MAACCN: An Intelligent Decoupling Diagnosis Method for Compound Faults in Electrohydrostatic Actuators

• Published in: IEEE transactions on instrumentation and measurement Vol. 74; pp. 1 - 11
• Main Authors: Hu, Yuanhao, Song, Yibo, He, Xiansong, Zhao, Xiaoli, Yang, Xiaowei, Yao, Jianyong, Wang, Zhaoqiang, Pei, Hong, Hu, Changhua
• Format: Journal Article
• Language: English
• Published: New York IEEE 2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Actuators; Algorithms; Attention convolutional capsule network; Complex compounds; Complexity; Compounds; Convolution; Convolutional neural networks; Data collection; Decoupling; decoupling classification; electrohydrostatic actuator (EHA); Fault detection; Fault diagnosis; Faults; Feature extraction; Heuristic algorithms; Hydraulic systems; Information systems; maximized aggregation; Routing; Vectors; Working conditions
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2025.3563047

Electrohydrostatic actuators (EHAs), as complex integrated systems of mechanical, electrical, and hydraulic components, play a crucial role in aerospace and other fields. However, due to the complexity of their internal structure and harsh working environments, the compound faults are the most common fault types. Considering the high cost of data collection for uncertainties and complex compound faults in EHA, an intelligent decoupling diagnosis method for compound faults in EHAs based on maximized aggregation attention convolutional capsule network (MAACCN) can be proposed. First, the multidimensional sensor information from the EHA can be collected, the feature-level data are fused through a 1-D convolutional layer combined with an efficient channel attention (ECA) mechanism. Second, it utilizes the capsule network to extract features deeply and introduces a maximized aggregation routing algorithm between capsule layers. Finally, a decoupling classification layer can be added, and its model is optimized by minimizing the margin loss function, enabling the network to more accurately identify and decouple compound faults. Validation on the EHA fault dataset demonstrates that the proposed method achieves higher subset accuracy under different working conditions, which can diagnose compound faults by learning data from single faults.