A Statistical Approach to Link Flux and Fouling to Sludge Characteristics for an Anaerobic Membrane Bioreactor Treating Dairy Cheese Wastewater

Abstract

Fouling is the main limitation to the application of membrane bioreactors (MBRs). Understanding the complexity of fouling has led to better decision making for design and operation of MBRs. However, studies have shown contradictory results of the impact of sludge characteristics on membrane filtration performance and fouling propensity. The purpose of this study was to characterize the sludge under different operating conditions of an anaerobic membrane bioreactor (AnMBR) treating dairy wastewater, to assess the impact on the filtration. The real flux method was used to determine the flux, while the characteristics of the sludge varied in time. The real flux method is when the feed, retentate and transmembrane pressures are controlled to induce similar hydrostatic conditions applied in full-scale anMBR in crossflow configuration. Total solids (TS), volatile solids (VS), total suspended solids (TSS), volatile suspended solids (VSS), viscosity, and different fractions of the chemical oxygen demand (COD) were performed for sludge characterization. The specific resistance to filtration (SRF), capillary suction time (CST), and supernatant filterability were used as parameters for filterability of the sludge and supernatant. Principal component analysis (PCA) was used to determine the correlation between the sludge characterization and the filterability methods. Five principal components (PC) attributing to 91% of the variance were extracted, based on an eigenvalue greater than 1. The principal components showed the correlation between the different variables studied. PC1 consisted of fraction of solids (total dissolved solids, VSS/TSS, and fixed suspended solids), the CST, and the normalized versions of the CST (CST/TSS, CST/Viscosity, CST/TSS/Viscosity). Five of the variables in PC1 are derived from the TSS concentration most likely indicating why they were grouped under this principal component. PC2 consisted of the particle size distribution of particles ranging from 0 to 10 micrometer. PC3 consisted mainly of a different fraction of solids (VSS, VS, TS, and TSS) and SRF. SRF is a function of TSS, this can explain why these variables were grouped together in PC3. PC4 consisted of the soluble and colloid particles. PC5 consisted of the hydrostatic conditions of the membrane. A multiple linear regression of the PC revealed statistically significance (ANOVA, p-value <0.05) to estimate the flux. A stepwise multiple linear regression was done to determine what variables can be used to estimate the flux based on the data obtained. The selection of the best model from the multilinear regression was based on the highest R squared value, statistical significance from ANOVA, and the variance inflation factor to take into consideration collinearity. Based on the criteria from the PCA and the multiple linear regression, the independent variables for predicting the flux were the CST/TSS, crossflow velocity, the SRF, and the VS/TS.