摘要
防火切断阀作为飞机液压系统中的关键组件,一旦发生堵塞或阀芯偏移等故障,会使主回路供油不足、压力降低,导致液压负载元件故障,造成严重后果。因此,提出了一种基于多层融合-卷积神经网络(CNN)-Transformer的模型,用于防火切断阀的故障诊断。首先,由于切断阀出口处缺乏压力测点,无法利用压差信号进行故障诊断,需采集三轴加速度信号并对其进行特征层预处理;然后,将处理好的数据输入至CNN-Transformer网络进行了训练与分类,CNN的小卷积层能够有效提取局部特征,Transformer则能够对全局特征进行捕捉;最后,针对发动机泵和增压泵等其他元件的振动干扰,利用Dempster-Shafer(DS)证据理论对位于切断阀入口处和出口处的两个加速度传感器的训练结果进行了决策层融合,以提高最终诊断的准确性和可靠性;在搭建的飞机液压系统试验台上对基于多层融合-CNN-Transformer的防火切断阀的故障诊断方法进行了实验验证,并与现有主流方法进行了对比。研究结果表明:基于多层融合-CNN-Transformer的防火切断阀的故障诊断方法在防火切断阀故障诊断中表现出最高的准确率,实验数据在阀芯开口为70%、80%、90%和100%的工况下的平均识别准确率达到99.5%。该方法可为飞行器液压系统中关键元件的智能诊断提供一种高可靠性的技术路线。
As a critical component of the aircraft hydraulic system,the fire shut-off valve can cause insufficient oil supply in the main loop,pressure drop,and failure of hydraulic actuators when faults such as blockage or spool deviation occur,potentially leading to severe consequences.To address this issue,a multi-level fusion-convolutional neural network(CNN)-Transformer model was proposed for fault diagnosis of the fire shut-off valve.First,due to the lack of pressure measurement points at the outlet of the shut-off valve,differential pressure signals could not be utilized for diagnosis.In this case,triaxial acceleration signals were collected and subjected to feature-level preprocessing.Then,the processed data were fed into a CNN-Transformer network for training and classification.The local features were extracted effectively by the small convolution kernels in CNN,while the global dependencies were captured by the Transformer.Finally,to mitigate vibration interference from other components such as the engine pump and booster pump,Dempster-Shafer(DS)evidence theory was employed to perform decision-level fusion of the diagnostic results obtained from two acceleration sensors placed at the inlet and outlet of the shut-off valve,thereby improving the accuracy and reliability of the final diagnosis.The multi-level fusion-CNN-Transformer model was experimentally validated on a constructed aircraft hydraulic system test bench and compared with existing mainstream approaches.The research results show that the proposed method achieves the highest diagnostic accuracy for fire shut-off valve faults,with an average recognition accuracy of 99.5%under spool opening conditions of 70%,80%,90%,and 100%.This method provides a highly reliable technical solution for intelligent diagnosis of key components in aircraft hydraulic systems.
作者
何阳
熊晓燕
王伟杰
李翔宇
兰媛
HE Yang;XIONG Xiaoyan;WANG Weijie;LI Xiangyu;LAN Yuan(School of Mechanical Engineering,Taiyuan University of Technology,Taiyuan 030024,China)
出处
《机电工程》
北大核心
2026年第2期269-279,共11页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金资助项目(52205065)
山西省基础研究计划联合资助项目(202403011212005)。
