Date of Award
Spring 2012
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Civil Engineering
First Advisor
Melching, Charles S.
Second Advisor
Feng, Xin
Third Advisor
Crandall, Clifford
Abstract
Sludge bulking is the most notable cause of activated sludge plant failure (i.e. exceeding discharge permit quality limits) worldwide. Numerous mathematical methods have been applied to detect and provide warning for the prevention of sludge bulking. However, these models often fail to reliably forecast sludge bulking events because they focus on the point-by-point "curve-fitting" strategy, while the number of bulking event data points is relatively small in comparison with the large amount of data in the time series. Therefore, three machine learning approaches which focus on detecting the temporal pattern data before the sludge bulking events are considered in this study.
The main objective of this research is to apply machine learning and statistical methods to detect the hidden temporal patterns in the sludge volume index (SVI) data and related water-quality parameters occurring before high SVI values (sludge bulking) occur, and then the hidden temporal patterns can be used to forecast high SVI values in the future. Three methods are applied in this research, the improved Time Series Data Mining (TSDM) method, the Hidden Markov Models (HMMs) method, and the combined method of Hidden Markov Models and multinomial logistic regression (MLR).
The results and analysis show that the improved TSDM method and the HMMs method are capable to detect and predict sludge bulking events. The improved TSDM method can have a sludge bulking event prediction accuracy between 60% and 100%. The HMMs method could provide warning information to the WWTP operators, even if the HMMs method only detects the first state of the pattern leading to sludge bulking. Once the first pattern state was detected, there was high probability (>80% in all cases, mostly > 90%) that sludge bulking would occur. However, both of these methods have limitations because they are new methods applied to the sludge bulking problem. For the combined method, although the results are not useful for the detection of sludge bulking, some wastewater quality parameters are found to have significant impact on the sludge bulking, i.e., sludge retention time (SRT) and effluent pH for all three batteries.