Real-Time System Identification of a Small Multi-Engine Aircraft

Wesley DeBusk, Eric N. Johnson, Girish ChowdharyAIAA 2009 · DOI

TL;DR

This paper demonstrates the feasibility of using a recursive Fourier Transform Regression (FTR) method for real-time system identification of a small, twin-engine Unmanned Aerial Vehicle (UAV). By applying this frequency-domain technique to flight test data, the researchers successfully estimated the aircraft's aerodynamic stability derivatives, paving the way for adaptive control and fault detection systems.

What Problem Does This Paper Address?

Accurately modeling an aircraft's dynamic behavior is crucial for advanced control systems, especially for adaptive control schemes that need to adjust to changing flight conditions or unexpected failures. Traditional system identification methods often require offline processing, making them unsuitable for real-time applications. The challenge is to develop a method that can continuously update the aircraft's dynamic model during flight, providing essential information for onboard controllers and health monitoring. This paper specifically tackles this by aiming to identify the aerodynamic stability derivatives of a small, twin-engine UAV in real-time.

Methods

The core of this paper lies in the application of the Recursive Fourier Transform Regression (FTR) method for system identification. This technique operates in the frequency domain, which offers advantages in handling noisy data and can inherently remove biases and trim conditions by omitting the zero frequency.

Here's a breakdown of the key methodological aspects:

  • State-Space Model: The aircraft dynamics are represented by a linearized, six-degree-of-freedom state-space model. This model is then decoupled into longitudinal and lateral-directional subsystems.
  • Fourier Transform Regression (FTR): This method estimates unknown parameters (aerodynamic derivatives and actuator effectiveness) by performing a least-squares fit in the frequency domain. It leverages the discrete Fourier transform of the system's signals.
  • Recursive Implementation: The FTR is implemented recursively, meaning it can update parameter estimates as new data arrives, making it suitable for real-time applications. This is achieved by propagating the Fourier transform in a recursive manner.
  • Data Filtering: A fourth-order Butterworth filter is used to reduce noise and unwanted frequency content from the flight data before system identification. The Butterworth filter was chosen for its flat pass-band response.
  • System Identification Maneuver: To gather rich data for identification, a "sum of sines" input was used for the elevator, aileron, and rudder control surfaces. The phase angles of these inputs were optimized for fast convergence and to avoid high amplitude peaks.

Datasets

The primary dataset used in this paper is flight test data collected from a Multiplex Twinstar, a small, twin-engine, fixed-wing UAV. This data was gathered during a system identification maneuver where optimized "sum of sines" inputs were applied to the control surfaces while the aircraft was under closed-loop control. The paper also mentions validation using computer simulation data, but the focus is on the real flight data.

Key Insights and Ideas

  • Frequency Domain for Real-Time: The paper strongly advocates for the frequency domain approach (FTR) for real-time system identification, highlighting its ability to inherently handle biases and noise by selectively ignoring certain frequencies.
  • Recursive FTR for Onboard Application: The recursive nature of the FTR method is key to its real-time applicability. This allows for continuous model updates, which is essential for adaptive control and fault detection.
  • Linear Model Sufficiency: The study validates the assumption that a simple linear model is sufficient for capturing the essential behavior of the GT Twinstar, at least for the maneuvers tested. This simplifies the identification problem.
  • Optimized Inputs for Rich Data: The use of optimized "sum of sines" inputs proved effective in generating rich excitation across the control surfaces, leading to faster convergence of parameter estimates.
  • Closed-Loop Identification: Performing system identification in a closed-loop condition (as done with the UAV) helps maintain flight stability and allows for richer data collection, even with external disturbances.

Key Results

  • Successful Parameter Convergence: The FTR method successfully estimated both longitudinal and lateral-directional aerodynamic stability derivatives. The parameters began to converge within approximately 2 seconds of the start of the system identification maneuver.
  • Good Model Match: The estimated model showed a good match with the measured data in both time and frequency domains. Figures 8 and 9 demonstrate satisfactory time-domain comparisons, while Figures 15 and 16 show good agreement in the frequency content.
  • Validation with Unused Data: Parameter estimates were validated using flight data not used for system identification, and the match was found to be satisfactory (Figures 17 and 18).
  • Comparison to Batch Least Squares: The final estimated values from the FTR method fell within a tolerable region compared to values obtained from batch-processed data using least squares estimation (Table II).
  • Computational Feasibility: The paper concludes that the method is computationally feasible for real-time implementation on the onboard autopilot system (AFI FCS 20).

Limitations and Future Work

  • Linear Model Assumption: The paper acknowledges that while a linear model was sufficient for the tested maneuvers, heavy damage or extremely aggressive maneuvers could lead to the buildup of nonlinear effects.
  • Full State Knowledge: The FTR method, as presented, assumes accurate full state knowledge.
  • Forgetting Factor: For situations where aircraft dynamics change significantly over time, a forgetting factor would be necessary to discount old data.
  • Future Work: The paper implies that the successful demonstration of real-time FTR opens doors for its integration into adaptive control and fault detection systems for UAVs. Further research could explore the impact of nonlinearities and the implementation of forgetting factors for more dynamic scenarios.

Summary updated

Abstract

In-flight identification of an aircraft’s dynamic model can benefit adaptive control schemes by providing estimates of aerodynamic stability derivatives in real time. This information is useful when the dynamic model changes severely in flight such as when faults and failures occur. Moreover a continuously updating model of the aircraft dynamics can be used to monitor the performance of onboard controllers. Flight test data was collected using a sum of sines input implemented in closed loop on a twin engine, fixed wing, Unmanned Aerial Vehicle. This data has been used to estimate a complete six degree of freedom aircraft linear model using the recursive Fourier Transform Regression method in frequency domain. The methods presented in this paper have been successfully validated using computer simulation and real flight data. This paper shows the feasibility of using the frequency domain Fourier Transform Regression method for real time parameter identification.
BibTeX
@inproceedings{debusk2009real,
  author = {Wesley DeBusk and Eric N. Johnson and Girish Chowdhary},
  title = {{Real-Time System Identification of a Small Multi-Engine Aircraft}},
  booktitle = {AIAA},
  year = {2009},
  doi = {10.2514/6.2009-5935},
}
Figure