An intelligent data capturing framework to improve condition monitoring and anomaly detection for industrial machines

Abstract

Companies understand the need to monitor their machines when operational under real life conditions, since it provides valuable information on how these machines and their processes can be improved and why they might fail. Currently, due to bandwidth and storage constraints (costs), companies are limited by how much data they can send from their machines into the cloud, resulting in transmission of only a reduced set of aggregated features. Such a reduced set often misses the critical information required to analyse the machine's behaviour, such as possible defects, which is often only present in the high frequent raw data signals. Because sending all high frequency data to the cloud is not possible, the work in this paper proposes a hybrid approach where we do transmit high frequency data, resulting in a complete yet compact set, aiming to reduce the redundancy of the transmitted data as much as possible. As a result a valuable and up to date dataset will be available in the cloud for machine monitoring and anomaly detection purposes while restricting ourselves to feasible transmission and storage requirements. This hybrid approach has been implemented and applied on a grid monitoring application, focusing on grid disturbances.