Model based IVHM system for the solid rocket booster

We report progress in the development of a model-based hybrid probabilistic approach to an on-board IVHM for solid rocket boosters (SRBs) that can accommodate the abrupt changes of the model parameters in various nonlinear dynamical off-nominal regimes. The work is related to the ORION mission program. Specifically, a case breach fault for SRBs is considered that takes into account burning a hole through the rocket case, as well as ablation of the nozzle throat under the action of hot gas flow. A high-fidelity model (HFM) of the fault is developed in FLUENT in cylindrical symmetry. The results of the FLUENT simulations are shown to be in good agreement with quasi-stationary approximation and analytical solution of a system of one-dimensional partial differential equations (PDEs) for the gas flow in the combustion chamber and in the hole through the rocket case.

💡 Research Summary

The paper presents a comprehensive model‑based framework for an on‑board Integrated Vehicle Health Management (IVHM) system tailored to solid rocket boosters (SRBs) used in the Orion mission. Recognizing that SRB internal flow dynamics are highly nonlinear and that fault conditions can cause abrupt changes in model parameters, the authors focus on the case‑breach fault—a scenario where a hole forms in the rocket case, allowing hot gas to leak, while simultaneously the nozzle throat undergoes ablation and the metal case melts.

A high‑fidelity model (HFM) of this fault is built in ANSYS FLUENT using axisymmetric two‑dimensional CFD. The HFM captures gas leakage through the breach, nozzle wall recession, metal melting, and insulator ablation. Simulation results are compared with a quasi‑steady analytical solution derived from a set of one‑dimensional partial differential equations (PDEs) describing gas flow in the combustion chamber and in the breach. The CFD and analytical results agree closely, validating the reduction of the three‑dimensional physics to a 1‑D representation for the purpose of low‑dimensional modeling.

From the HFM, a low‑dimensional performance model (LDPM) is derived by integrating the 1‑D PDEs along the axial direction of the chamber, nozzle, and breach. The LDPM consists of six coupled ordinary differential equations (ODEs) that describe the evolution of chamber pressure, gas density, temperature, breach radius, nozzle throat radius, and propellant burn surface. Key features include: (1) a prescribed burn‑area curve S_b = f_R(R) linking propellant surface area to web‑burn distance, (2) explicit equations for metal melting at the breach wall, (3) an effective nozzle area term that incorporates the changing geometry due to breach and ablation, (4) a volume‑change term for the combustion chamber, and (5) optional inclusion of insulator‑layer dynamics. This LDPM can reproduce the time‑varying geometry of the SRB while remaining computationally inexpensive enough for real‑time use.

Two real‑time inference algorithms are built on top of the LDPM. The first, a self‑consistent iterative algorithm, solves the LDPM under a quasi‑adiabatic assumption, updating pressure, density, and temperature without explicit numerical integration. This method is extremely fast and suitable when the fault evolves smoothly. The second algorithm employs a dynamic inference approach: the LDPM is cast as a system of stochastic differential equations, and Bayesian filtering (a Kalman‑filter‑type estimator) is applied to estimate the hidden state vector and model parameters from measured stagnation pressure and thrust. This probabilistic method explicitly accounts for measurement noise and model uncertainty, providing confidence intervals for the estimates and handling abrupt parameter jumps.

The authors validate both algorithms using two representative fault scenarios derived from recent ground‑test data. In Scenario 1, a sudden large hole appears in the metal case while the insulator hole remains smaller; the breach dynamics are dominated by nozzle ablation and insulator erosion. The self‑consistent algorithm quickly converges and accurately identifies the fault onset within a few seconds. In Scenario 2, a slower crack propagates through the insulator, leading to gradual enlargement of the breach; metal melting becomes the dominant mechanism, and the dynamics deviate from the classical Bartz approximations. Here, the dynamic inference algorithm captures the non‑linear growth curve and provides reliable prognostics of thrust loss. In both cases, the algorithms diagnose the fault and predict future thrust trends using only pressure and thrust time‑series, meeting the stringent SRB IVHM requirements of sub‑second decision windows and limited sensor suites.

The paper concludes that the combination of (i) high‑fidelity CFD validation, (ii) a physics‑based low‑dimensional model that retains essential geometric and thermodynamic effects, and (iii) Bayesian‑inspired real‑time inference yields a robust on‑board FD&P (fault detection and prognosis) capability for SRBs. The authors suggest future work to extend the framework to multiple fault modes (combustion instability, nozzle choking, etc.), incorporate additional sensor modalities (temperature, vibration), and develop lightweight hardware implementations suitable for flight. This research represents a significant step toward safer, more reliable solid‑propellant launch vehicles.