Optimizing dimensionless input-output flow sensing system for embedded platform

Accurate perception of flow data in flight is critical for the navigation, guidance, and control of aircraft. Flush Air Data Sensing (FADS) is a new type of non-intrusive flow sensing technology that mainly uses distributed pressure sensors on the surface of an aircraft to indirectly sense flow parameters without deployment mechanisms. For wide-speed range flight air data measurement, this research constructs a multi-physical parameter flow sensing system based on a dimensionless input-output neural network model. The flow sensing system combined with the data-driven model can realize high-fidelity real-time perception of multiple flow parameters such as Mach number, angle of attack, and sideslip angle under wide-speed range conditions. This study adopts a data-driven feature sparsity method to realize the model compression without compromising accuracy. The data-driven feature sparsity method decreases the storage space and computing time of the flow sensing algorithm for a resource-constrained embedded computing platform. The system performance and reliability of the proposed framework were tested on the embedded computing platform. The experimental results show that: the measurement error of the Mach number is less than 0.05; the measurement errors of the angle of attack and sideslip angle are both less than 0.6 degree.