Enhancing Predictive Maintenance in Water Treatment Plants through Sparse Autoencoder Based Anomaly Detection

Abstract

The deployment of Machine Learning (ML) for improving Water Treatment Plants (WTPs) predictive maintenance is investigated in the present article. Proactively detecting and fixing functional difficulties which might cause catastrophic effects has historically been an endeavour for reactive or schedule-based maintenance methods. Anomaly Detection (AD) in WTP predictive maintenance frameworks is the primary goal of this investigation, which recommends a novel approach based on autoencoder (AE)-based ML models. For the objective of examining high-dimensional time-series sensor data collected from a WTP over a long time, Sparse Autoencoders (SAEs) are implemented. The data collected involves an array of operational measurements that, evaluated together, describe the plant's overall performance. With the support of the AE, this work aims to develop a practical framework for WTP operation predictive maintenance. Anomalies are all system findings from testing that might result in flaws or malfunctions. The research article analyses January and July 2023 WTP data from Jiangsu Province China. The AE paradigm had been evaluated using F1-scores, recall, accuracy, and precision. SAE has substantially improved AD functionality.