1.Introduction Scramjet is pivotal for achieving hypersonic flight,enabling vehicles to operate at speeds exceeding Mach number 5.This engine is crucial for next-generation aerospace systems,includ-ing hypersonic missiles,spaceplanes,and high-speed aircraft.However,the development of scramjet is fraught with signifi-cant technical challenges,such as supersonic combustion,ther-mal management,and dynamic flow control.In recent years,advancements in Artificial Intelligence(AI)have opened new avenues for addressing these challenges,offering innovative solutions that complement traditional engineering methods.Progress and prospects of AI in supersonic combustion have been summarized in our previous works,1 especially the appli-cation in scramjets.

This study presents an innovative parameter optimization framework that integrates swarm intelligence techniques with residual neural networks to enhance the accuracy and efficiency of ethylene combustion mechanisms in scramjets under extreme conditions. By leveraging machine learning and data-driven optimization, the framework achieves rapid mechanism simplification, intelligent optimization, and comprehensive validation. An adaptive pre-partition algorithm, based on a residual neural network surrogate model, enables self-adaptive search space partitioning, achieving 80 % precision with a minimal training dataset of 320 samples and only 7.5 % missed high-value samples, thereby accelerating the optimization process by 59.99 %. A novel objective function incorporates physical constraints and a penalty mechanism to ensure calculation convergence. Comparative analyses reveal significant improvements: ignition delay time mean relative error (MRE) is reduced by 98.6 %, equilibrium temperature MRE by 0.8 %, and equilibrium product concentration MRE by 10.36 %. Validation through numerical simulations and experiments demonstrates high agreement, with a 90 % correlation between optimized and detailed mechanisms and 98.94 % for wall pressure measurements. This framework advances combustion science by showcasing machine learning's role in optimizing complex mechanisms. Future research will focus on higher-accuracy single-step mechanisms and macromolecular hydrocarbon fuels, contributing to high-speed combustion and sustainable energy solutions.

This study experimentally investigates the influence of kerosene equivalence ratio (ER) on flame flashback characteristics in a supersonic combustor with spanwise uniform fuel injection. Synchronous high-speed imaging (chemiluminescence and schlieren) combined with high-frequency pressure measurements captured flame evolution, flow-field dynamics, and wall pressure data. Analysis of this multi-source data identified four distinct combustion modes: pure scramjet, scramjet, ramjet, and transitional. Based on observed phenomena, combustion states were categorized into five types: combustor flashback, combustor-stable combustion, isolator-stable combustion upstream of injectors, isolator flashback upstream of injectors, and upstream-downstream flame oscillation near the combustor inlet. Results demonstrate that while adjusting ER alone shifts combustion modes, it cannot fully prevent flashback. However, within a given mode, increasing ER suppresses flashback onset. Additionally, a quasi-one-dimensional analysis equation was discretized using a first-order forward difference scheme, yielding an empirical solution method that simplifies computation. Pressure signal acquisition delays due to flame flashback were quantified, with measured delays reaching 19.86 ms.

Abstract: To enhance the resistance of the compression surface boundary layer to adverse pressure gradients and mitigate the risk of non-start in air-breathing engines, promoting an early transition to turbulence is crucial. This study investigates the use of wall-injected jets to trigger boundary layer transition on inlet compression surfaces, proposing a novel analytical framework that incorporates Kelvin-Helmholtz (KH) instability. The methodology involves establishing grid independence using the γ-Reθ transition model and validating numerical simulations against wind tunnel data. Key flow parameters, including streamwise velocity and density, were extracted from the near-wall region downstream of micro-jets injecting monatomic gases of varying molecular weights. Analysis of the evolving jet shear layers revealed how the molecular weight of the jet injectant influences the transition process and wall heat flux. By examining the underlying mechanisms through KH instability and vorticity theory, it was determined that higher molecular weights intensify shear layer KH instability. These enhanced disturbances accelerate the boundary layer transition, leading to increased skin friction and wall heat flux. Compared to helium (the lightest gas), krypton (the heaviest) shifted the transition onset and completion locations upstream by up to 32.8% and 55.7%, respectively. Despite the transition promotion, the jets reduced wall skin friction relative to the no-jet case by diminishing Reynolds stress and the wall-normal gradient of streamwise velocity, thereby partially counteracting the associated increase in heat flux. These findings provide valuable theoretical support for advancing active flow control strategies based on jet-induced transition.

The isolator is a critical component in scramjet engines, playing a vital role in supersonic propulsion systems. However, it faces significant challenges, including flow separation and shock wave/boundary layer interactions, particularly under varying inflow conditions. Among the potential solutions, boundary layer suction technology has shown promise, but the complex interaction between suction parameters and flow characteristics remains poorly quantified. This study employs an air throttling technique to perform high-fidelity numerical simulations, systematically investigating the effects of various suction parameters on the leading-edge position of the shock train, closed region formed by u = 0 streamline and the wall, and suction flow rate under Mach 4.0 and Mach 7.5 conditions. The results indicate that suction slot length, aperture of suction slots, and suction back pressure have negligible effects on the flow field structure. In contrast, the number of suction slots is identified as the most significant influencing factor, while the suction angle emerges as a secondary but noteworthy parameter. Under Mach 4.0 conditions, asymmetric suction exacerbates flow separation, resulting closed region of 7.36%-more than double the 3.13% observed without suction. Meanwhile, the leading-edge position of the shock train shifts downstream by up to 31%. Conversely, symmetric suction reduces the closed region to 0.73% and induces a downstream displacement of the shock train leading edge by over 50%. At Mach 7.5, the variation pattern of the shock train leading-edge position after suction remains consistent with that observed at Mach 4.0. However, under higher-speed inflow conditions, the increased flow inertia suppresses separation phenomena, resulting in a continuous reduction of the closed region formed by the u = 0 streamline and the wall following suction. Through systematic simulations of the effects of various suction parameters on internal flow field characteristics, this study elucidates the influence mechanisms of boundary layer suction on shock train dynamics and flow structure modulation within the isolator section. These findings provide critical insights into optimizing adaptive suction control strategies, thereby expanding the engine's operational margin and enhancing flow stability in compressible flow systems.

To address the need for enhanced fuel mixing and improved combustion efficiency in a supersonic combustor, a modified aerodynamic ramp injection scheme, namely cascaded injection (L2W2), is proposed in this paper, and the combustion characteristics of kerosene with L2W2 injection, spanwise uniform injection (W), and streamwise distributed injection (L2) are investigated experimentally. Synchronous measurements were performed using high-speed schlieren, a high-speed camera, and pressure sensors, and a systematic analysis was conducted by combining digital image processing, Fourier transform, and wavelet analysis. The results indicate that the interaction between the supersonic incoming flow and the cavity induces an oscillation of approximately 161 Hz in the flow field, while both L2W2 and L2 can effectively suppress this oscillation, thereby achieving stable combustion. The flame flashback at the initial ignition stage is attributed to the low equivalence ratio, while the injection configuration is the key factor determining the subsequent combustion stability. For W injection, the densely arranged injector orifices in the spanwise direction cause the flow field to become two-dimensional, leading to poor mixing and thus resulting in sustained flame flashback. In contrast, L2W2 and L2 injections, by virtue of the sparse arrangement of the orifice groups and the deceleration effect of the upstream bow shock, effectively enhance the three-dimensional characteristics of the flow field and increase the fuel penetration depth, thereby achieving stable combustion. Among them, L2W2 injection yields the highest wall pressure and the smallest coefficient of variation of pressure, and is therefore proven to be an effective injection scheme that balances high heat release rate with good combustion stability. In addition, the low-pass filtering effect of the pressure transmission tube causes the pressure sensor to mainly capture low-and mid-frequency signals. Nevertheless, these low-frequency signals still provide an effective cross-validation basis for analyzing flame evolution, indicating that multi-parameter synchronous measurement is of great value for a more comprehensive characterization of the combustion process.

Accurate characterization of the combustor outlet temperature field is important for integrated combustor-turbine design and turbine thermal protection. However, direct measurement of the combustor outlet temperature field is challenging due to extreme thermal environment. To address this issue, a convolutional neural network (CNN)-based cross-domain reconstruction framework is proposed to infer the combustor outlet temperature field from downstream turbine guide vane (TGV) outlet temperature measurements. The proposed model incorporates skip connections and multi-scale feature extraction to improve the reconstruction of transport induced thermal distortion and localized hot streak structures. Temperature field datasets are generated from three dimensional unsteady simulations of a coupled combustor-TGV system. Results show that, using 50 temperature measurement points, the proposed model achieves an average peak signal to noise ratio (PSNR) of 28.83 dB, a mean outlet temperature error of 0.81%, and a relative outlet temperature distribution factor (OTDF) error of 8.9%. The proposed method demonstrates strong capability for cross-domain thermal field reconstruction and provides a feasible solution for sparse sensing in practical combustor-turbine systems.

As the core propulsion system of supersonic vehicles, the scramjet engine experiences unstable combustion phenomena in the combustor under high-speed operating conditions, which can lead to performance degradation and structural damage. Therefore, the development of supersonic flame stabilization structure identification technology is urgently needed. A Heterogeneous Feature Fusion Module (HFFM) is proposed to achieve nonlinear and organic fusion of heterogeneous data. Flame Structure Data (FSD) characterize key flame features, while Combustor Wall Pressure Data (CWPD) supplement the missing flame structure features in FSD, generating Flame Heterogeneous Feature Fusion Data (FHFFD). Additionally, a Supersonic Flame Stabilization Identification Module (SFSIM) is proposed, which combines a horn-shaped convolutional neural network with a Simplified Low Latent Transformer (SLLT) to enable dynamic and adaptive multi-scale integration of flame stabilization structure features. Experimental results indicate that HFFM effectively extracts and consolidates stable flame structure features within FHFFD during the training phase, demonstrating the ability to generalize key physical principles from FHFFD. SFSIM achieves a recognition accuracy of 97.09% through parameter optimization and attention-based dimensionality reduction. The low latent space improves training efficiency by 10.3%, while its parameter count accounts for only 29.73%. While maintaining high accuracy, this approach provides efficient and robust technical support for real-time monitoring of supersonic combustion.

This study investigates optimal control strategies for fuel pulse injection in scramjets under varying inflow conditions and thrust demands. A one-dimensional mathematical model was developed through CFD simulation data to capture the nonlinear relationships between injection parameters and performance metrics, enabling efficient control simulations. A pulse injection model was established, and an active disturbance rejection control (ADRC) algorithm was proposed to effectively manage injection parameters, demonstrating strong disturbance rejection capabilities essential for variable conditions. To address real-time tuning challenges, a real-time adjustment method (RL-ADRC) based on the twin delayed deep deterministic policy gradient (TD3) algorithm was introduced, allowing dynamic ADRC parameter adjustments based on environmental and operational states, achieving superior performance over traditional methods. Furthermore, a residual neural-network-based optimal control strategy was proposed to determine pulse injection parameters for varying thrust demands, ensuring desired thrust output, preventing inlet unstart, and maintaining high combustion efficiency. Comprehensive simulations validated the proposed strategies, confirming the robust performance of the pulse injection model, RL-ADRC, and optimal control strategy. These methods demonstrate strong potential to enhance the reliability and efficiency of scramjet operations under complex and dynamic conditions.

Pulse injection enhances mixing and promote combustion in a combustor, but its effects are influenced by many design parameters that are not fully understood. Traditional computational fluid dynamics (CFD) simulations are expensive for investigating these effects. This study combines traditional simulation with machine learning to explore the influence of pulse injection parameters on a supersonic combustion combustor. The results indicate that a pulsed injection equivalence ratio greater than 0.5 reduces combustion efficiency. Furthermore, increasing the pulse injection angle and modifying the backflow area improve fuel mixing and combustion. The neural network model predicts the shock train's front edge better, while the Kriging surrogate model excels in predicting thrust. Sensitivity analysis shows the injection equivalence ratio has the most notable influence on the flow field's structure and performance, followed by the injection angle, with frequency primarily impacting the total pressure recovery coefficient.

To address light-round failure during ignition in the vaporizer combustor of micro and small gas turbine engine (MS-GTE), this study investigates the effects of multiple parameters on atomization characteristics using PDPA and shadowgraphy. The results indicate that the inward contraction of spray distribution promotes flame propagation within a single vaporizer but hinders it between adjacent vaporizers. Increasing the exit diameter improves turbulent shear and fuel mist distribution while reducing SMD at the spray boundary. Circumferential holes at the exit induce localized wall-attached vortices that expand spray coverage outward while reducing SMD and mitigating inward spray contraction. Higher air pressure differential promotes continuous-phase breakup and spray expansion. Elevated wall temperatures enhance atomization, though this improvement tends to saturate at higher levels. Furthermore, ignition and light-round experiments were conducted using a three-sector combustor with high-speed imaging to validate the atomization findings. Vaporizers with larger exit diameters improve atomization uniformity and fuel mist dispersion, enabling shorter ignition and light-round timing at lower fuel-air ratios. The circumferential holes expand fuel mist dispersion, filling fuel vacuum regions between adjacent vaporizers and significantly enhancing ignition and light-round performance. The dominant mechanisms in each phase determine how various vaporizer configurations influence ignition and light-round behavior.

The traditional forward design method of the scramjet nozzle is difficult to obtain good performance under strong geometric constraints. Meanwhile, the existing optimal design methods rarely design from the perspective of the overall torque balance of the engine, and often only take into account the performance of the nozzle itself. This paper introduces an innovative inverse design method for the pitching moment of Single Expansion Ramp Nozzles (SERN). The core of this method integrates the Particle Swarm Optimization (PSO) algorithm with the Grey Wolf Optimization-based Kernel Extreme Learning Machine (GWO-KELM). A high-precision surrogate model of nozzle performance is constructed using a data-driven approach. Based on this surrogate model, performance constraints for PSO are established according to the desired moment. Nozzle design parameters are then iteratively optimized to achieve maximum thrust and minimum moment. The proposed method's effectiveness and accuracy are verified using Computational Fluid Dynamics (CFD). In twelve inverse design experiments, the average absolute percentage error between the designed and expected moment is 0.75 %. Compared to the reference nozzle profile, these designs achieve precise moment control while significantly improving thrust and reducing drag under strict geometric constraints. In conclusion, this paper presents an effective SERN design method, enhancing integration in hypersonic vehicles.

This study proposes a new time-space adaptive implicit-explicit (IMEX-TSA) method to accelerate unsteady detonation simulations. The proposed IMEX-TSA has a locally linear implicit treatment and adapts local time steps to local stiffness while preservinsg second-order accuracy in time. The IMEX-TSA consists of three key parts: time step estimation, dynamic subdomain determination and a second-order linearized implicit-explicit local time stepping (IMEX-LTS) scheme to update subdomain. In estimating local time steps at each cell, both the flow and reaction time scales are considered. To mitigate the computational expense associated with local time steps at each cell, cells with similar time steps are clustered into dynamic subdomains primarily through a hierarchical clustering algorithm, which limits the local time step reduction at a minor cost. The IMEX-LTS scheme achieves second-order temporal accuracy by adding an additional stage on cells at subdomain boundaries. This scheme ensures flux conservation through flux accumulation and correction. The stability of sub-computations is also maintained. The time step limits and the maximum efficiency of IMEX-TSA are compared with the original method by adjusting the time step parameters (e.g., CFL number). Numerical examples show that the proposed method achieves a ratio of the local time step in the non-reactive zone to that in the reactive zone ranging from 2.9 to 241.7, and boosts the maximum efficiency by factors of 2.3 to 61.0. The relative errors of selected physical variables are between 0.1 % and 2.2 %.

The flow characteristics of scramjets are crucial for their overall performance, necessitating the precise and efficient reconstruction of the combustor's flow field for accurate predictions. This paper tackles the challenges posed by the multitude of parameters and lengthy inference times associated with existing deep learning models by introducing a lightweight architecture specifically designed for supersonic combustor flow reconstruction. By combining the image generation capabilities of generative adversarial networks with the sophisticated feature extraction provided by a multi-head attention mechanism, the proposed model significantly enhances inference speed while preserving high performance. Validation was performed using a high-fidelity dataset derived from hydrogen combustion experiments conducted in a direct-connect pulse combustion wind tunnel, at a constant Mach number with systematically varied equivalence ratios. The results demonstrate substantial improvements, achieving a Structural Similarity Index Measure (SSIM) of 0.554 and a Peak Signal-to-Noise Ratio (PSNR) of 19.972, all while reducing the parameter count by 95 % and boosting inference speed by 56.58 % compared to the previously developed Multi-Branch Fusion Convolutional Neural Network (MBFCNN) model.

Traditional nozzle design methods are time-consuming and challenging to achieve good performance under strong geometric constraints. In contrast, the optimization design of nozzles aided by machine learning methods requires massive datasets. This study proposes an innovative interactive active learning (IAL) surrogate model construction framework to achieve high accuracy using small datasets. The investigations indicate that the mean absolute percentage errors of the surrogate model developed using the IAL approach for predicting thrust (T), pitching moment (M), and drag (D) are 0.26 %, 2.71 %, and 0.14 %, respectively. The superiority of IAL is verified by comparing it with other surrogate model construction methods. Utilizing the particle swarm optimization algorithm assisted by the surrogate model, the multi-objective optimization of the nozzle is further investigated. By taking the optimized profile of the nozzle as the research object, the influence of design variables on the nozzle performance is studied in detail with the aid of an IAL-constructed surrogate model. Comparing the optimized typical profile with the reference profile, the T increases by 5.86 %, D decreases by 5.67 %, and M reduces by 38.79 %. The IAL proposed in this study effectively overcomes the need for large datasets to construct a high-accuracy surrogate model and improves the efficiency of nozzle design. The methods proposed herein can be widely utilized in engineering optimization problems.

Combustion instability remains a critical barrier to the reliable operation of scramjet engines, especially under high-speed, high-enthalpy conditions. This study investigates the use of unsteady fuel injection as a control strategy to suppress such instabilities in a kerosene-fueled scramjet combustor. Experiments were conducted in a direct-connect supersonic combustion facility at Mach 3.0, with fuel injection modulated at frequencies (171, 216, and 260 Hz) and equivalence ratios of 0.5 and 0.6. High-speed chemiluminescence imaging and pressure transducer measurements were employed to analyze flame dynamics and pressure oscillations. Results demonstrate that unsteady injection at 216 Hz, combined with an equivalence ratio of 0.6, induces a transition from an unstable mode to a stable ram mode, significantly suppressing pressure fluctuations and enhancing combustion intensity. In contrast, unsteady injection at lower equivalence ratios or non-optimal frequencies failed to mitigate instabilities and, in some cases, exacerbated them. To interpret these findings, a reduced-order model (ROM) was developed based on dynamic coupling and time-delay effects between the fuel supply and combustion zones. The ROM qualitatively predicts system stability by evaluating eigenvalues derived from experimental conditions and shows agreement with observed instability trends.

This paper investigates the combustion oscillation characteristics of kerosene fuel under varying equivalence ratios in a supersonic combustor, operating with an inflow Mach number of 3, a stagnation temperature of 1920 K, and a stagnation pressure of 2.9 MPa. The findings reveal that as the equivalence ratio increases, the oscillation amplitude decreases. Two distinct flame stabilization modes were identified during the oscillation process: cavity recirculation-stabilized combustion and shear-layer-stabilized combustion. At higher equivalence ratios, combustion oscillations are mitigated, and overall flame intensity diminishes, particularly in the shear-layer-stabilized mode. The spectral analysis of pressure oscillations identified three primary frequency ranges: 2.7-100 Hz, attributed to unstable fuel supply; 152 Hz, associated with flame flashback and blow-off; and 700-1000 Hz, linked to flow separation. Additionally, detailed schlieren and oil mist visualizations were used to capture the dynamic changes during the flame flashback process. A one-dimensional thermodynamic analysis further revealed the presence of a thermodynamic throat within the combustor, with thermal choking at the combustor identified as the main cause of flame flashback. The experimental phenomena, combined with theoretical analysis, show that appropriately increasing the equivalent ratio can maintain the shear-layer-stabilized combustion.

The traditional nozzle design method needs more time and economic cost, and it is difficult to obtain good performance under strong geometric constraints. In contrast, the nozzle optimization design process assisted by artificial intelligence technology is shorter in time, but requires a large dataset. In this paper, the multi-layered optimization framework design method is innovatively proposed to reduce the need for high-precision numerical simulation of nozzle in the design process. Experimental results demonstrate that the surrogate model employing this framework achieves mean absolute percentage errors of 0.02 %, 0.3 %, and 0.026 % for predicting thrust, pitching moment, and drag, respectively, in high-performance nozzle regions. Utilizing a multi-objective nonlinear grey wolf optimization algorithm, the nozzle's design is optimized. The optimized nozzle shows a 13.3 % increase in thrust, a 20.3 % reduction in moment, and a 10.8 % reduction in drag compared to the reference profile. Compared with the traditional optimization frame method, it is found that the multi-layered optimization frame method can greatly shorten the multi-objective optimization design time of the scramjet nozzle without loss of performance. The multi-layered optimization framework method proposed in this paper can effectively improve the efficiency of multi-objective optimization design of scramjet nozzle, and can be widely used in the engineering optimization design of hypersonic vehicle and its components.

The numerical calculation method has greatly promoted the process of optimal design of scramjet, but it still needs extremely heavy calculation for the model with complex thermochemical reaction. Data-driven deep learning relies heavily on a large amount of data in the face of complex nonlinear features. Therefore, combining ‘‘data-driven model” and ‘‘Navier-Stokes equation”, an intelligent prediction model of supersonic combustion flow process is constructed. This algorithm integrates the theory priors of combustion flow into the neural network model, and uses convolutional grouping and rearrangement to reduce the feature redundancy calculation, so as to achieve high-precision and high-efficiency prediction of velocity, density, pressure and temperature fields.This study makes a comprehensive comparison from two aspects of performance and efficiency.Unsteady scramjet multi-physical field dataset is constructed under different incoming Mach number conditions. The experimental results show that compared with other methods, the proposed algorithm can achieve the maximum Peak Signal-to-Noise Ratio(PSNR) improvement of 38.75% and Learned Perceptual Image Patch Similarity(LPIPS) improvement of 68.13% in predicting the average quality of images, and the computational cost of the model is reduced by 30.36% compared with other models. In addition, the high model can also effectively predict the unknown incoming flow condition.

This study investigates combustion-flow interactions within a scramjet combustor, examining how varying fuel equivalence ratios influence flow characteristics and combustion performance. By employing numerical simulations and experimental validations, we explore the dynamic interplay between combustion flames, shock waves, and localized flow fields, aiming to elucidate the evolution laws governing flow dynamics and flame propagation. The findings reveal that the recirculation zone within the cavity undergoes continuous disassembly and merging, inducing periodic oscillations in the flow field with a cycle duration of 3.6 +/- 0.2 ms. Hydrogen injection stabilizes the flow field by balancing upstream and downstream pressures, thereby creating a structured mixing flow field. Furthermore, under reacting flow conditions, a pre-combustion shock train forms due to the combined effects of heat release from combustion and compression caused by fuel injection. Notably, at a fuel equivalence ratio of 0.398, reducing hydrogen supply causes a shift in the combustion regime, accompanied by decreased heat release and weakened shock train intensity. The specific impulse reaches its maximum at a ratio of 0.453, while flow uniformity at the combustor exit is optimized at 0.496. This study contributes to the understanding of complex combustion-flow interactions in scramjet engines, offering valuable insights into optimizing engine performance.