Population Based Pumps Monitoring and Benchmarking Using IoT and Edge ML Learning Methods

Abstract

Machinery monitoring is typically applied to a single machine based on sensor integration and data analysis. Such an approach to a set of machines operating in similar conditions allows for a multivariate analysis for condition monitoring based on a single machine as well as based on group analysis. This paper describes an Industrial Internet-of-Thing (IIoT) concept for condition monitoring of machinery population based on water pumps. The first part provides an introduction to unsupervised anomaly detection based on population modeling with using features calculated from the: mechanical (based on vibration sensors), electrical (voltage and current signals collected from electric motors that drive monitored pumps) and operational processes (such as pressures, flows) signals. Finally, the preliminary results from laboratory testing and demonstration at a wastewater processing plant are presented.