D5.3 – Monitoring and Fault Recovery in Highly Dynamic IoT Environments
Monitoring of IoT sensors measuring different phenomena in different contexts is a crucial task especially in large IoT environments to ensure uninterrupted operations of the systems that depend on them. However, developing domain agnostic algorithms for monitoring is a highly complex task as the desire for higher accuracy also demands more domain knowledge which differs from scenario to scenario. This work describes the faults that can occur in IoT sensors and presents developed algorithms for the detection of faults and their recovery mechanisms. Algorithms implemented in this deliverable are developed to work in different environment with as little domain knowledge as possible. This deliverable also makes use of the concepts developed earlier in IoTCrawler, like getting information from metadata and utilising Quality of Information of sensors to know their working status, to gain better insights of the data and for better tuning of learning algorithms.
This deliverable concludes the work defined in T5.4. The objective is to develop strategies for the monitoring of IoT sensors to detect faults before they propagate in the system and cause operation failures and downtimes. Additionally, strategies and algorithms are developed to provide support to data providers and users by alerting them of the fault status of their subscribed sensors. As a temporary solution, recovery mechanisms are also implemented which facilitates the users by sending the predicted data from the learned data distribution which may reflect the ground truth better.
Domain knowledge is important when detecting faults in any IoT environment. However, as IoTCrawler is designed to deal with a large number of IoT sensors deployed in different environments, it is of paramount importance that the developed solutions should work with as little knowledge about the sensor and its environment as possible. Current state of the art discussed later overcomes this problem by monitoring sensor states annotated by the sensor providers and often defines solutions which are computationally expensive.
Acknowledgement of other key-contributors: Eushay bin Ilyas, Thorben Iggena, Marten Fischer (UASO), Aurora Gonzalez (UMU) Narges Pourshahrokhi (UniS)