Model–Based Techniques for Virtual Sensing of Longitudinal Flight Parameters

Abstract

Introduction of fly-by-wire and increasing levels of automation significantly improve the safety of civil aircraft, and result in advanced capabilities for detecting, protecting and optimizing A/C guidance and control. However, this higher complexity requires the availability of some key flight parameters to be extended. Hence, the monitoring and consolidation of those signals is a significant issue, usually achieved via many functionally redundant sensors to extend the way those parameters are measured. This solution penalizes the overall system performance in terms of weight, maintenance, and so on. Other alternatives rely on signal processing or model-based techniques that make a global use of all or part of the sensor data available, supplemented by a model-based simulation of the flight mechanics. That processing achieves real-time estimates of the critical parameters and yields dissimilar signals. Filtered and consolidated information is delivered in unfaulty conditions by estimating an extended state vector, including wind components, and can replace failed signals in degraded conditions. Accordingly, this paper describes two model-based approaches allowing the longitudinal flight parameters of a civil A/C to be estimated on-line. Results are displayed to evaluate the performances in different simulated and real flight conditions, including realistic external disturbances and modeling errors.

Towards Certification of a Reduced Footprint ACAS-Xu System: A Hybrid ML-Based Solution

“…Learning a Surrogate NN of a LUT Neural networks are a new trend for approximating complex functions as a replacement of LUT, for example for control command systems [27]. Significant work has been performed in the framework of deep Q-learning, where the Q table is approximated by neural network.…”

An Innovative Angle of Attack Virtual Sensor for Physical-Analytical Redundant Measurement System Applicable to Commercial Aircraft

“…[22] The authors of the present paper deem that a suitable customization of the AoA estimator developed for GA aircraft and presented in [1] and [17] could help addressing the issues discussed for Commercial Aviation aircraft. Indeed, it could be exploited as an additional virtual sensor that, coupled with existing physical sensors and suitable voting algorithms [25]- [27], allows to implement a physical-analytical redundant measurement system that is robust to a single fault of the physical sensors. This measurement system could significantly improve the safety of the aircraft, reducing the occurrence of LOC-I related accidents, saving human lives, and avoiding the economic impact on the airliners and the aircraft manufacturers consequent to those accidents.…”

Angle of attack prediction using recurrent neural networks in flight conditions with faulty sensors in the case of F-16 fighter jet

“…In [5], Extended Kalman Filter (EKF) is used to estimate the AOA using GPS, IMU, and Pitot tube measurements. In [6], Adaptive Kalman Filter is used to estimate the AOA. In [7], GPS receiver, an inertial reference system and a sensor measurement for the static and dynamic pressure are used to estimate the AOA.…”

“…Furthermore, most literature uses linear models like the Kalman Filter for AOA estimation [5,6,9]. Nonlinearities have a significant impact on dynamic responsiveness in modern high-performance aircraft.…”

“…2. The majority of AOA estimation literature [5,6,9] uses an extended Kalman filter which works for some nonlinear dynamical systems that can be represented by linear operations but are not well suited when the system is complex and nonlinear. We have used RNN, which can model complex nonlinear dynamical systems [25].…”