Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.777790
Title: Biological investigation and predictive modelling of foaming in anaerobic digester
Author: Kanu, Ifeyinwa Rita
ISNI:       0000 0004 7963 5630
Awarding Body: Heriot-Watt University
Current Institution: Heriot-Watt University
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Anaerobic digestion (AD) of waste has been identified as a leading technology for greener renewable energy generation as an alternative to fossil fuel. AD will reduce waste through biochemical processes, converting it to biogas which could be used as a source of renewable energy and the residue bio-solids utilised in enriching the soil. A problem with AD though is with its foaming and the associated biogas loss. Tackling this problem effectively requires identifying and effectively controlling factors that trigger and promote foaming. In this research, laboratory experiments were initially carried out to differentiate foaming causal and exacerbating factors. Then the impact of the identified causal factors (organic loading rate-OLR and volatile fatty acid-VFA) on foaming occurrence were monitored and recorded. Further analysis of foaming and nonfoaming sludge samples by metabolomics techniques confirmed that the OLR and VFA are the prime causes of foaming occurrence in AD. In addition, the metagenomics analysis showed that the phylum bacteroidetes and proteobacteria were found to be predominant with a higher relative abundance of 30% and 29% respectively while the phylum actinobacteria representing the most prominent filamentous foam causing bacteria such as Norcadia amarae and Microthrix Parvicella had a very low and consistent relative abundance of 0.9% indicating that the foaming occurrence in the AD studied was not triggered by the presence of filamentous bacteria. Consequently, data driven models to predict foam formation were developed based on experimental data with inputs (OLR and VFA in the feed) and output (foaming occurrence). The models were extensively validated and assessed based on the mean squared error (MSE), root mean squared error (RMSE), R2 and mean absolute error (MAE). Levenberg Marquadt neural network model proved to be the best model for foaming prediction in AD, with RMSE = 5.49, MSE = 30.19 and R2 = 0.9435. The significance of this study is the development of a parsimonious and effective modelling tool that enable AD operators to proactively avert foaming occurrence, as the two model input variables (OLR and VFA) can be easily adjustable through simple programmable logic controller.
Supervisor: Adeloye, Adebayo J. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.777790  DOI: Not available
Share: