SUMMARY
For many years, plant engineers have used data collected from industrial sensors for supporting the diagnosis of failures. Recently, data scientists are using these data to make predictions on industrial processes. However, the meaning and the relationships of each specific sensor is unknown to people outside the engineering context. Conventional approaches to create a semantic layer for industrial sensors require a rigid “term alignment” followed by a lot of manual efforts. Hence, the problem is frequently set aside by industries. However, this condition limits the usage of advanced analytics tools in industries, preventing the capture of potential benefits. Since there are naming conventions and some other rules defined by engineers, this study takes these standards into account and analyze the metadata of sensors intending to automate the creation of a semantic middleware able to indicate the meaning of each sensor and its relationships with other sensors, equipments, areas, plants and other entities. This study intends to answer the following research question: Which approach could automate the creation of a semantic middleware for industrial sensors? In order to address the objectives of this study, we performed an empirical research using sensor metadata from three different plants from a mining company. As a result, we present MINDSense, a method that creates an ontology capable of describing the meaning of industrial sensors and its relationships. We conclude that this method contributes to leverage advanced analytics in industries and to increase the potential of new studies on top of industrial sensors data.