Modeling Machine Health Using Gated Recurrent Units with Entity Embeddings and K-Means Clustering

We describe a machine learning approach for predicting machine health indicators two weeks into the future. The model developed uses a neural network architecture that incorporates sensor data inputs using gated recurrent units with metadata inputs using entity embeddings. Both inputs are then concatenated and fed to a fully connected neural network classifier. Furthermore, our classes are generated by clustering the continuous sensor values of the training data using K-Means. To validate the model we performed an ablation study in order to verify the effectiveness of each of the model’s components, and also compared our approach to the typical method of predicting continuous scalar values.