using Natural Language Processing (NLP), and automating object type classification. Results from case studies show significant reductions in maintenance hours and improved asset reliability. This paper discusses the novel contributions of AI in maintenance optimization and its implications for the industry. Methods/Processes/Procedures Traditionally, maintenance activities in offshore operations rely heavily on manual classification and interpretation of extensive historical maintenance data. This labour-intensive process often results in misclassification, inaccurate reliability assessments, and suboptimal maintenance strategies. Our AI-based approach addresses these issues by first classifying maintenance actions into three distinct categories: corrective (failure-induced), preventive, or life-resetting. AI-driven Natural Language Processing (NLP) techniques utilizing LLMs then identify precise failure modes from unstructured maintenance reports and historical logs, significantly reducing analysis time by over 95%. Critical to accurate reliability modelling and predictive maintenance strategies is the identification of equipment types (object type classification). Use of AI automates the process of assigning appropriate object types to failures previously logged without specific equipment references. By accurately linking failures to the correct asset type, the AI significantly enhances the precision of reliability analyses and failure predictions. Additionally, AI-driven functional location (FLOC) reclassification ensures maintenance records initially assigned to higher-level tags within the Computerized Maintenance Management System (CMMS) hierarchy are reassigned appropriately. Accurate FLOC classification is essential, as it directly influences the granularity and quality of predictive modelling, enabling more precise targeting of maintenance efforts and improved operational efficiency. Results/Observations/Conclusions Through case studies we will show how this approach generated a reduction in annual maintenance hours of 100,000 hours for a large oil & gas operator in the region. This release of capacity from low-value maintenance has allowed the core maintenance teams to focus on targeted reliability improvement & maintenance backlog, reducing overall risk. We will also share details of a long-term study into impact of maintenance optimisation on the performance of a UKCS asset cluster. Following an initial project in 2022 we have captured performance data from the CMMS which has allowed us to utilise AI to assess the reliability and performance of the asset and validate our methodology. Novel/Additive Information By enhancing the data available for decision-making, the integration of AI to enhance unstructured asset management data supports subject matter experts (SMEs) with data-driven insights, enabling them to focus on strategic decision-making and optimizing management of the asset; a true human-AI collaboration. AI also support decision-making at pace, allowing for increased value from maintenance activities.