From Predictive Analytics to Decision Governance: An AI-Enabled Framework for Condition-Based Maintenance of Industrial Gas Turbines

Abstract

Abstract The objective of this paper is to reframe AI-enabled predictive maintenance from a failure prediction activity into a structured decision support capability. It describes how artificial intelligence, combined with physics-based models and real operating data, supports faster interpretation of asset behavior when systems operate close to technical limits. Rather than replacing engineering expertise, AI enables more disciplined, traceable, and defensible maintenance and operational decisions within a Condition-Based Maintenance (CBM) framework. The proposed method integrates AI-based analytics within a structured Condition-Based Maintenance (CBM) decision process. Physics-based models, OEM domain knowledge, and real-time operating data are combined through AI-enabled pre-processing to identify anomalies, characterize degradation behavior, and continuously update asset condition assessments. Decisions related to inspection interval extensions, temporary operating limit exceedances, and deferral of non-critical work are governed through a formal decision process based on departure records. These records document underlying assumptions, residual risk, agreed mitigation actions, and clearly defined re-entry conditions, ensuring accountability, technical traceability, and shared risk ownership across the asset operating lifecycle. Scenario-based "what if" analyses are used to assess decision robustness and reversibility before the implementation, with final decision authority retained by human experts. Application of the proposed approach has enabled customers to extend maintenance intervals in a controlled and risk-based manner across a broad range of assets. Evidence from 104 documented departure records issued in 2025 shows that condition-based decisions have been implemented at fleet scale, with 54% of departures related to aeroderivative units and 46% to heavy duty gas turbines. The industrial relevance of these results is illustrated by large LNG applications, including implementations with ExxonMobil where condition-based interval extensions were formally adopted following OEM analytics, engineering validation, and structured risk assessment. In these applications, inspection intervals for critical gas turbines were extended from standard semi-annual cycles (4,000 fired hours) to condition-based intervals exceeding 15,000 fired hours, avoiding multiple days of production downtime per train while maintaining asset integrity. These results support the transition toward Performance-Based Contract (PBC) models, where CBM-driven decisions, digital insights, and engineering governance are directly linked to shared performance outcomes such as availability, reliability, and production continuity rather than prescriptive maintenance schedules. This paper adds to the existing literature by demonstrating that the adoption of an AI-enabled CBM decision process, grounded in formal risk assessment and decision governance, enables new collaborative operating and contractual models, including Performance-Based Contracts (PBC). The approach shifts OEM–operator relationships from transactional service delivery to partnership-based models, where priorities, risk acceptance, and operational flexibility are jointly defined. This enables operators to sustain production under constrained conditions while maintaining traceable, accountable decision-making.