Application of Machine Learning to Characterize the Permeate Quality in Pilot-Scale Vacuum-Assisted Air Gap Membrane Distillation Operation

Abstract

Membrane distillation (MD) is a thermal desalination technique proposed for the valorization of residual brines that other operations such as reverse osmosis cannot treat. Previous studies have shown that vacuum-assisted air gap (V-AGMD) operation in commercial multi-envelope modules improves the performance of MD noticeably. However, the permeate quality at pilot scale has not been thoroughly characterized so far. The aim of this study is, therefore, to assess and model the effect of the main operating conditions (feed flow rate, inlet temperatures, and feed salinity) on the permeate quality. Results from different steady-state experiments allowed to estimate descriptive metrics such as the salt rejection factor (SRF) and the membrane leak ratio (MLR). Given their non-linear behavior, these metrics were subsequently modeled using artificial neural networks (ANN) to estimate the permeate quality in the whole scope of operating conditions. Acceptable SRF results with MLR values lower than 0.2% confirmed the validity of MD as an operation for the treatment of concentrated brines, although the salinity of the resulting permeate does not comply in all cases with that permitted for human consumption.

Keywords

• membrane distillation
• brine treatment
• experimental
• zero-liquid discharge
• machine learning
• modeling