An In-situ Solid Fuel Ramjet Thrust Monitoring and Regulation Framework Using Neural Networks and Adaptive Control

Ryan DeBoskey, Parham Oveissi, Venkat Narayanaswamy, Ankit Goel · 2025

Read An In-situ Solid Fuel Ramjet Thrust Monitoring and Regulation Framework Using Neural Networks and Adaptive Control by Ryan DeBoskey, Parham Oveissi, Venkat Narayanaswamy, Ankit Goel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-08157.txt.

Introduction

Ryan DeBoskey and co-authors propose a way to watch and control thrust in solid fuel ramjets, in a 2025 paper. These engines burn solid fuel in fast air, but their inner burning is hard to tame, which limits wide use. Direct thrust sensing in flight is also tough.

A neural network learns to infer thrust from a few in-place sensor readings, trained on data from a simplified engine model. A self-tuning controller then adjusts the inlet to hold thrust at target, learning online from past costs. Tuning studies set the best parameters for both parts. In simulations, the paired system held thrust well using sparse sensor data. The authors see a path to safer, wider use of this engine type.

Catalog record

CREATORRyan DeBoskey, Parham Oveissi, Venkat Narayanaswamy, Ankit Goel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-08157

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

An In-situ Solid Fuel Ramjet Thrust Monitoring and Regulation Framework Using Neural Networks and Adaptive Control Ryan DeBoskey, Parham Oveissi, Venkat Narayanaswamy, Ankit Goel Abstract Controlling the complex combustion dynamics within solid fuel ramjets (SFRJs) remains a critical challenge limiting deployment at scale. This paper proposes the use of a neural network model to process in-situ measurements for monitoring and regulating SFRJ thrust with a learning-based adaptive controller. A neural network is trained to estimate thrust from synthetic data generated by a feed-forward quasi-one-dimensional SFRJ model with variable inlet control. An online learning controller based on retrospective cost optimization is integrated with the quasi-one-dimensional SFRJ model to regulate the thrust. Sensitivity studies are conducted on both the neural network and adaptive controller to identify optimal hyperparameters. Numerical simulation results indicate that the combined neural network and learning control framework can effectively regulate the thrust produced by the SFRJ model using limited in-situ data. I INTRODUCTION Solid fuel ramjet (SFRJ) platforms promise greater propulsive performance and capability over traditional rocket engines by ingesting freestream air for combustion with high-density solid fuels and without any complex turbomachinery [ 1 ] . Combustion involves complex interaction between turbulent flame-dynamics, solid fuel pyrolysis, and complex hydrocarbon chemistry. A review of the progress and challenges in understanding the SFRJ combustion phenomena is compiled in several review articles [ 2 , 3 , 4 ] . Due to the multi-physics nature of the dynamics, precise control over the combustor equivalence ratio and other performance metrics is extremely challenging limiting deployment at scale. For SFRJ platforms, direct command over system performance requires modifying the inflow conditions to the combustor subsystem. The majority of studies have considered the use of bypass air channels to modulate system thrust [ 5 , 6 ] . Experimentally, bypass ratio control has shown increased combustor efficiency [ 7 ] , but requires the addition of bypass piping which increases the complexity of the SFRJ geometry whilst decreasing the theoretical capacity for solid fuel loading. Recent work has investigated the potential of using variable inlet design as an alternative means for regulating SFRJ performance [ 8 ] . Variable inlet control presents potential advantages by minimizing the actuator hardware footprint and providing a novel mechanism to regulate air ingested for combustion and, thus, thrust. However, due to the complex dynamic relationship between the inlet geometry and thrust generated, which is sensitive to geometric variations of the SFRJ, as well as variations in flow conditions and dependence on flight conditions, designing a control system to regulate the SFRJ thrust remains a challenging problem. The focus of this work is thus the investigation of an adaptive control technique to regulate the thrust generated by an SFRJ. Previous controllers designed for thrust modulation of SFRJ engines have primarily assumed the generated thrust is available as an input to the control system [ 9 , 10 , 11 , 12 , 13 ] . However, in-flight measurement of generated thrust is extremely challenging, particularly for supersonic and hypersonic vehicle platforms [ 14 ] . Many previous works have been conducted to characterize SFRJ performance, resulting in the development of empirical relations relating inflow conditions to system performance [ 1 , 15 , 16 ] . Although useful in elucidating principal determinants, continued improvement in the understanding of the native complex multiphysics process, through experiments and numerical simulations [ 17 ] , is still necessary to develop highly accurate phenomenological models. Alternatively, thrust models for the SFRJ system can be constructed using data-driven and machine learning approaches. An inherent advantage of machine learning is that a thrust model can be developed from any available input set. This enables the creation of a model based entirely on in-situ measurements obtained from a real-world supersonic platform. This work utilizes artificial neural networks (ANNs) to predict the thrust of a model SFRJ platform using in situ measurements. Previous studies have utilized neural networks to model and improve the performance of SFRJ and related propulsive platforms in a variety of ways. Thrust prediction in hybrid-rocket [ 18 ] and solid-rocket [ 19 ] engines has been successfully demonstrated using neural networks trained on both synthetic and experimental data, respectively. Many studies have considered neural networks in surrogate-based analysis and optimization [ 20 ] , where neural networks are used to model and improve the design of propulsion systems [ 21 , 22 ] . Analogously, neural networks have also been used to determine model sensitivities for solid fuel combustion to optimize experimental design [ 23 ] . Fundamentally, ANNs are known to be universal function approximators [ 24 ] , making them well-suited to quickly and accurately estimate thrust from limited measurements as considered in this work. This work investigates the application of an online, learning-based control design technique called retrospective cost adaptive control (RCAC) [ 25 ] to regulate the thrust generated by an SFRJ model with variable-geometry inlet as the input and ANN-predicted thrust with in situ measurements. RCAC has been recently demonstrated as a viable technique to synthesize an adaptive control system to regulate the thrust and prevent inlet unstart in a liquid-fuel scramjet engine [ 9 , 10 , 26 ] and to regulate thrust in an SFRJ [ 11 , 12 , 13 ] . However, in [ 11 , 12 , 13 ] , the SFRJ input was assumed to be the heat flux and the measurement was assumed to be direct thrust measurements. In contrast, this work considers a realistic input, that is, a variable-geometry inlet and in situ measurements to compute thrust. The paper is organized as follows. Section II describes the numerical SFRJ model and the ANN used to predict SFRJ thrust, Section III describes the data-driven, learning-based control system to regulate the thrust, Section IV-A1 describes several numerical examples demonstrating the application of the proposed approach to regulate the SFRJ thrust, and Section V concludes the paper. II Numerical Methodology This study considers steady cruise operation of a model 140 mm external diameter SFRJ at a flight Mach number of 3.25 and altitude of 30 km. A constant freestream temperature and pressure are assumed based on the 1976 US Standard Atmosphere model [ 27 ] . Table I provides details of the selected cruise condition. TABLE I: Overview of selected cruise condition M H [km] P_{t0} [Pa] T_{t0} [K] 3.25 30 63677 748.6 II-A SFRJ Model This work considers a static quasi-one-dimensional feed-forward SFRJ with variable cowl control, modified based on the model developed by DeBoskey et al. [ 8 ] . Figure 1 shows a schematic of a SFRJ platform with thermodynamic stations labeled. Freestream air, station “0”, is ingested and compressed through the inlet and isolator system, station “1”, and enters the combustor entrance, station “2”. The solid fuel is embedded within the combustion chamber in station “3”; the port radius r_{3} increases as the fuel is consumed during combustion. Additional mixing and combustion occurs at the combustor aft-end, station “4”, before expansion and exhaust through the nozzle, station “e”. Fig. 1: Schematic of model SFRJ projectile with thermodynamic stations labeled. System performance is controlled through the translation of a variable cowl mechanism. A generic axisymmetric nose-spike center body inlet

Read the complete text