针对液压系统故障数据采集困难和数据量有限导致其故障诊断模型精度低的问题,提出基于子领域自适应的故障诊断方法.通过液压系统仿真模型模拟不同的故障类型,获得丰富的故障仿真数据,并与有限的实测故障数据结合用于模型构建;将1维卷积神经网络(One-dimensional-Convolutional Neural Network,1D-CNN)与子领域自适应方法结合,利用1D-CNN构建多传感器信息融合网络全面提取液压系统的故障信息,在域自适应层采用局部最大平均差异(Local Maximum Mean Discrepancy,LMMD)减小源域和目标域特征之间的全局分布差异和子领域分布差异,构建出故障诊断模型.结果表明,所提方法在各类故障实测样本仅有1个的情况下,故障的诊断准确率达到100％,优于所对比的其他智能故障诊断方法.
Fault Diagnosis Method of Hydraulic System Based on Subdomain Adaptation
Addressing the challenge of difficult data collection and limited datasets for hydraulic system faults,leading to lower accuracy in fault diagnosis models,a fault diagnosis method based on subdomain adaptation is proposed.Using a hydraulic system simulation model,various fault types to generate rich simulation data.Combining this simulated data with limited real-world fault data for model construction,we integrate a 1D-CNN with subdomain adaptation.This approach,employing a 1 D-CNN network and a multi-sensor information fusion strategy,comprehensively extracts fault information from different domains of hydraulic systems.In the domain adaptation layer,LMMD is utilized to reduce global and subdomain distribution differences between source and target domains.The research results demonstrate that our proposed method achieves a 100％diagnostic accuracy for each fault type with only one real-world sample per fault type,outperforming compared other intelligent fault diagnosis methods.
国家科技期刊平台
NSTL
万方数据
