Multivariate statistical analysis of a high rate biofilm process treating kraft mill bleach plant effluent

• Published in: Water science and technology 2007-01, Vol.55 (6), p.47-55
• Main Authors: Goode, C, LeRoy, J, Allen, D.G
• Format: Article
• Language: English
• Subjects: biochemical oxygen demand; biofilm; Biofilms; Biofilms - growth & development; Bioreactors; Bioreactors - microbiology; Bleaches; Bleaching; Dynamics; Equipment Design; Error analysis; Faults; Hydrogen-Ion Concentration; Industrial Waste; Mathematical models; Membranes, Artificial; Modelling; moving bed biofilm reactors; Multivariate Analysis; Multivariate statistical analysis; partial least squares; Performance prediction; pH control; pH effects; prediction; Prediction models; principal component analysis; Principal components analysis; Pulp; Pulp & paper mills; pulp and paper mill effluents; Reactors; shutdowns; Statistical analysis; Statistical methods; statistical models; Statistics; Time Factors; Variables; Waste Disposal, Fluid - methods; Wastewater; Wastewater treatment; Water Pollutants, Chemical - analysis; Water Purification - instrumentation; Water Purification - methods; Wood
• ISSN: 0273-1223; 1996-9732
• DOI: 10.2166/wst.2007.211

This study reports on a multivariate analysis of the moving bed biofilm reactor (MBBR) wastewater treatment system at a Canadian pulp mill. The modelling approach involved a data overview by principal component analysis (PCA) followed by partial least squares (PLS) modelling with the objective of explaining and predicting changes in the BOD output of the reactor. Over two years of data with 87 process measurements were used to build the models. Variables were collected from the MBBR control scheme as well as upstream in the bleach plant and in digestion. To account for process dynamics, a variable lagging approach was used for variables with significant temporal correlations. It was found that wood type pulped at the mill was a significant variable governing reactor performance. Other important variables included flow parameters, faults in the temperature or pH control of the reactor, and some potential indirect indicators of biomass activity (residual nitrogen and pH out). The most predictive model was found to have an RMSEP value of 606 kgBOD/d, representing a 14.5% average error. This was a good fit, given the measurement error of the BOD test. Overall, the statistical approach was effective in describing and predicting MBBR treatment performance.