Predictive Maintenance of Industrial equipment using Temporal Graph Neural Networks: A step towards Novelty

Predictive maintenance is important for improving industrial activities and achieving minimum downtime. The most common methods focuses on temporal features that are static and extracted from a set of sensors which may not account for the dynamic of time evolution in the degradation of the equipment .In this paper , we develop a Temporal Graph Neural Networks (TGNN) model to address the limitations. Since sensor data comprises of temporal data which can be structured as graph with time points as nodes and relationships between time points as edges, TGNN can learn advanced timing configurations to enhance accuracy of the model’s forecast. The method has been tested on dataset, which has a real industrial applications and conducted a comparison of its result with the traditional system. Our findings confirm that TGNN are effective than other baseline methods and thus, generalising the type of tasks for predictive maintenance.