基于IndRNN-1DLCNN的负载口独立控制阀控缸系统故障诊断

IndRNN-1DLCNN based fault diagnosis of independent metering valve-controlled hydraulic cylinder system

为了解决负载口独立控制阀控液压缸系统故障信息相似表征下的故障元件识别难题,提出基于独立循环神经网络(IndRNN)和一维大核卷积神经网络(1DLCNN)结合的故障诊断方法.构建负载口独立控制阀控液压缸系统,针对系统提出压力与位移信号的状态感知方案,分析了系统故障的信号特征.设计一种基于IndRNN-1DLCNN的深度神经网络模型,模型引入残差结构进行多层IndRNN设计并引入1DLCNN增强全局信息捕捉能力,实现多源信号的融合,识别发生故障的具体元件.结果表明在不同的负载工况下,利用提出的方法均能够准确地将系统故障定位至4个先导阀、2个主阀、1组位移传感器以及1个液压缸共8类具体元件,系统的整体诊断准确率最高达到96%,单一元件的故障识别准确率均大于93%.

A fault diagnosis method was proposed to address the problem of fault components identification under similar fault information representations in an independent metering valve-controlled hydraulic cylinder system.An independently recurrent neural network(IndRNN)and a one-dimensional large-kernel convolution neural network(1DLCNN)was combined in the method.An independent metering valve-controlled hydraulic cylinder system was constructed.A state-sensing scheme was presented for capturing pressure and displacement signals.The signal characteristics under different fault conditions were analyzed.A deep neural network model utilizing IndRNN-1DLCNN was established.The deep network architecture of multi-layer IndRNN with a residual structure was adopted.The 1DLCNN was developed to enhance the global information capture capability.The model structure facilitated multi-sensor information fusion and specific fault component identification.Results showed that the proposed method could accurately distinguish eight specific fault components,including four pilot valves,two main valves,displacement sensors and a hydraulic cylinder in the case of different working conditions.The overall diagnostic accuracy of the system could reach up to 96%for the discussed working conditions.The fault identification accuracy of one component was above 93%under the working condition.