Multivariate statistically-based modelling of a membrane bioreactor for wastewater treatment using 2D fluorescence monitoring data

This work presents the development of multivariate statistically-based models for monitoring several key performance parameters of membrane bioreactors (MBR) for wastewater treatment. This non-mechanistic approach enabled the deconvolution of 2D fluorescence spectroscopy data, a powerful technique that has previously been shown to capture important information regarding MBR performance. Projection to latent structure (PLS) modelling was used to integrate 2D fluorescence data, after compression through parallel factor analysis (PARAFAC), with operation and analytical data to describe an MBR fouling indicator (transmembrane pressure, TMP), five descriptors of the effluent quality (total COD, soluble COD, concentration of nitrite and nitrate, total nitrogen and total phosphorus in the permeate) and the biomass concentration in the bioreactor (MLSS). A multilinear correlation was successfully established for TMP, CODtp and CODsp, whereas the optimised models for the remaining outputs included quadratic and interaction terms of the compressed 2D fluorescence matrices. Additionally, the coefficients of the optimised models revealed important contributions of some of the input parameters to the modelled outputs. This work demonstrates the applicability of 2D fluorescence and statistically-based models to simultaneously monitor multiple key MBR performance parameters with minimal analytical effort. This is a promising approach to facilitate the implementation of MBR technology for wastewater treatment. [Display omitted] ► Statistically-based models were developed for monitoring a membrane bioreactor. ► 2D fluorescence data was incorporated in addition to operation and analytic data. ► PLS modelling was applied, after PARAFAC compression of fluorescence data. ► Indicators of permeability, effluent quality and biomass were successfully modelled. ► Multiple performance descriptors can be simultaneously monitored with few analysis.