Development of an intelligent IoT-based framework for boiler water quality monitoring and early warning using statistical process control and machine learning

ABSTRACT Boiler water quality directly influences the efficiency, reliability and service life of industrial steam generation systems, and inadequate monitoring can cause scaling, corrosion, reduced heat transfer efficiency and unplanned shutdowns. This study presents an intelligent Internet of Things (IoT)-based framework for continuous boiler water quality monitoring and early warning that integrates real-time sensing, statistical process control (SPC), rule-based decision support and machine learning. Six key parameters are monitored continuously: pH, conductivity, chloride concentration, hardness, dissolved oxygen and temperature. SPC performs real-time process monitoring, while a rule-based decision support system generates corrective recommendations from predefined operational limits. To validate the framework, nine supervised machine learning models were evaluated using representative operating scenarios; Random Forest and LightGBM achieved the highest classification performance, with 99.5% accuracy under 10-fold cross-validation. Correlation analysis showed strong positive relationships between conductivity and hardness (r = 0.73), conductivity and chloride (r = 0.70), and hardness and chloride (r = 0.72). Feature importance analysis identified temperature (31.9%), pH (27.9%) and hardness (18.1%) as the most influential variables. The framework enables continuous monitoring, rapid identification of abnormal boiler conditions and timely operational decision support, demonstrating proof-of-concept capability for further validation with long-term industrial boiler data.