Modelling of Fine Chemical Manufacturing Effluent and Emissions: Implementation of Mathematical and Physical Models for Wastewater Management

Abstract

This thesis consists of a portfolio of four projects that investigates, and addresses challenges faced by fine chemical and pharmaceutical manufacturers based in the UK to achieve greater control of effluent and air emissions in an environment of tightening legislation. In part one of this research, a mathematical model based on modified continually stirred tank reactor (CSTR) in series equations was created using MATLAB and deployed as a standalone tool with graphical user interface (GUI) to allow non-experts to predict the concentration of substances through a complex plant effluent system and activated sludge wastewater treatment plant (WWTP). The model showed an average prediction accuracy to within 9% of the true value on-site when validated against a tracer compound. In Part two of the thesis, a 100 litre small-scale activated sludge wastewater treatment plant was designed and built to enable extended testing of treatment efficiency for potential new wastewater streams for a large-scale (2400 m3 reactor) plant, to reduce the risk of breaching site effluent consent limits or large scale microorganism poisoning. Plant variables and conditions required to maintain equivalent wastewater treatment performance from the large-scale to the small-scale plant were researched and found to include temperature, pH, dissolved oxygen (DO) concentration, hydraulic retention time (HRT) and solids retention time (SRT). The resulting pilot plant matched performance of the large-scale plant to within 12.8% of effluent chemical oxygen demand (COD) concentration. Long term treatment efficiency testing of a methylene-blue dye containing wastewater was conducted and shown to be satisfactory after a period of microorganism acclimation, where equivalent short-term lab respirometry testing had previously shown no digestion of the blue wastewater. In part three of this thesis, mathematical models were researched and developed in MATLAB to study prediction of the effluent chemical oxygen demand concentration of an activated sludge wastewater treatment plant. Existing methods of non-linear modelling using partial least squares (PLS) and artificial neural network (ANN) structures were investigated and compared against a newly proposed structure that replaces the linear inner regressor of the PLS algorithm with a combination of multiple neural networks (mNNPLS). The mNNPLS model showed an improved correlation coefficient (R2) of 0.855 between observed variables and predicted effluent chemical oxygen demand. In the final section of the thesis, a methodology was created for conducting consequence and risk modelling using DNV’s PHAST software to compliment the studies in parts one and two by allowing estimation of substance input concentrations to the effluent system from potential major accidents to aid emergency response planning and test the robustness of the effluent system. Sources for input data including individual and societal risk criteria are researched and presented along with techniques of scenario identification, results presentation and overall report format.