TL;DR
This paper introduces a novel Linear Parameter-Varying (LPV) model reduction technique to create consistent, lower-order models of flexible aircraft across their flight envelope. This approach successfully reduces a complex 148-state aeroservoelastic model of a Body Freedom Flutter (BFF) vehicle to a more manageable 26 states, crucial for designing effective flight control systems.
What Problem Does This Paper Address?
Modern aircraft, especially those with high-aspect-ratio wings, exhibit increased flexibility. This flexibility leads to complex interactions between rigid body dynamics and structural modes, a phenomenon known as Body Freedom Flutter (BFF). While detailed models of these aeroservoelastic (ASE) systems can accurately capture this behavior, they result in very high-order state-space representations (e.g., 148 states). These high-order models are computationally prohibitive and practically difficult to use for designing controllers for flutter suppression and improving handling qualities. Existing model reduction techniques often produce different reduced-order models at different flight conditions, making controller scheduling complex. The core problem is to develop a method that reduces model order while maintaining a consistent set of states across the entire flight envelope, thus simplifying controller design and implementation.
Methods
The paper proposes an LPV model reduction method based on parameter-varying balanced realization techniques. The core idea is to achieve a consistent set of states across the flight envelope, unlike traditional methods that might eliminate different states at each operating point.
The approach involves:
- Initial Reduction: Standard truncation and residualization techniques are applied to reduce the initial 148-state model to 43 states. This step aims to remove states not critical to the flutter dynamics.
- LPV Balanced Realization: The key innovation is the application of a parameter-varying balanced realization. This method uses generalized controllability and observability Gramians to find a balancing state transformation that is invariant across the flight envelope. This ensures that the same states are retained regardless of the operating condition.
- Coprime Factorization and Singular Perturbation: The LPV model reduction is framed within coprime factorization and singular perturbation theories, extending existing LTI techniques to the LPV domain. This involves solving Linear Matrix Inequalities (LMIs) to compute the generalized Gramians.
The paper also compares this LPV approach with classical LTI model reduction techniques, including Therapos' and Zhou's methods for unstable systems, and point-wise balanced reduction.
Datasets
The paper uses an experimental Body Freedom Flutter (BFF) vehicle model developed by the U.S. Air Force Research Laboratory (AFRL). This model is derived from MSC/NASTRAN simulations and represents a high-aspect-ratio, lightweight flying wing aircraft. The model is characterized by its complex aeroservoelastic dynamics, including rigid body modes, structural bending and torsion modes, and unsteady aerodynamics. The data used for evaluation spans a range of airspeeds from 40 to 90 KEAS.
Key Insights and Ideas
- Consistent State Representation is Crucial: The primary insight is that for practical controller design and scheduling, a reduced-order model must retain a consistent set of states across the flight envelope. LPV model reduction, by using an invariant state transformation, achieves this, unlike LTI methods applied point-wise.
- LPV Balanced Realization for Stability: The proposed LPV balanced realization method effectively handles mixed-stability systems (common in aeroservoelasticity) by considering both stable and unstable modes within a unified framework, which is a significant advantage over traditional methods requiring stable systems.
- Trade-off Between Order and Consistency: While LTI methods can achieve lower state orders (e.g., 20 states), they result in different states being eliminated at each flight condition. The LPV method achieves a higher state order (26 states) but provides a consistent representation, which is more valuable for controller design.
- LPV Dynamics Drive Complexity: The significant changes in aircraft dynamics across the flight envelope (as seen in Figure 3) make achieving very low-order, consistent LPV models challenging, leading to higher state orders than might be achieved with LTI methods at specific points.
Key Results
- The initial 148-state aeroservoelastic model of the BFF vehicle was successfully reduced to 43 states using standard techniques.
- The proposed LPV model reduction method yielded consistent, control-oriented models with 26 states across the flight envelope.
- These 26-state LPV models capture the essential dynamics for active flutter suppression controllers.
- Frequency response comparisons (Figures 11 and 12) show that the 26-state LPV models provide good accuracy compared to the 43-state modal residualization model, with errors generally less than 10%.
- LPV models achieved better accuracy and consistency across the flight envelope compared to LTI methods applied point-wise, although LTI methods could sometimes achieve lower state orders (e.g., 20 states) at the cost of state consistency.
Limitations and Future Work
- Numerical Issues with LMIs: The paper highlights significant numerical challenges in solving the LPV model reduction problem due to the large number of decision variables and constraints in the Linear Matrix Inequalities (LMIs). This feasibility problem becomes more pronounced with higher-order models and more flight conditions.
- Marginally Stable Systems: The accuracy of the reduced models can degrade at flight conditions where the system is marginally stable, leading to larger errors.
- Future Work: The authors suggest that future work should focus on addressing LMI feasibility issues for aeroservoelastic systems. Additionally, exploring methods to further reduce the state order while maintaining consistency and accuracy would be beneficial. Investigating the impact of parameter variation rates on the reduction process is also a potential area for future research.
Summary updated