With the development of industrial IoT, industrial systems are becoming more and more complex, and industrial equipment operation and maintenance data are becoming increasingly large, so it is particularly important to predict the posture of industrial equipment. Knowledge graph has been proved to have significant advantages in dealing with a large amount of heterogeneous data, and can be applied to the processing of industrial IoT equipment operation and maintenance data. Aiming at the temporal logic problem of predicting the failure posture of industrial equipment operation, this paper proposes a temporal knowledge graph inference representation learning model that uses a local recursive graph encoder network to model the historical dependencies of events at adjacent time points and a global history encoder network to collect repeated historical facts. Experiments show that the proposed model outperforms baseline inference methods in metrics such as mean reversal ranking (MRR).