Modeling and control of Industrial Wastewater Physico-Chemical treatment processes

The purpose of the present thesis is to shed light on a very complex subject such as the modeling and control of industrial wastewater physico-chemical treatments, with particular attention to the coagulation-flocculation process with simoultaneous precipitation of heavy metals. After a brief but comprehensive introduction to the theory of coagulation, some concerns about the applicability of DLVO theory in the context of process simulation are raised, thus urging the need to find simpler but equally explanatory formulations that can be practically implemented within numerical models. In this regard, a fairly in-depth review of the progress that flocculation theory has made over the past forty years is given, going as far as modeling flocs as porous and hydraulically permeable fractal aggregates. After recalling the philosophy behind a paper published by Hogg in 1984 about a modelling approach for the collision efficiency factor for particle flocculation using polymers, an extension to this model is proposed, to take into account for more mechanisms of particle destabilization such as sweep coagulation, adsorption and interparticle bridging that possibly act simultaneously. A discretized population balance equation is then proposed to model the coagulation-flocculation-precipitation process using the above formulation in the framework of a linear driving force model. The topic of process control is subsequently introduced, showing the main reasons and challenges behind industrial wastewater treatment. After a brief overview of the main physical sensors and analyzers related to coagulation and flocculation available on the market, the problem of fouling is remarked as an issue that places considerable limits on the choice of the amount, types, sampling points and sampling frequency of sensors to be installed. The issue of dead time is also mentioned, as a potential source of disturbance for the control system. A brief review of the main studies conducted in the last 30 years in the field of online monitoring and control of the coagulation process is then given, raising particular attention on the so-called "software sensors". The important role that machine learning and data-driven techniques could play for model and/or control laws discovery is then highlighted, showing that the relatively high cost associated with pilot plant testing to train these systems usually pays off in the long term. Finally, the results of an experimental study on the coagulation-flocculation of washing wastewaters from the automotive painting plant of FCA Mirafiori (Turin, Italy) are illustrated. An equally effective but less expensive alternative to the treatment currently performed in the Mirafiori plant is proposed, also showing the existence of a relationship between turbidity, solids content and the chemical oxigen demand (COD) of the influent, as well as between COD and the integral of the UV absorption spectrum below 280 nm. To conclude, some uses of these formulas are proposed for automatic control purposes.