Reasoning over OPC UA Information Models using Graph Embedding and Reinforcement Learning

Abstract

With the fourth industrial revolution, the OPC Unified Architecture (OPC UA) [1] has emerged as the standard communication framework for the implementation of cyber-physical production systems since it can be used for both, communication and information modeling. Even though OPC UA helps bridge the interoperability gap at the automation level, its semantic has not yet been formally defined and generated or manually created OPC UA information models are often incomplete and inconsistent making an efficient automated reasoning and knowledge inference on the OPC UA address space particularly challenging.In this paper, we show that it is possible to train a machine learning system on OPC UA information models, such that it performs automated reasoning over OPC UA knowledge graphs with high precision and recall. More specifically, we present a reinforcement learning based solution that learns to reason on semantically incomplete OPC UA information models by constructing multi-hop relational paths along an embedded vector space of the knowledge graph representing the information model. The construction of such relational paths allows the discovery of missing relations between the entities and at the same time the evaluation of the truth of the encoded triples, thus enabling consistency checks and questions answering.