Section: New Results
Introducing Data Quality to the Internet of Things
Participants : Jean-Marie Bonnin, Jean-François Verdonck, Frédéric Weis [contact] .
The Internet of Things (IoT) connects various distributed heterogeneous devices. Such Things sense and actuate their physical environment. The IoT pervades more and more into industrial environments forming the so-called Industrial IoT (IIoT). Especially in industrial environments such as smart factories, the quality of data that IoT devices provide is highly relevant. However, current frameworks for managing the IoT and exchanging data do not provide data quality (DQ) metrics. Pervasive applications deployed in the factory need to know how data are "good" for use. However, the DQ requirements differ from a process to another. Actually, specifying/expressing DQ requirements is a subjective tasks, depending to the specific needs of each targeted application. As an example this could mean how accurate a location of an object that is provided by an IoT system differs from the actual physical position of the object. A Data Quality of 100% could mean that the value represents the actual position. A Data Quality of 0% could mean that the object is not at the reported position. In this example, the value 0% or 100% can be given by a specific software module that is able to filter raw data sent to the IoT system and to deliver the appropriate metric for Dev apps. Building ad hoc solutions for DQ management is perfectly acceptable. But the challenge of writing and deploying applications for the Internet of Things remains often understated. We believe that new approaches are needed, for thinking DQ management in the context of extremely dynamic systems that is the characteristic of the IoT.
In 2018, we started to define DQ software services that are able to query data and retrieve a collection of DQ metrics that the developer need. The goal is to enable developers to access, configure and tweak any DQ mechanisms in an easy way. Facilitating embedding of DQ capabilities will demand a new type of "endpoint" services, deployed to industrial pervasive environments. We obtained first results of our work towards establishing metrics and tools to enable IoT developers to know and use the quality of data they obtain from the IoT. Our approach combines continuous data analytics with modeling expected behavior of sensors in order to weight the inputs of different sensors to reduce the overall error. Key challenges of our work are semantic modeling of the data quality and modeling the expected behavior of sensors. We illustrated our approach at the example of localizing production robots in a factory. We demonstrated the potential of our first solutions with a demonstration at the AdHoc Now conference (see figure 5). We managed to significantly reduce the error introduced by faulty sensors. This should lead to publishing on both DQ and programming aspects of our approach.
This work has been done in collaboration with Technical University of Munich.