Predictive maintenance system for membrane replacement time detection using AI-based functional profile monitoring: Application to a full-scale MBR plant

Abstract

The significant operational cost of membrane maintenance is an important issue to overcome alongside the membrane replacement time. We propose a membrane lifetime estimation method via functional machine learning (FML) using 12 biological-chemical-physical functional parameters and a predictive maintenance system for membrane replacement based on AI-driven functional profile monitoring (FPM). Biological, chemical, and physical information regarding membrane aging were reflected in the 12 functional parameters extracted from transmembrane pressure (TMP) and the chemical dosage. Membrane aging was diagnosed using hidden patterns and trends in the functional parameters, then the membrane lifetime was estimated by interpreting their relationships using FML. Influential functional parameters regarding membrane lifetime were monitored using FPM, and membrane replacement time was suggested according to newly suggested 4 types of membrane replacement rule. The predictive maintenance system improved the performance of permeability, pumping energy, and NaOCl cost by 0.66%, 0.45%, and 3.55%, respectively while extending membrane lifetime by 18 days. The proposed predictive maintenance system was considered efficient in comparison with manual maintenance and an incompetent maintenance system. • Functional machine learning extracted functional parameters from membrane fouling. • Membrane replacement rules were suggested to detect proper membrane replacement time. • A predictive maintenance system saved energy by 0.45% and chemical dosage by 3.55%. • The membrane could be operated 18 days longer by the predictive maintenance system.