Complex Flow Field Research Articles

Enhancing high-resolution reconstruction of flow fields using physics-informed diffusion model with probability flow sampling

The application of artificial intelligence (AI) technology in fluid dynamics is becoming increasingly prevalent, particularly in accelerating the solution of partial differential equations and predicting complex flow fields. Researchers have extensively explored deep learning algorithms for flow field super-resolution reconstruction. However, purely data-driven deep learning models in this domain face numerous challenges. These include susceptibility to variations in data distribution during model training and a lack of physical and mathematical interpretability in the predictions. These issues significantly impact the effectiveness of the models in practical applications, especially when input data exhibit irregular distributions and noise. In recent years, the rapid development of generative artificial intelligence and physics-informed deep learning algorithms has created significant opportunities for complex physical simulations. This paper proposes a novel approach that combines diffusion models with physical constraint information. By integrating physical equation constraints into the training process of diffusion models, this method achieves high-fidelity flow field reconstruction from low-resolution inputs. Thus, it not only leverages the advantages of diffusion models but also enhances the interpretability of the models. Experimental results demonstrate that, compared to traditional methods, our approach excels in generating high-resolution flow fields with enhanced detail and physical consistency. This advancement provides new insights into developing more accurate and generalized flow field reconstruction models.

Preventing scour of monopile foundations using a vertical rotation device

Scour around monopile foundations is a major cause of offshore wind turbine failures, necessitating effective scour prevention measures to ensure the reliability of monopiles. This study investigates a new scour prevention strategy utilizing a vertical turbine rotor. This measure can not only effectively reduce scour around monopile foundations in complex flow field environments but also generate clean electricity, adding value. The effectiveness of the device in mitigating scour is evaluated through experiments. Subsequently, the impact of the flow field around the pile foundation and the seabed shear stress are examined using numerical simulations. The influence of the device’s installation position on seabed shear stress is also discussed. The results indicate that the vertical rotating device significantly reduce scour around the pile foundation, thereby enhancing the reliability of monopile foundations.

Precise and accurate speed measurements in rapidly flowing dense suspensions

We introduce a method for precise and accurate measurements of particle speeds in dense suspensions flowing at high rates and demonstrate the utility of the approach for revealing complex flow fluctuations during shearing in a setup that combines imaging with a confocal microscope and shearing with a rheometer. We scan the focal point in one dimension, aligned with direction of flow, producing absolute measurements of speed that are independent of suspension structure and particle shape. We compare this flow-direction line scanning approach with a complementary method we introduced previously, measuring speed using line scanning in the vorticity direction. By comparing results in various flow conditions, including shear-thinning and thickening regimes, we demonstrate the efficacy of our new approach. We find that both approaches exhibit qualitatively similar flow profiles, but a comparative analysis reveals a 15%–25% overestimation in speed measurement using vorticity line scanning, with discrepancies generated by anisotropic suspension microstructure under flow. Moreover, in the thickening regime where complex flow fields are present, both approaches capture local speed fluctuations. However, line scanning in the flow direction reveals and precisely captures stagnation and backflows, a capability not achievable with vorticity line scanning. The approach introduced here not only provides a refined technique for speed measurement in fast-flowing suspensions but also emphasizes the significance of accurate measurement techniques in advancing our understanding of flow behavior in dense suspensions, particularly in contexts where strong non-affine flows are prevalent.

Emissions and Flame Stability Assessment of Hydrogen Addition and Air Dilution in a Micro Gas Turbine Combustor Using 0D/1D Modeling by Chemical Reactor Network

Abstract Hydrogen emerges as a promising fuel for clean and sustain- able electricity production when utilized in micro gas turbines (mGTs). However, some challenges linked to hydrogen usage must be overcome as its high reactivity can lead to increased NOx emissions and flame instabilities. Literature suggests that combustion air humidification and/or exhaust gas recirculation (EGR) may be methods to reduce this reactivity. However, these measures are limited due to the small operating range of the mGT combustor. In this context, a 20 kWth mGT combustor is investigated un- der various inlet conditions to assess the effect of EGR towards increased flame stability and emission control when operating under hydrogen-enriched methane firing, using a Chemical Re- action Network (CRN) model. The CRN modeling is a fast and low computational complexity tool to model combustion perfor- mance and emissions by representing complex reactive flow fields through idealized reactor models, leading to reduced computa- tional costs. This study examines partial load conditions, fuel compositions, and the effect of combustion air dilution through EGR. The CRN model of the combustion process is designed based on combustion zones extracted from the main flow fields with similar thermo-chemical states from CFD, while emission predictions are experimentally validated for different operating points. Predicted NOx and CO emissions by the proposed CRN model show an acceptable agreement with the experimental data at high power loads. However, its accuracy diminishes for loads below 70% due to the variability in model tuning parameters across different loads, whereas the

Numerical simulation and experimental study on guidance performance of hypersonic seeker under aerodynamic optical transmission effects

When a hypersonic seeker flies at high speed within the atmosphere, intense interaction with the incoming flow gradually develops into a complex turbulent flow field. This interaction results in complex thermal responses at the seeker window, causing aerodynamic optical effects such as image shift, jitter, and blur of the target image, thereby restricting the seeker's detection capability and accuracy. This paper uses a numerical simulation model for the guidance performance of a hypersonic seeker under aerodynamic optical transmission effects. The study focuses on an ellipsoidal seeker, with its supersonic flight simulation on the basis of the Reynolds-Averaged Navier-Stokes (RANS) equations to get a non-uniform gradient flow field. The correctness of the flow filed results can be verified by wind tunnel experiments. The transient temperature field of the seeker is solved using an unsteady thermal conduction-radiation coupled fluid-solid heat transfer method. Finally, the guidance performance of the hypersonic seeker under aerodynamic optical effects is predicted using the ray tracing method, which employs wavefront aberration, point spread function, degraded images, and image shift.

Dynamic Docking Anti‐Disturbance Control of Overactuated AUV: System, Method, and Lake Trails

ABSTRACTDynamic docking control technology is crucial for autonomous underwater vehicles (AUV) to perform tasks underwater. To enhance the docking success rate of AUVs during dynamic docking, this paper presents a robust anti‐disturbance control algorithm specifically designed for overactuated AUV dynamic docking scenarios. During a dynamic docking mission, the AUV's depth control is adversely affected by the complex flow field generated by the underwater recovery device. To address this issue, this research proposes an AUV control scheme that combines an extended state observer (ESO) with a combined disturbance rejection method of the elevator‐vertical tunnel controller. First, an ESO is constructed to estimate and compensate for complicated disturbances such as model uncertainty and environmental disturbances. These estimations are then incorporated into the control law to mitigate the effects of the complicated flow field interference experienced during the AUV's dynamic docking process. Second, as turbulence intensifies at the end of the docking stage, the vertical thrust allocation is achieved using a hyperbolic tangent transition function. This ensures the stability of the AUV's attitude and depth, thereby enabling precise docking. Finally, the effectiveness of the proposed control algorithm is verified through lake trials and compared against the classic proportional‐integral‐differential (PID) and active disturbance rejection control (ADRC) methods. The trial results indicate that the proposed control algorithm significantly reduces the pitch and depth errors of the AUV, resulting in a remarkable 91% success rate for dynamic docking (based on 45 tests). The lake trials demonstrate that the proposed control algorithm is highly precise and robust.

Swirl-induced hysteresis in a sudden expansion flow

Swirling sudden expansion flows are complex flow fields containing several coherent structures that depend on the swirl number and can exhibit hysteresis behavior between increasing and subsequently decreasing swirl levels. While these flows have extensively been studied in simple geometries, results involving special designed nozzles are scarce. Therefore, this paper aims to provide insights into a more complex geometry, specifically a two-step conical expansion with a converging outlet. Experimental data is acquired for changing swirl numbers at a Reynolds number in the range of 35, 000. Stereoscopic Particle Image Velocimetry is employed to characterize downstream flow structures, while Laser Doppler Velocimetry is used to characterize upstream structures and to determine the inlet swirl number. Several distinct flow patterns are found as a function of the swirl number and the identified flow patterns include, in order of increasing swirl, a Closed Jet Flow, an Open Jet Flow (OJF), and a Coandã Jet Flow (CoJF). A central positive axial velocity is noted for both OJF and CoJF downstream of the expansion due to the converging outlet geometry. At higher swirl numbers, Vortex Breakdown moves upstream into the nozzle until a negative axial velocity is noted in the inlet tube. For these higher swirl numbers, no hysteresis is observed in the inlet tube between increasing and subsequently decreasing swirl. However, downstream of the nozzle, it is observed that the CoJF detaches at a lower swirl number than the swirl number required for attachment, indicating a hysteresis effect between in- and decreasing swirl.

Three-dimensional particulate volume fraction reconstruction in the fluid based on the Lambert–Beer physics information neural network

The measurement of particle volume fraction in flow fields is of great significance in scientific research and engineering applications. As one of the particle detection techniques, the light extinction method is widely used in measuring nano-particles volume fraction in flow fields due to its simplicity and non-contact nature. In particular, in complex reactive flow fields like combustion reactions, the volume fraction of soot particulate and other particles can be accurately measured and reconstructed via the light extinction method that based on the Beer–Lambert law. This is crucial for exploring combustion phenomena, understanding their internal mechanisms, and reducing pollutant emissions. However, due to the enormous computational burden, current algebra reconstruction techniques struggle to achieve high-precision three-dimensional (3D) reconstruction of particles. Therefore, this paper originally proposes a 3D reconstruction algorithm based on the Beer–Lambert law physical information neural networks (LB-PINNs). By incorporating physical information as constraints into the particle reconstruction process, it is possible to achieve high-precision 3D reconstruction of particles in complex flow field environments with low computational cost. Meanwhile, to address the trade-off issues of reconstruction accuracy and smooth noise resistance in previous reconstruction algorithms, i.e., Tikhonov regularization, this paper employs dynamically adjusted regularization parameters in the LB-PINN algorithm. This approach ensures smooth noise-resistant processing while maintaining reconstruction accuracy, significantly reducing computation time and resource consumption. According to the experimental results, LB-PINNs demonstrate superior performance compared to previous reconstruction algorithms when reconstructing the soot volume fraction in complex reacting flow fields, i.e., combustion flame scenarios.

Fast Spatiotemporal Sequence Graph Convolutional Network-based transient flow prediction around different airfoils

A novel Spatiotemporal Sequence Graph Convolutional Network (ST-SGCN) data-driven model is proposed to predict transient fluid dynamics around airfoils using complex and unstructured flow field data, with the aim of reducing dimensions and expediting predictions. Graph Neural Networks directly interact with the flow field grid, capturing spatiotemporal physical features of grid nodes and their interconnections, while eliminating the need for complex preprocessing steps. The ST-SGCN model integrates a Graph Convolutional Network and a Graph Attention Network with a Deep Recurrent Neural Network that uses a Gate Recurrent Unit as the kernel, adeptly extracting spatial and temporal physical features of the flow field to accurately predict transient flow states. Preliminary airfoil flow experiments demonstrated the model's ability to continuously predict transient flow fields, achieving an average accuracy of 97% for both velocity and pressure field predictions, with a maximum error of approximately 10% in the testing dataset. Further experiments, varying angles of attack, airfoils, and Reynolds numbers, demonstrated the model's generalizability, extensibility, and adaptability, with prediction errors below 5% and a speedup of over 20 times.

Numerical study on flow and heat transfer of combustion chamber and turbine under thermal shock test

In order to more accurately analyze the influence of the interaction between the aero-engine combustion chamber and the turbine, this paper establishes a combustion chamber-turbine cross-component model based on a high-pressure first-stage guide vane thermal shock test device for numerical research. The SST k-ω turbulence model, non-premixed combustion model and P1 radiation model are selected, and the test conditions are used as boundary conditions to carry out fluid-solid coupling unsteady calculation in ANSYS FLUENT. The numerical results are in good agreement with the experimental results. The complex flow field information between the combustion chamber and the turbine is obtained by analyzing the numerical results. It is found that there is an interaction between the combustion chamber and the turbine. The results show that the hot spot at the outlet of the combustion chamber will cause a stronger thermal shock to the downstream of the high-pressure first-stage guide vane during the migration process. The influence of the high-pressure first-stage guide vane on its upstream can be traced back to the position 0.3 times the chord length from the outlet of the combustion chamber.

Numerical Simulation and Experimental Research on Effect of Flow Regimes for Refrigeration Performance of Wave Rotor

Abstract Wave rotors exchange energy through the interaction of fluids with different pressures to produce a cooling effect without the application of mechanical parts, which makes them suitable for the complex flow field environments. But the flow situation inside the wave rotor is complex, and research on the impact of various flow regimes for the refrigeration efficiency is currently incomplete. This study deals with the numerical simulation and the experimental verification of a four-port fixed rotor gas wave refrigerator. Various typical flow regimes within the wave rotor were obtained through simulation to analyze their impact on the refrigeration efficiency. To further demonstrate the effectiveness of the simulation results, experimental verification was conducted. The simulation results indicate that separation bubbles and vortices will be respectively generated on the upper and lower walls of the wave rotor tube during the gradual opening stage. Vortex affects the airflow flow and causes detachment of separation bubbles as the airflow moves, which results in the loss of pressure energy and kinetic energy of the incident gas. As a result, the loss caused by the gradual opening process leads to a decrease in the self-expansion ability of high-pressure gas during the gradual closing stage, thereby affecting the refrigeration effect. Therefore, reducing the influence of vortices plays an important role in improving the cooling efficiency of wave rotors. Finally, this study analyzes the main factors that affect the cooling effect within the wave rotor through flow regime and provides new insights for improving refrigeration efficiency.

Model optimization and mechanism analysis of two-stage ejector considering nonequilibrium condensation

Two-stage ejectors are widely used in multi-effect distillation and refrigeration systems owing to their vacuum and pressurization capabilities. However, most studies on two-stage ejectors are based on the dry-gas hypothesis, which neglects the universal physical phenomenon of condensation. A wet-steam model of two-stage ejectors is established and compared to the dry-gas model in this paper. Simulation results indicate that the proposed wet-steam model more accurately describes the performance and complex flow field characteristics of the two-stage ejector. Moreover, the nonequilibrium condensation mechanism of the fluid within the two-stage ejector under variable conditions is revealed. The results show that as the two-stage main nozzle pressure increases from 300 kPa to 550 kPa, the liquid mass fraction increases, and the droplet nucleation rate of the two stages decreases by 14.17 % and 13.14 %. When the first-stage actuating pressure is increased, the first-stage fluid restricts the expansion state of the second-stage main jet core and weakens the vapor condensation in the region of the shock chain. Furthermore, the experimental results demonstrate that compared with the dry-gas model, the prediction error of entrainment ratio and the secondary flow obtained by the proposed model are reduced by 26.67 % and 43.04 %, respectively.

Flow Field Characteristics at the Spanwise Edge of a Vegetative Canopy Model

This study investigates the complex flow field across a spanwise vegetative model canopy edge focusing on turbulent transport processes. Utilizing stereoscopic particle image velocimetry, the three velocity components were measured in wall-parallel planes at various elevations within canopy across the spanwise canopy edge. Conventional ensemble averaged results were contrasted with those obtained by conditionally averaged flow properties across instantaneous internal interfaces in the flow to understand their contribution to the ensemble average. The conditional average captured the strong gradients in mean velocities, Reynolds stresses, vorticity, swirling strength, and turbulent kinetic energy production across the dynamically changing instantaneous interface. In contrast, the conventional ensemble average smeared out the strong gradients. Small magnitudes of advective terms in the turbulent kinetic energy transport equation suggested weak secondary transverse flows in the present model canopy. The turbulent flow structure across the spanwise canopy edge was further investigated using Quadrant-Hole analysis for both averaging approaches. Conventional ensemble averaged results indicated a shift from sweep to ejection dominance when moving from canopy into the open patch, while the conditional average showed only sweep dominated transport. In contrast to a homogeneous canopy layout, below canopy height at the canopy edge, sweeps and ejections lose their dominance in vertical turbulent transport. The present results show that the dynamics of internal interfaces govern the ensemble averaged results and a possible implementation into existing models is proposed. The present results are expected to increase understanding of spanwise turbulent transport and aid in developing strategies to mitigate desertification.

Rapid identification and early warning of axial compressor stall based on multiscale CNN-SVM-FC model

Early prewarning of compressor stall and surge is crucial to avoid aircraft engine instability, yet it is challenging due to the complex and unstable flow field characterized by multiple modes and multiscale features. To enhance the multi-scale feature representation capability of Convolutional Neural Network-Support Vector Machine (CNN-SVM) algorithm, a novel classifier modelling method combined multiscale windows with CNN-SVM is introduced for stall prewarning in this paper, named Multiscale CNN-SVM-FC. Multiscale detection windows are utilized to adaptively identify various pressure features during the stall process. Additionally, to reduce the false alarm rate, a fuzzy control algorithm is integrated with the temporal accumulation of prediction results from the multi-branch network for joint analysis. A series of test data from a five-stage axial compressor at different operating speeds is used to verify this method. The results indicate that the proposed Multiscale CNN-SVM-FC method enhances the accuracy of classification and reduces the false alarm rate compared to the standard CNN-SVM model, achieving over 99% accuracy in identifying unstable states under various speeds. Compared to three traditional stall prewarning methods, the Multiscale CNN-SVM-FC model provides an average warning signal 164 milliseconds ahead of stall, and reduces the uncertainty associated with threshold selection, which typically relies on engineering experience.

Complex flow field analysis in Multi-Shaft stirred Reactors: Dynamics of Wave-Vortex coupling revealed by POD and DMD methods

Insufficient understanding of complex flow structures in multi-shaft stirred reactors hinders industrial adoption. In this work, Proper Orthogonal Decomposition (POD) and Dynamic Mode Decomposition (DMD) methods were used to analyze the Large-Eddy Simulation (LES) results of the stationary and rotating zones for three types of stirred reactors, including multi-shaft configurations. The results indicate that the flexible spatial distribution of the impellers in multi-shaft stirred reactors enhances fluid interactions, leading to the superior performance in energy cascading, flow field self-regulation, and wave-vortex coupling. Frequency analysis of single modes underscores DMD’s efficacy in interpreting complex flow phenomena. Based on this, the DMD modal coefficients were fitted to equations, clarifying the development process of wave-vortex coupling within the stirred reactors. Additionally, by integrating fundamental fluid mechanics equations with the outcomes of the DMD method, a wave-vortex coupling model was developed, which is anticipated to yield a more accurate representation of the flow field.

Complex lava tube networks developed within the 1792–93 lava flow field on Mount Etna (Italy): insights for hazard assessment

Lava tubes are powerful heat insulators, allowing lava to practically keep the initial temperature and travel longer distances than when freely flowing on the ground surface. It is thus extremely important to recognize how, when and where these structures form within a lava flow field for hazard assessment purposes, in order to plan possible interventions should a lava flow approach inhabited areas. Often being formed within thick and complex lava flow fields, lava tubes are difficult to detect, study and explore. In this study, we analyse the 1792–93 Etna lava flow field emplaced on a steep slope (>4°) which comprises several lava tubes located at different distances from the eruptive fissure, at different levels within the lava flow field, and showing various inner morphologies, with peculiar inner features related to their maturity and eruptive history. Our aim is to verify whether it is possible to connect the underground features with features observed on the lava flow surface in order to reconstruct the extension of the tube network and unravel the genetic processes. Our results show that, in the studied lava flow field, a clear correspondence is possible between shallow tubes emplaced late during the lava flow field growth and surface textures. In addition, vertical and horizontal tube capture is very widespread, and might be the primary process for lava tube persistence and long life. Our results might be applicable to other lava tubes on Earth and other rocky planets.

Influence of the shape and number of obstacles on the excitation effect of dust-containing gas explosion

To investigate the excitation effects of different shapes and quantities of obstacles in dust-containing gas explosions, this study utilized an Euler-Lagrangian model. This model simulated explosions involving dust-laden gas, applying various governing equations to address the roles of coal dust at different combustion stages and its dynamic forces within the flow field. The research examined a 9.5 % gas concentration explosion with 5 g/m3 coal dust in a straight pipeline, assessing the impacts of spherical, regular tetrahedral, and cuboidal obstacles, as well as combinations of 1–3 cuboidal obstacles and three differently shaped obstacles on the explosion's flow field structure, overpressure, and flame propagation speed. Key findings include: dust-containing gas explosions result in more complex flow field structures, and the excitation effects of coal dust and obstacles on flame propagation speed exhibit mutual inhibition. Obstacles increase the maximum overpressure locally without affecting the overall trend of increasing overpressure with distance in closed conduits. Cuboidal obstacles show the most pronounced excitation effects, followed by regular tetrahedral and spherical obstacles, respectively. The number of obstacles also progressively enhances their excitation effects. When three obstacles of different shapes are present, cuboidal and regular tetrahedral obstacles primarily affect the flow field structure, with flame propagation speed and maximum explosion overpressure slightly lower compared to the presence of three cuboidal obstacles alone. These findings offer valuable insights into the destructive effects of gas explosions and the excitation effects of obstacles, providing theoretical support for disaster prevention and recommendations for underground equipment design and layout.

Investigating Flow around Submerged I, L and T Head Groynes in Gravel Bed

Riverbank erosion poses a significant threat to the stability and integrity of river training structures. River training structures such as groynes are important components of sustainable development as they play a crucial role in mitigating flood risks, controlling erosion, and supporting the habitat for aquatic organisms. The habitats vary largely according to the groyne type. A comprehensive comparative analysis of the flow field around the I, L, and T head groynes in the gravel bed is drawn. This study will be of immense use for riverbank protection in hilly terrain where streams are mostly dominated by the gravel bed. Laboratory experiments were conducted in a channel with a sediment bed as gravel of size 9.36 mm. Consistent flow conditions were maintained, with a flow depth (D) of 0.136 m and Froude no (Fr) of 0.61. The performance of these groynes, quantified using Lp (length of bank protection), was investigated. LHG and THG, notably, instigate more profound scour depths, recording values of 0.295 D and 0.29 D, respectively, while IHG trails with the value of 0.21 D. The complex flow field involving velocity peaks, decelerated, and negative flow is discussed and is attributed to flow separation at the groyne tip and the horseshoe vortex. The Lp for each groyne was estimated, with the IHG providing the maximum bank protection of 1.2 L1, L1 being the transverse length of the groyne. The cost–benefit analysis revealed IHG as the most cost-effective structure. These findings contribute to optimization of riverbank stabilization efforts, enhancing the resilience of hydraulic infrastructure and ensuring the safety and wellbeing of affected communities and ecosystems. The results also provide valuable insight into bank protection by various groynes and highlight their contribution to enhancing the resilience of river systems.

Numerical Simulation of the Unsteady Airwake of the Liaoning Carrier Based on the DDES Model Coupled with Overset Grid

The wake behind an aircraft carrier under heavy wind condition is a key concern in ship design. The Chinese Liaoning ship’s upturned bow and the island on the deck could cause serious flow separation in the landing and take-off area. The flow separation induces strong velocity gradients and intense pulsations in the flow field. In addition, the sway of the aircraft carrier caused by waves could also intensify the flow separation. The complex flow field poses a significant risk to the shipboard aircraft take-off and landing operation. Therefore, accurately predicting the wake of an aircraft carrier during wave action motion is of great interest for design optimization and recovery aircraft control. In this research, the aerodynamic around an aircraft carrier (i.e., Liaoning) was analyzed using the computational fluid dynamics technique. The validity of two turbulence models was verified through comparison with the existing data from the literature. The upturned bow take-off deck and the right-hand island were the main areas where flow separation occurred. Delayed detached eddy simulation (DDES), which combines the advantages of LES and RANS, was adopted to capture the full-scale spatial and temporal flow information. The DDES was also coupled with the overset grid to calculate the flow field characteristics under the effect of hull sway. The downwash area at 15° starboard wind became shorter when the hull was stationary, while the upwash area and turbulence intensity increased. The respective characteristics of the wake flow field in the stationary and swaying state of the ship were investigated, and the flow separation showed a clear periodic when the ship was swaying. Comprehensive analysis of the time-dependent flow characteristic of the approach line for fixed-wing naval aircraft is also presented.

Deep learning-based source term estimation of hydrogen leakages from a hydrogen fueled gas turbine

Hydrogen fueled gas turbine is an efficient and environmentally friendly energy conversion equipment. However, it is prone to gas leakages from corrosion-prone components and pipe connections. In order to address gas leakage problems, source term estimation (STE) can be applied to estimate the location and intensity of the leakage sources. Traditional STE methods are based on an optimization procedure using an atmospheric transport and dispersion model, which requires considerable computation efforts and cannot meet the need of real-time implementation. The complex flow field around the gas turbine, the possible existence of multiple leakage sources, and the high-dimensional time series data further increase the difficulty of the STE problem. To address these challenges, in the present study, a deep learning based STE approach is proposed. The basic idea is to use long short-term memory auto-encoder (LSTM-AE) network to extract the encoded features of multi-sensor data and then use a deep neural network (NN) to establish the relationship between the encoded features and the source term parameters of hydrogen leakages. The multi-sensor data are generated using computational fluid dynamics (CFD) analysis to simulate relevant hydrogen leakage scenarios of concern. The results show that compared with other machine learning algorithms, the proposed approach has an overall best performance in terms of the STE accuracy. Specifically, its hydrogen leakage source localization accuracy is 0.9798, and the leakage strength estimation R-squared (R2) can reach 0.9632. When the training data proportion is only 20–30%, the proposed approach can still perform well, indicating the ability of the model to effectively predict the location and strength of the leakage source even with relatively less training data.