Hydraulic systems, known for their inherent complexity and nonlinearity, are integral to a vast array of modern industrial applications. The precise diagnosis of faults within hydraulic systems is very important to ensuring the safety and reliability of industrial operations. The fault diagnosis of the hydraulic system is still challenging because of its strong nonlinearity and high secrecy. This paper constructs a residual neural network (ResNet) to recognize the fault types of different hydraulic components. ResNet alleviates the gradient dis-appearance problem in deep learning by introducing a shortcut mapping structure, which enables the network to better fit the correlation between system operating parameters and hydraulic component failure types. Experimental results show that the proposed method based on ResNet has higher fault diagnosis accuracy than other deep neural networks.