Deep Learning for Modeling Filamentous Sludge Bulking

Biological wastewater treatment has been widely used to remove contaminants from water to avoid pollution. The activated sludge process (ASP) is the most extensively utilized biological wastewater treatment technology to remove organic matter and nutrients due to its economic and technological viability. Filamentous bacteria are common constituents of activated sludge biomass, where the presence of a percentage of filamentous bacteria is vital and beneficial in the formation of flocs by acting as the floc-backbone to which other bacteria can adhere. However, filamentous bulking sludge, a term used to characterize excessive filamentous bacteria proliferation, frequently leads to decreasing sludge settleability, lower operational performance, and greater treatment costs. Filamentous bulking sludge is regarded as the most critical issue that frequently occurs in wastewater treatment plants that use ASP.

The practical way used to characterize the sludge bulking problem is by measuring the Sludge Volume Index (SVI). Currently, water treatment plant specialists in Trentino are in charge of measuring the SVI index on samples collected in the purification tanks once a week. The chemical measurement requires up to 2 days to be completed.

Predictive models have been presented to obtain early warning for filamentous sludge bulking to avoid major degradation of sludge settleability. In this context, deep learning (DL) has emerged as an innovative method for improving data-driven modelization in a wide range of applications where many diverse variables can impact the final output.

The goal of the presented project is to use innovative DL algorithms to predict the SVI as precisely as possible. Special attention will be given to the whole workflow for developing a strong and sophisticated DL model, from data cleaning and feature engineering to produce the most accurate model feasible. The model will be created utilizing real data collected in a Trentino water treatment facility.

Instructor: Andrea Di Luca (University of Trento), Marcus Vuk (Niris, IT), Andrea Turso (Niris, IT)

Participants:

  • Aurora Poggi, University of Verona
  • Alba Gurpegui, Lund University
  • Armando Assembleia, University of Lisbon – Instituto Superior Técnico
  • Thomas Trinh, Chalmers University of Technology
  • Dennis Modesti, University of Verona

Room:

Room S.8, Veronetta – Polo didattico Zanotto, Viale Università 4, 37129 Verona VR G-MAP: https://goo.gl/maps/iDPvePC6gdKv4xfN9