摘要
针对压燃式活塞发动机气门间隙故障振动信号样本少以及跨工况故障诊断困难的问题,提出一种基于元学习和迁移学习的压燃式活塞发动机气门间隙异常故障诊断方法。元学习采用MAML作为学习器,对目标域的支撑集进行数据扩展,提升其泛化能力;迁移学习采用ResNet34作为特征提取网络,并通过SSL替代SL损失函数,压缩源域特征向量之间的距离,为目标域任务提供更多的特征嵌入空间,提升其跨域诊断能力。将预训练和微调后的元学习和迁移学习模型进行决策融合后作为诊断结果输出,并使用发动机台架进行实验数据验证。结果表明,所提方法能在小样本情况下有效识别跨工况气门间隙故障,且效果明显优于单独使用元学习或迁移学习的诊断方法。
In allusion to the challenges of limited vibration signal samples and difficulties in diagnosing faults across different operating conditions in valve clearance fault diagnosis of compression-ignition piston engine,a method of compression-ignition piston engine valve clearance abnormal fault diagnosis based on meta-learning and transfer learning is proposed.In the meta-learning,MAML is used as learner to expand the support set data of the target domain,thereby enhancing its generalization ability.In the transfer learning,ResNet34 is used as the feature extraction network to replace the SL loss function with SSL to compress the distance between feature vectors of the source domain,providing more feature embedding space for the target domain task and enhancing its cross-domain diagnostic capability.The decision fusion of pre-trained and fine-tuned meta-learning and transfer learning models are used as the diagnostic result output,and experimental data validation is conducted by means of engine bench.The results show that the proposed method can effectively identify cross working condition valve clearance faults in small sample situations,and the effect is significantly better than diagnostic methods that use meta-learning or transfer learning alone.
作者
何鹏飞
万洪平
黄国勇
HE Pengfei;WAN Hongping;HUANG Guoyong(Faculty of Civil Aviation and Aeronautics,Kunming University of Science and Technology,Kunming 650500,China;Yunnan Branch of CGN New Energy Holding Co.,Ltd.,Kunming 650200,China)
出处
《现代电子技术》
北大核心
2024年第18期29-34,共6页
Modern Electronics Technique
基金
南通常测机电设备有限公司科技项目(KKF0202165365)。
