Sensor Fault Detection and RUL Estimation for Drinking Water Pumping Stations

Predictive maintenance is a very important need and is used frequently in many areas. One of them is monitoring sensors of water critical infrastructure. In this article, we focus on sensor data fault classification and remaining useful life (RUL) estimation of sensors of water management infrastructures. We implement different data-driven models to classify sensor faults and estimate the RUL of sensors on our synthetically created datasets that accurately match real predictive maintenance data. The best model, decision tree (DT), has an accuracy of 99% with the smallest training and prediction times for sensor data fault classification. In addition, the best RUL estimator is a convolutional neural network (CNN) with long-short term memory with 86% accuracy value. Experimental results show that our datasets will lead to works in the field of sustainable water governance in the literature.