Health assessment method of industrial robot reducer based on deep belief network

Industrial robots are the most representative equipment in smart manufacturing system. Reducers, which are one of the key components of industrial robots, account for a significant portion of failures in industrial robots. It is thus important to evaluate health status of reducers during the operation of industrial robots. A deep-level probability directed graph model-Deep Belief Network (DBN) is used to assess health status for industrial robot reducer in this paper. First, in the pre-training stage of the deep belief network, the weights closer to the optimal are trained layer by layer through the Restricted Boltzmann Machine (RBM) from bottom to top. Secondly, in the fine-tuning stage of the deep belief network, the weights are tuned through the backpropagation algorithm. Finally, the deep belief network model is verified through experiments, which shows that the DBN has a high diagnosis accuracy for health assessment of industrial robot reducer. © 2021 IEEE