Prediction of sludge volume index bulking using image analysis and neural network at a full-scale activated sludge plant

Abstract

Sludge volume index parameter should be monitored daily for the performance of wastewater treatment plants. It was aimed to estimate this parameter using image processing and artificial intelligence techniques for full-scale wastewater treatment plant. The activated sludge samples were collected from the aeration tank of the activated sludge process in Konya Domestic Wastewater Treatment Plant. Sludge characteristics and settling properties were observed microscopically via the measurements of flocs and filaments. The 49 images per slide were taken by an image-analysis system developed for automated image acquisition. A total of 120 samples were examined over a period of year. The floc and filament structures were analyzed using Cellular Neural Networks (CNN). Iteration value of the CNN was modified according to the image. Then, a number of morphological operations were applied for an accurate identification of the floc and filaments separately. Textural, shape, and statistical approaches were utilized for creating a set of data for each sample. After preparing the training and test data by shuffling the data randomly, a fivefold cross-validation method was applied. And, these training and test data were applied to an artificial neural network. The weights of the neural network were trained using the Levenberg–Marquardt, Genetic, and Artificial Bee Colony algorithms.