PENSEC, William (UBS / Lab-STICC), ESPES, David (Université de Bretagne Occidentale), Dezan, Catherine (Université de Bretagne Occidentale)

Smart Anomaly Detection and Monitoring of Industry 4.0 by Drones

Scheduled for presentation during the Regular Session "Manned/Unmanned Aviation" (ThA5), Thursday, June 23, 2022, 11:10−11:30, Elafiti

2022 International Conference on Unmanned Aircraft Systems (ICUAS), June 21-24, 2022, Dubrovnik, Croatia

This information is tentative and subject to change. Compiled on April 18, 2024

Abstract

Nowadays, industry 4.0 can be distributed over a large area. To monitor their processes, they use sensors that periodically gather data on the system. Based on them, operators can detect that an anomaly occurs on the system. However, it is not always easy to know the causes of the anomaly because the operator has no visual information on the system. To help operators to identify the root of the anomaly, drones are very useful because they are fast enough to intervene in large-scale industry and embed a large variety of sensors to offer complementary data (images, video...) that are necessary for the diagnosis. However, drones have to be synchronized with the industrial process to know where the anomaly occurs and to go there in an automated way. We propose a new architecture to automate the displacement of the drone to reach safely the place where the anomaly is located and to confirm it using a deep-learning approach. The drone embeds a small computing system (Raspberry Pi) which communicates with the supervisory control and data acquisition system in order to be aware of anomalies that occur on the industrial process. To function properly indoor or outdoor, the drone is positioned either using a precise positioning system based on ultra-wide band (UWB) or on the GPS. The drone can take pictures of the potentially detected anomaly and thanks to a neural network algorithm, it analyzes the images to confirm or deny the anomaly. The results show an error on the indoor position of about 5 cm, and a precision of about 90% to detect anomalies.