Investigation of a MBR membrane fouling model based on time series analysis system identification methodsстатья из журнала
Аннотация: Flux stepping tests were carried out on a novel sidestream MBR pilot plant treating industrial wastewater, and a membrane filtration unit treating tertiary quality municipal effluent. This included offline tests measuring mixed liquor concentrations, as well as soluble microbial product (SMP) levels in the sludge water which is the main irreversible foulant on the membrane [1]. A basic phenomenological dead-end filtration model that includes the three main fouling mechanisms mentioned in Hermia (i.e., cake build-up, complete pore blocking, and pore constriction) and that was based on a constant TMP operation was extensively modified [2,3]. Modifications and add-ons to this basic model included: alteration so that it could be used for varying flux and varying TMP operations; inclusion of a backwash mode; it described pore constriction (i.e., irreversible fouling) in relation to the concentration of SMP in the liquor; and, it could be used in a crossflow scenario by the addition of scouring terms in the model formulation. Using data collected from both the pilot plant and the filtration unit, this modified deterministic model was calibrated and validated in Matlab©. In order to see whether a simpler model could be formulated for advanced control purposes that was based wholly upon measured historical data sets for both the pilot plant and the filtration unit, a further conceptual model was developed based on system identification procedures and input-output times series analysis methods [4]. This model form utilised an autoregressive subspace state-space formulation. Again using the same data collected from both the pilot plant and the filtration unit, this alternative model was calibrated and validated in Matlab©. A very good correlation was shown between the measured and the expected flux decline/recovery for the phenomenological model, although a complex genetic algorithm procedure was needed for parameter estimation. The subspace model was almost as accurate as the phenomenological model even though it only used a single shot fast algorithm for parameter estimation. Further and longer historical data sets are needed to ascertain whether this second simpler modelling approach can be improved upon.
Год издания: 2011
Авторы: Parneet Paul
Издательство: Taylor & Francis
Источник: Desalination and Water Treatment
Ключевые слова: Fault Detection and Control Systems, Mineral Processing and Grinding, Wastewater Treatment and Nitrogen Removal
Открытый доступ: closed
Том: 35
Выпуск: 1-3
Страницы: 92–100