Robust Fault-Tolerant Flush Air Data Sensing Algorithm via Incorporating Physical Knowledge

Abstract

The flush air data sensing (FADS) system resolves air data state issues through redundant measurements of surface pressure distributions on the vehicle, with its fault-tolerant algorithm being crucial for ensuring flight safety. However, voting-based fusion strategies for redundant measurements may lead to incorrect judgments under specific conditions, along with limitations such as high algorithmic complexity and underutilization of pressure signals. To address these challenges, this manuscript introduces a fault-tolerant FADS algorithm based on dimensionless input and output convolutional neural networks (FT-DIONNFADS). We trained the neural networks with a fault dataset designed for adaptability, enabling it to work with various pressure port layouts. For each layout, the algorithm incorporates physical knowledge to assess the discrepancy between predicted and true air data states. This approach, based on the principle of minimal error, facilitates the selection of an optimal layout that improves fault diagnosis and tolerance. This algorithm undergoes assessment employing a simplified supersonic model, demonstrating its capability for accurate fault diagnosis and air data estimation across different bias levels. The manuscript also discusses the impact of varying bias levels on FT-DIONNFADS performance.