A Real-Time Anomaly Detection System for Industrial IoT Sensors

Abstract

Reliability in Industrial Internet of Things (IIOT) systems is affected by several reasons. One of the significant cause is a faulty sensor. Industrial machineries have sensors as their integral part, which are for their control and behaviour. Due to a fault in sensor it may generate erroneous values, which may cause unexpected triggering of actuators leading to damage to the equipment and production downtime. In this paper, we introduce a system developed that detects sensor anomalies or faults by analyzing its data output. The system developed is based on the Node-RED platform which handles the communication through the Message Queue Telemetry Transport Protocol (MQTT) to collect sensor data and the processing tasks. In Node-Red function nodes are programmed to analyze sensor data and detect faulty behaviour of the sensor. MySQL database management system is used to store and separate out anomalous data. Initially the system is designed to detect issues that are observed commonly in a faulty sensor by analyzing their output data values, that is sudden value spike and flat sensor reading. The system is not limited to this and is designed such that additional anomaly detection filters whether of rule-based type or machine learning based can be easily developed and integrated. Importance of system is that, the actuator should not get triggered by any faulty control signal.