Operationalisation of FT-NIRS based real-time monitoring for optimisation of anaerobic digestion

  • Michael Darke

    Student thesis: Doctoral Thesis


    Anaerobic Digestion (AD) technology is an industrialised solution that mimics the naturally occurring decomposition of organic materials by anaerobic metabolisms and produces useable products in the form of biogas and digestate. Previous work has identified FT-NIRS technology for applications in real-time multi-parameter monitoring of sludges. This thesis addresses the gap between 'proof of concept' and operationalisation, using case studies of sewage-derived digestates and feedstocks from Advanced Anaerobic Digestion (AAD) wastewater treatment plants (WWTPs) and a Plug Flow Digester AD plant, fed using Source Segregated Waste (SSW) and Agricultural Wastes (AgW). FT-NIRS monitoring programmes of a commercial Ultrasound (also known as "Sonication") (US) trial and a laboratory evaluation of Plug Flow (PF) versus a Continuous Stirred Tank Reactor (CSTR) extended the calibration intervals.

    The study utilised a decomposition modelling approach, extending the range of calibrations using a 'Whole of System' (WoS) approach (input-output). Principal Component Analysis (PCA) models and Partial Least Squares regressions (PLS-R), using matched FT-NIRS and matched physicochemical parameters, were evaluated using root mean square error of cross-validation (RMSECV) and R2. Confidence Intervals (CI), using uncertainty modelling, and prediction errors are defined using Mean Absolute Prediction Error (MAPE) and the CI. With the extended ranges under investigation using WoS principles, additional metrics are required to assess model performance. "Normal Operating Condition" (NOC) models using PCA were most effective at early recognition of out-of-range samples.

    The research shows that physicochemical parameters (e.g. Total and Volatile solids (TS, VS), volatile fatty acids (VFAs) and Carbon Oxygen Demand (COD) can be determined from fresh samples, with no sieving, drying, or filtering. The mean RMSECV was 0.47% for TS (0.38 – 0.55%, over a calibration range of 4.04 – 8.82%) and 0.32% for VS (0.29 – 0.36%, with a calibration interval of 2.55 – 7.06%). The preferred models utilised Support Vector Machines (SVM), Generalised Least Squares Weighting (GLSW), and Mean Centering (MC) preprocessing. Results using SSW-AgW appear higher due to a wide calibration interval. Within theviii preferred models, GLSW-MC performed best; the mean RMSECV was 0.86% for TS and 0.87% for VS over calibration intervals of 1.33 – 15.82% and 0.66 – 13.73%, respectively.

    In addition, a Total Volatile Fatty Acids (TVFA) model using wastewater sludges delivered a RMSEC of 211 mg.L-1 and RMSECV of 281 mg.L-1 over a calibration range of 68 – 4746 mg.L-1. At the same time, Soluble Carbon Oxygen Demand (CODS) gave a RMSEC of 1574 mg.L-1and RMSECV of 2018 mg.L-1 over a calibration range of 4821 – 35183 mg.L-1.

    Other techniques developed include recognising Ultrasonic intensity in wastewater treatments from PCA analysis, the potential to monitor PCA/PLS loadings as a data reduction tool to simplify analysis and recognise digester changes related to VFAs and using the relationships in the raw spectra behaviour to improve VFA build-up detection and speciation through an understanding of their interaction with other physicochemical parameters. The models developed would be helpful for the industrial monitoring of AD plants. Independent datasets show that the models could not be applied globally. However, the WoS method reduces calibration lead times by establishing the AD plant's process gradient. A simple at-line Spectra Acquisition Cell (SAC) improved spectral acquisition. Fully developed and deployed, the SAC could further reduce the impact of confounding variables, such as temperature, increase data collection and enhance interpretation consistency while avoiding the capital expenditure associated with online deployment.
    Date of Award2024
    Original languageEnglish
    SponsorsKESSII & Dŵr Cymru Welsh Water
    SupervisorSandra Esteves (Supervisor), James Reed (Supervisor) & Tim Patterson (Supervisor)

    Cite this