A framework for constructing a common knowledge base for human-machine system to perform maintenance tasks

A reliable and comprehensive maintenance is important to promise the system running in a normal state, but it is skill-intensive and heavily dependent on human labor. With the development of predictive maintenance in industry, an optimized solution can be posed for maintaining assets with less downtime and cost. However, most of current research on this topic is limited on a top-level algorithm design for prediction, but few consider how to perform the maintenance tasks according to the prediction results at a particular occasion and condition. Besides, the complexity of system is exploded, and it may take people much effort to cover every detail to achieve a credible maintenance result. Thus, machine is introduced to collaborate with human by undertaking some work and suggesting actions to take in order to reduce human physical and mental workload. This paper aims to present a framework to integrate human knowledge and machine learning into a common knowledge base to enable human and machine can contribute to shift the final maintenance decision from planning to performing. The proposed framework is based on a knowledge graph generated by ontology and machine learning, which can be conveniently retrieved by human via questions answering system or visualization platform and efficiently computed by machine via graph representation learning. Consequently, domain knowledge can be formally represented, systematically managed and easily reused by human-machine teaming to attack domain-specific problems. In a long term, the evolving knowledge based, with an accumulation on samples and information, can guide the team to draw a reasonable and delicate strategy for overhaul and recondition, moreover, ensure the next generation of maintenance: prescriptive maintenance.