基于DNN-CapsNet的液压泵故障程度诊断方法

Fault Degree Diagnosis of Hydraulic Pump Based on DNN-CapsNet

液压泵作为液压系统中的主要动力提供者,其内部若发生故障,将对液压系统运行稳定性和可靠性产生威胁。针对其在多故障模式下的故障程度诊断问题,提出一种将深度神经网络(Deep Neural Network,DNN)与胶囊网络(Capsule Network,CapsNet)相结合的液压泵故障程度诊断方法。首先,采用DNN网络替换胶囊网络中的特征提取层来充分挖掘液压泵故障数据中的关键特征;其次,利用胶囊网络数字胶囊层中的动态路由算法更新模型参数;最后,计算输出层输出向量模长实现对液压泵多故障模式下故障程度的准确识别。通过搭建液压泵数字孪生体采集压力故障数据来进行实验。结果表明:相比于传统深度神经网络、胶囊网络,该方法对于液压泵故障程度诊断的准确率达到99.67%。

As the main power provider in a hydraulic system,the internal failure of a hydraulic pump will threaten the operational stability and reliability of the hydraulic system.Aiming at the problem of fault diagnosis under multiple fault modes,a hydraulic pump fault diagnosis method combining Deep Neural Network(DNN) and Capsule Network(CapsNet) is proposed.Firstly,the DNN network is used to replace the feature extraction layer in the Capsule Network to fully explore the key features in the hydraulic pump fault data.Secondly,the dynamic routing algorithm in the digital capsule layer of the Capsule Network is used to update the model parameters.Lastly,the output vector modulus length of the output layer is calculated to achieve the accurate recognition of the hydraulic pump fault degree under multiple fault modes.Experiments are conducted by building a hydraulic pump digital twin to collect pressure fault data.The results show that the accuracy of this method for hydraulic pump fault degree diagnosis reaches 99.17% compared with the traditional deep neural network and capsule network.