Optimized convolutional neural networks for fault diagnosis in wastewater treatment processes

Abstract

During the long operation of wastewater treatment systems, there is a risk of various failures due to aging equipment and environmental climate change. Hence, to ensure the stability of the wastewater treatment procedure, it is essential to perform prompt and efficient fault diagnosis. This paper presents an optimized fusion deep learning network designed for feature extraction and fault diagnosis within wastewater treatment processes. The method combines the advantages of knowledge-based fault diagnosis approaches while establishing a reliable model. A multiscale convolutional neural network captures spatio-temporal trends across different ranges, enhancing the extraction of spatio-temporal features from the original data. Additionally, using proxy-nearest neighborhood component analysis as a loss function, rather than the conventional cross-entropy, can further optimize the model parameters to improve the fault classification performance. Thus, the multiscale convolutional neural network optimized with proxy-nearest neighborhood component analysis, that is, a fault classification model, was finally built to identify the fault types, and the fault classification performance of the model was validated using the faults generated from a simulation platform for wastewater treatment processes. The highest classification accuracy, with an average accuracy of 85.4%, was achieved when compared to conventional fault diagnosis models and a single sub-model using different loss functions. This study demonstrates the enhanced potential of the proposed model for fault diagnosis in wastewater treatment processes.