Experimental and data-driven PVDF hollow fiber direct contact membrane distillation module degradation prediction for water purification using a deep learning method

Polyvinylidene Fluoride (PVDF) Direct Contact Membrane Distillation (DCMD) systems offer an efficient solution for clean water production, but are prone to degradation under variable operational conditions such as salinity, temperature, and pressure. This study investigates the aging behavior of PVDF DCMD membranes using experimental cyclic operation data and introduces a predictive model based on long short-term memory (LSTM) neural networks. The results reveal both reversible and irreversible degradation mechanisms contributing to performance decline. A new empirical formula is developed to quantify the reduction in membrane efficiency over time. Experimental observations show a 0.17% decline in normalized flux per cycle during early operation and approximately 4% irreversible loss following each chemical cleaning cycle. The LSTM model achieves high accuracy in predicting degradation trends under various operational scenarios, with mean absolute percentage errors of 0.63% and 0.57% for temperature and pressure variations, respectively. This approach enables proactive control strategies and optimized operating cycles, thereby extending membrane lifespan and ensuring stable performance in real-world applications. By combining empirical observations with deep learning, this work advances predictive maintenance in membrane distillation, offering a practical tool for managing PVDF membrane aging in water purification systems.