摘要
液压轴向柱塞泵是液压系统的核心动力元件,对轴向柱塞泵进行故障诊断对于保证液压装备系统的安全可靠性运行至关重要。提出了一种改进的辅助分类生成对抗网络与模型迁移策略相结合的故障诊断方法,构建了故障诊断框架,并采用预训练-微调策略提高了模型在目标域任务中的泛化能力,解决了传统深度学习诊断方法在实际运行过程中正常数据与故障数据数量因数据不平衡导致效果不佳甚至失效的问题。试验证明,该方法在样本不均衡时,其结构相似性值提高了20.4%,峰值信噪比值提高了5.4%,三种数据集在F1分数评估指标上分别可以达到96.3%、94.4%、92.5%,能够有效提高生产样本的质量和轴向柱塞泵的故障识别率。
Hydraulic axial piston pumps are core power components in hydraulic systems,so effectively diagnosing faults in axial piston pumps is crucial for ensuring the safe and reliable operation of hydraulic equipment.This paper proposes an improved fault diagnosis method that combines an auxiliary classification generative adversarial network and a model migration strategy.A fault diagnosis framework and adopts a pre-training-fine-tuning strategy to improve the model's generalisation ability in the target domain task is proposed.This method solves the problem of traditional deep learning diagnostic methods having a poor effect,or even failing,in the actual operation process of normal and fault data due to data imbalance and insufficient quantity.Experimental results show that this method increases the structural similarity value by 20.4%and the peak signal noise ratio value by 5.4%when samples are imbalanced.The three datasets achieve F1 scores of 96.3%,94.4%and 92.5%,respectively,effectively improving the quality of production samples and the fault recognition rate of axial piston pumps.
作者
李兴东
向星
马诗浩
郭雨萱
潘宏鑫
宋明星
LI Xingdong;XIANG Xing;MA Shihao;GUO Yuxuan;PAN Hongxin;SONG Mingxing(School of Mechanical Engineering,Yanshan University,Qinhuangdao,Hebei 066004;School of Mechanical Engineering,Hebei University of Architecture,Zhangjiakou,Hebei 075000)
出处
《液压与气动》
北大核心
2025年第8期21-34,共14页
Chinese Hydraulics & Pneumatics
基金
国家自然科学基金(52275066)
河北省科技重大专项(23281901Z)
新疆生产建设兵团第十二师科技局2022年重大科技专项(SRS2022003)。
关键词
数据不平衡
生成对抗网络
残差网络
轴向柱塞泵
故障诊断
data imbalance
generative adversarial network
residual network
axial piston pump
fault diagnosis
