PlantReliability Fusion Algorithm for Predicting Equipment and Structural Support Failures in Combined-Cycle Power Plants Using Operational and Structural Data Compared with Weibull Reliability Analysis

Abstract

Combined-cycle power plants contain strongly coupled mechanical, thermal, electrical, and structural systems whose degradation mechanisms evolve under variable operating loads, cyclic thermal stresses, vibration, environmental exposure, and equipment ageing. Conventional reliability techniques such as Weibull reliability analysis are effective for estimating population-level failure distributions but have limited capability to incorporate high-frequency operational measurements, structural-condition indicators, nonlinear interactions, and time-varying degradation simultaneously. This study develops a novel PlantReliability Fusion Algorithm (PRFA) for integrated prediction of equipment and structuralsupport failures in combined-cycle power plants using operational and structural data. The proposed framework combines a temporal feature-learning module for processing equipment variables such as compressor discharge pressure, turbine exhaust temperature, vibration, bearing temperature, heat-recovery steam generator pressure, start-stop cycles, operating hours, and load fluctuations with a structural-condition module incorporating foundation settlement, support displacement, strain, vibration response, crack development, anchor-bolt condition, thermal expansion, and structural stress indicators. An attention-based multimodal fusion layer dynamically weights degradation signals from both data domains, while a reliability-learning layer estimates failure probability, hazard evolution, remaining useful life, and component reliability over time. PRFA is benchmarked against two-parameter Weibull reliability analysis, Cox proportional hazards regression, Random Forest, XGBoost, Long Short-Term Memory networks, and a conventional artificial neural network. Comparative performance is evaluated using accuracy, precision, recall, F1-score, area under the receiver operating characteristic curve, concordance index, Brier score, remaining-useful-life error, and computational efficiency. Graphical evaluation includes Weibull and predicted reliability curves, hazard-rate trajectories, receiver operating characteristic curves, predicted-versusobserved remaining useful life plots, structural-operational degradation maps, feature-importance plots, and comparative algorithm performance charts. The proposed fusion architecture is designed to provide stronger sensitivity to interacting mechanical and structural degradation mechanisms than conventional failure-time models or single-domain machinelearning approaches. The resulting framework provides a technically integrated basis for predictive maintenance, structural integrity management, inspection prioritization, outage planning, and lifecycle reliability optimization in combined-cycle power plants.