Applying Time-Constraints Using Ontologies to Sensor Data for Predictive Maintenance

Predictive maintenance depends on sources of data analysed through machine learning and data mining algorithms to identify and predict potential fault states. However, sources of raw data lack a layer of semantic abstraction providing a consistent way of interpreting data. The use of ontologies to describe data from structured and unstructured sources has been employed to provide a semantic layer delivering a consistent interpretation and meaning to data that can be exchanged between different entities. Data supplied by sensors is time-sensitive, as variations and fluctuations occur over periods and must be analysed regarding the period they occur. Transforming raw data and applying time constraints to fit the data to the semantic concepts is a process not frequently documented. In this paper, we present an architecture for the transformation of data acquired through different sources - from sensors to contextual information supplied by management software - from JSON to ontology instances that can be used in a predictive maintenance scenario, alongside with machine learning and data mining solutions. Here, the different data sources are transformed semantic time-sensitive data, represented through means of an ontology and stored in a triple store.