Abstract: In order to realize the fault state identification of the cooler in the hydraulic system, this paper proposes a fault diagnosis model that uses the deep belief network （PSO-DBN） optimized by the particle swarm algorithm to achieve multi-sensor information fusion. In the proposed model, the signals from different sensors are characterized and selected, and the multi-sensor fusion method is used to integrate the feature level into the deep belief network to identify the fault state of the cooler. At the same time, the particle swarm algorithm is used to adaptively select the hyperparameters of the deep belief network, including the number of hidden layer nodes, the number of reverse iterations and the reverse learning rate, to determine the optimal structure of the network, thereby improving the diagnostic accuracy of the deep belief network. In this paper, the hydraulic system dataset of the Center for Machine Learning and Intelligent Systems of the University of California, Irvine is used to verify, and the experimental results show that compared with the deep belief network, the deep belief network optimized by genetic algorithm, and the support vector machine optimized by particle swarm algorithm, PSO-DBN can effectively extract the inherent characteristics of the data, and the average fault state recognition accuracy of the cooler can reach 98.77%, which realizes the reliable identification of the fault state of the cooler.

Key words: cooler, fault diagnosis, particle swarm optimization, deep belief network, multi-sensor fusion