Interpretable Operational Regime Classification for a Wind Turbine PHM Digital Twin Architecture

Abstract

Wind turbines operate under varying environmental and control conditions that make reliable interpretation of SCADA data challenging. Without contextual information about the current operating state, normal variability may be mistaken for abnormal behaviour, reducing the reliability of diagnostic and prognostic analysis. This study develops a Digital Twin framework for wind turbines using SCADA data, where periodically updated models represent turbine structure, behaviour, and operating context. To address this, physics-informed operational zones are defined based on power curve characteristics and control logic, providing structured labels. These are used to train interpretable rule-based algorithms, Repeated Incremental Pruning to Produce Error Reduction (RIPPER) and First-Order Inductive Learner (FOIL), which generate explicit IF-THEN rules linking measured variables to operational states. Evaluation using overall accuracy, macro-F1 score, and per-class precision and recall shows that both methods achieve classification while producing compact, physically interpretable rule sets aligned with known turbine behaviour. The study demonstrates that rule-based learning enables transparent and effective operational regime classification, forming a critical contextual layer for prognostics and health management (PHM) oriented Digital Twins, with applicability beyond wind turbines to other industrial assets.