Large Language Models, ChatGPT, and Beyond: A New Paradigm for Condition Monitoring and Prognosis in Power and Energy Systems

Large language models and related foundation models offer new capabilities for interpreting heterogeneous information, retrieving engineering knowledge, and supporting high-level decision processes in power and energy systems. This Perspective examines their role in condition monitoring and prognostics and health management, with particular emphasis on remaining useful life estimation. To support language and multimodal model interpretation of structured and validated estimates, we propose a digital twin-centric architecture comprising physics-guided hybrid estimation, digital twin synchronization, rule-governed semantic translation, foundation model reasoning, output validation, and human supervision. The specific contribution is threefold. First, the proposed layering principle distinguishes physically consistent estimation, system integration, and higher-level reasoning functions. Second, an auditable semantic interface is defined to encode health indicators, degradation states, remaining useful life distributions, uncertainty information, operating context, provenance, retrieved evidence, and permitted reasoning scope. Third, a deployment-oriented reliability framework relates physical consistency, digital twin consistency, explanation fidelity, uncertainty preservation, robustness, traceability, safe deferral, latency, data sensitivity, and computational requirements to edge, on premises, cloud, and hybrid implementations. An illustrative battery state-of-health and remaining useful life use case demonstrates the resulting information flow. The foundation model layer is primarily intended for slow-timescale, human-supervised functions such as diagnostic explanation, degradation interpretation, maintenance prioritization, inspection planning, and fleet-level assessment, rather than protection or real-time control. The Perspective also examines complementary roles for small language models, tiny recursive models, world models, large language models, and multimodal foundation models and identifies priorities for their reliable evaluation and deployment in power and energy prognostics and health management.