Leveraging Artificial Intelligence (AI) for Benchmarking and Enhancing Maintenance and Reliability in the Energy Industry

Objectives/Scope This study aims to enhance maintenance and reliability benchmarking within the energy industry by employing advanced generative AI, specifically Retrieval-Augmented Generation (RAG). Benchmarking is critical for identifying best practices, optimising maintenance intervals, reducing operational costs, and ensuring asset reliability. However, traditional benchmarking processes face challenges such as inconsistent data structures, variability in maintenance builds, and manual-intensive comparison processes. Building upon previous research that utilised basic Natural Language Processing (NLP), this investigation seeks to significantly improve benchmarking accuracy, reduce processing time, and facilitate precise comparisons of diverse maintenance builds, thereby driving industry-wide improvements Methods, Procedures, Process The methodology incorporates state-of-the-art generative AI techniques, centred around the RAG framework. Initially, comprehensive data collection from various maintenance builds—including systems categorisation, task details, operational narratives, and extensive textual documentation—is conducted. Subsequently, advanced preprocessing and filtering procedures are applied to structure and prepare data optimally for RAG processing. This approach uniquely identifies and maps relationships within maintenance data, enhancing the clarity and quality of benchmarking comparisons. RAG leverages contextual retrieval and generative AI capabilities to systematically produce accurate, one-to-one relational insights across diverse asset maintenance builds. Results, Observations, Conclusions The implementation of Retrieval-Augmented Generation (RAG) demonstrated substantial improvements over the earlier basic NLP approach. Key findings include achieving an accuracy rate of approximately 94% in correctly matching maintenance items across disparate builds, representing a significant enhancement in the reliability of benchmarking outputs. Moreover, the advanced AI-driven process resulted in a dramatic reduction of analysis time—over 90% faster compared to previous methods—thus enabling rapid and frequent benchmarking cycles. The precision of RAG-facilitated outputs allowed for more effective identification of optimal maintenance intervals, significantly contributing to reliability enhancement and operational efficiency. Observations also underscored RAG's robustness in handling diverse and complex maintenance data, thereby offering greater flexibility and adaptability for benchmarking practices. These results conclusively demonstrate the capability of generative AI methods, specifically RAG, to substantially transform maintenance benchmarking, providing actionable insights and fostering best practices within the upstream oil and gas sector. Novel/Additive Information The novelty of this study lies in the pioneering use of Retrieval-Augmented Generation (RAG) within maintenance benchmarking, significantly advancing beyond previously published NLP methodologies. This approach unities and rates contextual retrieval with generative AI to enhance accuracy, significantly reduce processing times, and improve operational reliability assessments. Consequently, this innovative implementation establishes a new benchmark in the efficient and precise evaluation of maintenance and reliability within the energy industry.