FDENN: Health Indicator Extraction for Power Electronics of Electro-Mechanical Actuator Based on Feature-Enhanced Deep Elastic Neural Network

Electro-mechanical actuator (EMA) is increasingly recognized as safety-critical components in mechatronics of aviation applications. The power electronics, as key components of EMA, are essential for maintaining safe and stable operations. Thus, the research of degradation process for power electronics can significantly improve the reliability of EMA. However, the limitations on sensor placement in practical applications may result in inadequate feature extraction, making it difficult to capture the mapping relationship between features and the underlying degradation process. Consequently, this poses challenges in accurately assessing the EMA power electronics degradation characteristics. Therefore, research on the EMA health indicator (HI) extraction to characterize the degradation of EMA power electronics without adding specialized sensors is highly valuable. In this article, an innovative method for extracting HI from the EMA power electronics, based on a feature-enhanced deep elastic neural network (FDENN), is proposed, which integrates the mixed logic dynamic model of EMA phase currents with the deep elastic neural network without additional sensors. Experiments utilizing both simulation and practical data are conducted to assess the effectiveness of the proposed method. The results indicate that the FDENN model exhibits high performance in extracting the HI for EMA power electronics.