Published on 03/07/2017 | Use Cases
Data generated by your IoT sensors are a special case data source for Anomaly Detection. This case is even more interesting because a fault in the IoT infrastructure can be an anomaly itself.
You can have vehicles behaving normally (moving), but the anomaly would be to be stopped (either due to traffic or malfunction), as show in the image below:
Database techniques can be used to populate your data repository for priming an anomaly detection algorithm, but only real-time generation of precisely tailored data verifies that end-toend processing works as intended.
MIMIC MQTT Simulator can simulate large numbers of heterogeneous sensors generating desirable data patterns in real-time over MQTT. For example, you can have myriads of sensors generating MQTT payloads containing a "normal" pattern, and instruct a small subset of them to "misbehave" predictably, then observe how long it take to detect this anomaly. By deterministically varying the anomaly patterns in the simulator you are able to tune and regression test iterations in your detection algorithm. You are able even to explore boundary conditions of the infrastructure requirements, such as message rates, failure conditions, etc.
For reference, check these white papers:
https://aws.amazon.com/blogs/iot/anomaly-detection-using-aws-iot-and-aws-lambda
https://www.oreilly.com/ideas/the-elements-of-anomaly-detection-in-the-internet-of-things
---nyse--ge_1.jpg "General Electric")
