In this paper the results are presented of original research into the automatic and “intelligent” detection of breakpoints in Dissolved Oxygen (DO) profiles. The research has been based on a large body of data collected from laboratory SBRs operating on synthetic wastewater. Two different approaches were followed to identify the endpoints. The paper analyses and evaluates the results of automatic breakpoint detection on the basis of geometric features in the DO profiles. This was followed by classification of the detected breakpoints using different soft computing techniques based on Neural Network (NN), Fuzzy Neural Network (FuNN) and Evolving Fuzzy Neural Network (EfuNN) software systems for breakpoint classification. A high rate of successful detection and classification was obtained with up to 96% of the decisions made correctly.
In order to overcome the limitations of this system to adapt to dynamically changing process conditions, an intelligent control model was developed by a combination between an Evolving Fuzzy Neural Net (EfuNN) combined with a logic decision unit. This system has the ability to “learn on-the-fly” and adjust its response pattern in order to maintain a high rate of successful breakpoint detection under varying changing process conditions.
This software system has been successfully embedded on a small programmable controller for integration into larger process control systems for the operation of SBR plants.