Data-Driven Analysis of Gas Turbine Compressor Fouling and Water-Washing for Improving Efficiency Maintenance Planning and Wastewater Management in Sustainable Power Plant Infrastructure

Abstract

Gas turbine compressor fouling is a major source of progressive efficiency degradation, increased heat rate, reduced power output, higher fuel consumption, and avoidable maintenance expenditure in combined-cycle and industrial power plants. Compressor water washing can recover lost aerodynamic performance; however, conventional washing schedules based on fixed operating hours, differential pressure thresholds, or operator experience may result in premature cleaning, excessive water consumption, unnecessary downtime, or delayed intervention. This study develops a data-driven framework for jointly predicting compressor fouling severity and optimizing water-washing and maintenance decisions while incorporating wastewater-management constraints in sustainable power plant infrastructure. A novel Fouling–Wash Adaptive Optimization Network (FW-AON) is proposed to integrate multivariate operating data including compressor inlet temperature, ambient humidity, pressure ratio, corrected mass flow, compressor efficiency, exhaust temperature, power output, fuel flow, operating hours, particulate loading, water-wash history, and wastewater volume. The FW-AON combines temporal feature learning, nonlinear fouling-state estimation, degradation forecasting, and multi-objective maintenance optimization to determine the economically and environmentally preferred washing interval. Its predictive performance is compared with Random Forest, XGBoost, Long Short-Term Memory, Gated Recurrent Unit, and Support Vector Regression using coefficient of determination, root mean square error, mean absolute error, fouling-detection accuracy, computational time, and maintenancedecision effectiveness. System-level performance is further evaluated through compressor efficiency recovery, power-output recovery, heat-rate reduction, fuel savings, wash-water consumption, wastewater generation, maintenance cost, and avoided production losses. Results are presented using time-series degradation curves, predicted-versus-observed plots, algorithm comparison charts, sensitivity graphs, Pareto-front optimization plots, and water-use versus efficiency-recovery relationships. The proposed framework is intended to provide improved fouling prediction and condition-based washing decisions compared with fixed-interval and conventional machine-learning approaches while simultaneously minimizing unnecessary water consumption and wastewater generation. The study establishes an integrated pathway for coupling gas turbine performance analytics, predictive maintenance, water-resource efficiency, and environmental management within sustainable power plant infrastructure.