Real-time aerodynamic load estimation for hypersonics via strain-based inverse maps

Julie Pham, Omar Ghattas, Noel Clemens, Karen WillcoxarXiv 2024 · DOI · arXiv

TL;DR

Reconstructing aerodynamic loads on hypersonic vehicles in real time is notoriously difficult due to extreme aerothermal environments that destroy external surface sensors. This paper introduces a "vehicle-as-a-sensor" approach that pre-computes an explicit inverse map from internal structural strain measurements to external surface pressures. By leveraging the linearity of structural elasticity, the method delivers accurate, physics-informed load estimates and uncertainty quantification in milliseconds.

What Problem Does This Paper Address?

Hypersonic flight control systems need real-time aerodynamic data—specifically surface pressure fields—to ensure safe guidance, navigation, and control (GNC). However, directly measuring surface pressure at Mach 5+ is nearly impossible because the extreme heat and harsh aerothermal environment quickly degrade external sensors. Traditional alternatives like GPS or Inertial Measurement Units (IMUs) don't provide direct aerodynamic load data and can be denied or jammed in adversarial settings.

The authors tackle this by treating the vehicle itself as a sensor. When aerodynamic forces hit the vehicle, the airframe deforms. By measuring this internal deformation (strain) using robust internal fiber-optic sensors, they aim to back-calculate the external pressure field. The core challenge is that solving this partial differential equation (PDE)-constrained inverse problem in real time is typically computationally intractable.

Methods

The authors exploit a key physical insight: while hypersonic fluid dynamics are highly nonlinear, the structural response of the vehicle's airframe can be accurately modeled using linear elasticity.

Because the structural forward model is linear ($A\mathbf{u} - C\mathbf{q} = 0$), the inverse problem can be formulated as a statistical least-squares problem with linear constraints. This allows the authors to derive an explicit analytical solution via the normal equations.

The resulting estimator is a linear inverse map: $$\hat{\mathbf{q}} = T\mathbf{d} + \mathbf{k}$$ where $\mathbf{d}$ is the sparse strain data, $T$ is a pre-computed matrix, and $\mathbf{k}$ is a shifting term based on prior data. In real-time flight, evaluating this requires only a simple matrix-vector multiplication, which executes an order of magnitude faster than the typical 100 Hz frequency of onboard sensors.

The paper explores two distinct formulations for the pressure quantities of interest ($\mathbf{q}$):

  1. Over-determined ($n_q \le n_d$): The pressure field is projected onto a low-dimensional subspace using Proper Orthogonal Decomposition (POD) derived from offline Computational Fluid Dynamics (CFD) snapshots.
  2. Under-determined ($n_q > n_d$): Direct full-field estimation using a data-driven Gaussian prior covariance constructed from CFD snapshots, solved efficiently using the Woodbury matrix identity to handle rank-deficient matrices.
Figure 1: Inverse problem overview.

Datasets

The method relies on offline-generated simulation databases rather than standard public datasets:

  • Aerodynamic Data: Generated using CART3D, an inviscid steady fluid dynamics solver for compressible Euler equations. The authors collected 90 snapshots for the POD case and 243 snapshots for the full-field prior covariance case across varying Mach numbers (5–7), angles of attack, and sideslip angles at a 20 km altitude.
  • Structural Data: Built using FEniCS to discretize the linear elasticity equations over a finite element mesh of the Initial Concept 3.X (IC3X) conceptual hypersonic vehicle fuselage (60,538 surface degrees of freedom, 461,664 structural degrees of freedom).

Key Insights and Ideas

  • Linearity is a Superpower: Even though hypersonic aerodynamics are wildly nonlinear and complex, the structural response remains linear elastic. Shifting the inverse problem's constraint to the structural domain bypasses the need for expensive, iterative nonlinear PDE solvers during flight.
  • The Smoothing Effect of Physics: The forward physical process (pressure $\rightarrow$ strain) acts as a low-pass filter, smoothing out sharp pressure discontinuities like shocks. Consequently, the inverse problem is ill-posed; high-frequency spatial features are naturally attenuated and hard to recover from sparse data.
  • Integrated Quantities are Forgiving: Even when sensor noise and regularization prevent the exact reconstruction of sharp localized pressure shocks, the integrated aerodynamic loads (total force and moment coefficients) remain incredibly accurate because the global bending moments are well-captured.

Key Results

  • Speed: Because the real-time step is reduced to a single matrix-vector product, load estimation takes less than 10 milliseconds, easily satisfying real-time GNC requirements.
  • Reconstruction Accuracy (Low-Dimensional POD): Using 54 internal strain sensors, the median relative $L_2$ error for surface pressure reconstruction was just 5.5% across thousands of noisy synthetic test cases.
  • Full-Field Accuracy: In a noise-free setting, full-field reconstruction achieved a 2.8% relative error. With 1% sensor noise added, the full-field error increased to 13.3%–27.9% because the regularization smooths out the leeward shock wave.
  • Force & Moment Coefficients: Despite full-field smoothing errors, the integrated normal and side force coefficients maintained relative errors below 2%, and pitch/yaw moments stayed below 3.4%.

Limitations and Future Work

  • Inviscid Aerodynamics: The prior database used CART3D (Euler equations), meaning it ignores viscous drag due to skin friction. The authors note that future work must account for the viscous drag component.
  • Thermal Effects Omitted: Hypersonic flight generates extreme aerothermal heating, causing thermal expansion and temperature-dependent material property shifts. These introduce structural nonlinearities not captured by the current linear elastic model.
  • Sensor Placement: The study used heuristic sensor rows. The authors plan to investigate algorithmic optimal sensor placement to maximize information gain and minimize conditioning errors.

Summary updated

Abstract

This work develops an efficient real-time inverse formulation for inferring the aerodynamic surface pressures on a hypersonic vehicle from sparse measurements of the structural strain. The approach aims to provide real-time estimates of the aerodynamic loads acting on the vehicle for ground and flight testing, as well as guidance, navigation, and control applications. Specifically, the approach targets hypersonic flight conditions where direct measurement of the surface pressures is challenging due to the harsh aerothermal environment. For problems employing a linear elastic structural model, we show that the inference problem can be posed as a least-squares problem with a linear constraint arising from a finite element discretization of the governing elasticity partial differential equation. Due to the linearity of the problem, an explicit solution is given by the normal equations. Pre-computation of the resulting inverse map enables rapid evaluation of the surface pressure and corresponding integrated quantities, such as the force and moment coefficients. The inverse approach additionally allows for uncertainty quantification, providing insights for theoretical recoverability and robustness to sensor noise. Numerical studies demonstrate the estimator performance for reconstructing the surface pressure field, as well as the force and moment coefficients, for the Initial Concept 3.X (IC3X) conceptual hypersonic vehicle.
BibTeX
@misc{pham2024real,
  author = {Julie Pham and Omar Ghattas and Noel Clemens and Karen Willcox},
  title = {{Real-time aerodynamic load estimation for hypersonics via strain-based inverse maps}},
  year = {2024},
  eprint = {2408.15286},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2408.15286},
}
Figure