Shell QGC continually seek to lift production while lowering unit operating cost (UOC). At QGC, unplanned downtime in hydraulic power units (HPUs) is a major driver: across a large wellsite fleet, more than 10,000 unplanned HPU trips occur annually. Each trip triggers a corrective maintenance (CM) work order (WO), yet diagnosis often becomes a hit and miss exercise due to limited historical failure data and incomplete digital symptom mapping. Traditional diagnosis relies heavily on complex datasets across multiple tools, process trends, trip metrics, and Business Intelligence (BI) dashboards. This makes data analysis and root cause investigation a time consuming process. The fragmented approach is costly, nearly impossible to scale, and can misdirect strategic focus, e.g. low trip counts can mask long failure durations, unnecessary component replacements can distort genuine mean time to failure calculations. Presenting an artificial intelligence (AI) diagnostics model that identifies the probable failure mode within seconds, previously this took hours. Rather than treating trips in isolation, the model reasons over failure episodes, incorporating WOs, parts replaced, process data, operator commentary, and CM history. At its core is a bespoke AI-driven diagnostics engine, including a domain-specific knowledge graph built from Root Cause Analysis documents, causal analysis trees, and Managed Equipment Care procedures. Further enriched with bespoke advanced analytics outputs and SME vetted historical failures, the system delivers following: recommends probable failure modes for ongoing failure and provides fleet wide failure mode insights. This shift reduces CM cycles, shortens unplanned offline duration, lowers UOC, and lifts production, thereby translating AI into tangible business performance.
Gehlot et al. (Thu,) studied this question.