Cognitive Predictive Maintenance

Nowadays, conventional physics-based or data-driven Prognostics and Health Management (PHM) approaches alone fail to satisfy the diversified demand for complex equipment predictive maintenance. To address it, the Industrial Knowledge graph (IKG) and graph embedding (e.g. graph convolutional network) techniques have shown promising prospects in gathering and merging multiphysics-based information, as well as offering extendable representation and explainable reasoning of multi-disciplinary domain experience and knowledge in a data-driven context-based manner. Therefore, representing complex equipment in the IKG form can organize complex equipment in a graphic and logical manner, and conduct explainable querying for generating the optimal predictive maintenance strategies.

The IKG of complex equipment enables the generation of integrated strategies, including fault diagnosis, optimum maintenance plan, manual maintenance guidance, remaining useful life prediction, etc. Meanwhile, the visual equipment knowledge graph provides explanation for the integrated solutions. Furthermore, in the manufacturing process, the equipment knowledge graph is context-aware and interactive, which can adopt timely and effective reactions during the user interface and unexpected intervene, such as warning, adjustment, and recommendation.