Estimation of long-term resistance changes in a submerged MBR system using theoretical and empirical methods
Author
Tóth, Mátyás
Term
4. term
Education
Publication year
2017
Submitted on
2017-06-10
Pages
43
Abstract
Dette projekt afprøver og sammenligner matematiske modeller til at beskrive, hvordan membranmodstand ændrer sig på grund af sorption i membranbioreaktor (MBR) systemer. Sorption betyder her, at stoffer binder sig til membranen og øger modstanden mod gennemstrømning. Målet var at forudsige, hvor lang tid der går, før modstanden når forudbestemte niveauer. Data kom fra et neddykket MBR-anlæg kørt ved konstant tryk, med målinger logget hvert 30. sekund over et år. Til datarensning brugte vi et glidende gennemsnit og fjernede punkter med store afvigelser (baseret på standardafvigelsen). Hver model blev først tilpasset en lille del af tidsserien; efter hver tilpasning blev datamængden udvidet med én observation, og modellen blev tilpasset igen. Den udbredte førsteordens eksponentielle model gav ingen brugbare resultater. En strakt eksponentiel model klarede sig bedre, men gav stadig fejl. Begge modeller antager én enkelt stationær slutværdi (steady-state) for modstanden, hvilket i praksis ikke nødvendigvis gælder. En mere avanceret model blev udviklet for at håndtere dette, men den kunne ikke bruges, fordi en nødvendig type data manglede. Til sidst kunne en empirisk model, der bruger hastigheden hvormed modstanden stiger, forudsige cyklers sluttidspunkter mere præcist.
This project tests and compares mathematical models for describing how membrane resistance changes due to sorption in membrane bioreactor (MBR) systems. Here, sorption means substances stick to the membrane and increase opposition to flow. The aim was to predict how long it takes for resistance to reach preset levels. Data came from a submerged MBR operated at constant pressure, with measurements logged every 30 seconds over one year. To clean the data, a moving average was applied and points with large deviations (based on the standard deviation) were removed. Each model was first fitted to a small portion of the time series; after each fit, the data window was expanded by one observation and the model was fitted again. The widely used first-order exponential model did not yield usable results. A stretched exponential performed better but still produced errors. Both models assume a single steady-state resistance value, which may not exist in practice. A more sophisticated model was developed to address this, but it could not be used because a required type of data was unavailable. Finally, an empirical model that uses the rate of resistance increase was able to predict cycle end times more accurately.
[This abstract was generated with the help of AI]
Documents