A plant-wide analysis of an industrial flocculation process in biomanufacturing

Abstract

This master thesis project is a case-study of the biomass dewatering operations of the industrial waste treatment plant (iWTP) that is shared by the biotech companies Novozymes A/S and Novo Nordisk A/S in Kalundborg, Denmark. More than 4000 m3 of biomass sludge are dewatered every day in the plant using decanter centrifuges. The biomass sludge is a mix of spent biomass from the biotech production and waste activated sludge. The concentrated cake is sold to an external biogas facility and the clarified liquid remains in the plant for removal of the remaining pollutants. The efficiency of this separation depends largely on chemical dosing with a coagulant and a flocculant solution. The composition of the biomass sludge is subject of the changing biotech production upstream, which difficults optimization of the chemical dosing. Moreover, the particle processes in complex dynamic systems such as decanter centrifuges are poorly understood. Therefore, the aim of this project was to develop and test tools to aid on the monitoring, study and decision-making in the dewatering of biomass sludge. Several steps were followed to accomplish this. First, an on-line degassing system and a Random Forests prediction model were integrated into an on-line sensor for determination of reject water quality. The Random Forests model was trained with color features and total suspended solids (TSS) measurements of degassed reject water (dRW). Next a measuring campaign was carried out over 15 production batches to collect data from samples of biomass sludge before and after dewatering. Novel technology from ParticleTech® was used to collect particle images from these samples. The images were used to build and train an image analysis pipeline for the clustering of particles based on their morphology. Additionally, the new particle data was used to study the influence of pH and polymer dose on the particle population of the of the biomass sludge, as well as to find associations between particle cluster removal and reject water quality. The best prediction model achieved an accuracy of 80.81% on the prediction of TSS of dRW using the three color features that define the HSV color space. This system can potentially save time on the decanter optimization procedure. However, more data and re-training of the model are necessary before full-scale implementation. Clustering of particles images resulted in the identification of 10 main particle sub-populations. The study of flocculation based on the particle clusters showed a strong association between the removal of TSS in reject water and the removal of the larger particles from cluster K9. The removal of this cluster was also highly sensitive to polymer dose, benefiting from polymer overdosing most of the times. Thus, morphology-based data was demonstrated to be a new source of information that can potentially help elucidate the particle processes associated to the dewatering of the biomass sludge. This project was carried out within the frame of the Helix Lab Fellowship Program.