Artificial Neural Network Approach for Fault Recognition in a Wastewater Treatment Process

Abstract

The paper deals with fault detection and recognition for WWTP (Wastewater Treatment Plant). The chosen classifier is a feed-forward neural network. Its input is a high-size vector of measured variables, rather than a small-size compressed feature vector. The output of the network points to the recognized fault class. The test was performed on a simulated WWTP, disturbed by 6 different types of faults (sensors and actuators). The results of the test proved a good ability of the neural network to recognize the faults, in 97.2% of the analysed cases.