A Case Study on the Optimization of Cooling and Ventilation Performance of Marine Gas Turbine Enclosures: CFD Simulation and Experimental Validation of Key Inlet Parameters

Abstract

This study addresses the thermal management challenges of marine gas turbine enclosures by proposing an innovative optimization of the air intake design, enhancing thermal management capabilities without mechanical restructuring. Through Computational Fluid Dynamics (CFD), the research systematically optimizes key parameters including cooling air inlet pressure, positioning, and enclosure inlet diameter. The results demonstrate that elevating the cooling air inlet pressure to 300 Pa enhanced the entrainment ratio (η) by 9.55% and increased the pressure loss coefficient (PLC) by 2.06% compared to the baseline case (P_in = 0 Pa). An enclosure inlet diameter of 1100 mm optimizes entrainment efficiency (η = 0.331) and minimizes internal temperatures. The multi-objective optimization identifies the globally optimal configuration (D = 800 mm, P_in = 300 Pa, L = 1.6 m), which improves the entrainment ratio by 31.7% (η = 0.399) and reduces the average temperature at key monitoring points (T₁–T₅) by up to 14 K compared to the baseline, albeit with a marginal increase in PLC. This optimal configuration ensures that all local temperatures remain within the operational limit of 355 K. This research provides a theoretical foundation for enhancing marine power system performance and offers evidence-based guidance for engineering applications.

1. Introduction

The maritime industry faces increasing power demands driven by ship equipment modernization and growing requirements for autonomy and endurance. This trend imposes heightened requirements on gas turbine power generation capabilities while intensifying thermal management challenges within their enclosures [1,2,3]. Addressing these challenges requires implementing effective thermal control strategies, essential for maintaining gas turbine reliability and stability under high-temperature operating conditions [4,5,6].

Marine gas turbine designs traditionally incorporate mechanical drive units and critical electrical components [7,8,9]. The gas turbine is housed within a dedicated enclosure to protect these vital components from environmental factors while providing thermal insulation, soundproofing, and corrosion resistance. However, this enclosed environment often results in elevated internal temperatures, negatively impacting the performance and operational efficiency of both the gas turbine and adjacent electronic systems [10,11,12]. Moreover, the gas turbines’ high-temperature surfaces can generate secondary heating within the enclosure through radiative heat transfer, potentially causing equipment overheating and safety risks [13,14,15].

Addressing these thermal management challenges requires designing and implementing an effective cooling and ventilation system for marine gas turbines. Such a system must effectively reduce the internal enclosure temperature to maintain optimal gas turbine performance while limiting unnecessary energy losses and auxiliary power requirements from excessive ventilation [16]. The design must also consider installation costs, aiming to balance cost-effectiveness with thermal efficiency. Substantial research has been conducted in this field, employing numerical simulations and experimental tests to advance understanding and efficiency.

Current research in this domain primarily examines two fundamental approaches: active cooling ventilation and passive cooling ventilation. Active cooling ventilation relies on forced ventilation to reduce enclosure temperatures and mitigate fuel leakage explosion risks, with key findings centered on airflow distribution and structural optimization. For example, Bagheri and Vahidi [16] emphasized that optimal active ventilation requires uniform airflow and no stagnant zones, but their design focused on fan layout rather than inlet parameter tuning, achieving only a 5–8% reduction in average enclosure temperature without quantifying entrainment efficiency or pressure loss. Vahidi et al. [17] validated that baffles enhance ventilation performance, yet their study ignored the coupling effect of inlet pressure and diameter, leading to impractical retrofitting costs for existing vessels. Lucherini et al. [18] conducted scaled experiments to analyze flow characteristics, but their steady-state model failed to capture transient thermal fluctuations, and the reported entrainment ratio (η ≈ 0.28) was significantly lower than industrial demands. Common limitations of active cooling studies include overreliance on complex fan/baffle modifications, neglect of multi-parameter coupling, and failure to balance entrainment efficiency (η) with pressure loss coefficient (PLC).

Passive cooling ventilation utilizes high-temperature exhaust gases to induce external airflow, avoiding additional energy consumption. Key advances include nozzle/diffuser structural optimization: Shi et al. [19] integrated orthogonal design with genetic clustering to improve η by 10.19% and reduce peak temperature by 24.99 K, but their focus was on exhaust ejector geometry rather than inlet parameters. Sun et al. [20] proposed rectangular nozzles and multi-stage diffusers, which enhanced mixing efficiency but increased PLC by 12–15%, compromising overall engine efficiency. Maqsood and Birk [21] found that excessive mixing tube curvature degraded subsonic ejector performance (η reduced by 7%), highlighting the sensitivity of passive systems to structural complexity. Sheng et al. [22] reported that lobed nozzles improved thermal mixing efficiency but caused a sharp PLC increase (≥18%), making them unsuitable for low-cost retrofits. Passive cooling studies share critical limitations: overemphasis on complex ejector modifications (high cost, poor applicability to existing vessels), fragmented analysis of single inlet parameters (e.g., only diameter or position), and lack of a multi-objective optimization framework to balance η, PLC, and temperature distribution.

Both active and passive cooling research have laid a foundation for enclosure thermal management, but critical gaps remain that motivate this study: (1) Research focus bias: Most studies prioritize complex structural modifications (ejector nozzles, fans, baffles) over inlet parameter optimization—despite the inlet’s simpler structure and lower retrofit cost; (2) Incomplete parameter analysis: No systematic investigation of the coupled effects of three key inlet parameters (cooling air inlet pressure P_in, enclosure inlet diameter D, inlet position L) on comprehensive performance (η, PLC, T₁–T₅); (3) Lack of balanced optimization: Existing studies optimize single indicators (e.g., only η or temperature) without addressing trade-offs between conflicting metrics, leading to impractical designs (e.g., high η but excessive pressure loss).

Addressing this gap is critical because optimizing these readily modifiable inlet parameters offers a low-cost, non-intrusive pathway to enhance thermal management and eliminates the need for mechanical restructuring while directly improving the reliability, operational safety, and energy efficiency of marine gas turbines. This is particularly urgent for the marine industry, where aging fleets require cost-effective upgrades and modern vessels demand stricter thermal control to support advanced equipment and prolonged endurance.

To bridge this gap, the novelty of the present work is threefold. First, unlike prior studies centered on complex ejector geometries or overall ventilation layouts, we prioritize the cooling air inlet as the optimization target and leverage its structural simplicity to enable practical, low-cost modifications. Second, we conduct a systematic, coupled analysis of P_in, D, and L via CFD simulations and experimental validation, filling the gap of fragmented parameter research. Third, we integrate a multi-objective optimization framework (genetic aggregation prediction + entropy weight-TOPSIS method) to balance conflicting performance indicators, ensuring globally optimal solutions rather than single-criterion improvements, an approach rarely applied in previous enclosure cooling studies.

Furthermore, the methodological framework established herein is highly generalizable. Its core logic, which involves optimizing key inlet parameters to regulate internal flow uniformity, suppress vortex stagnation, and mitigate high-temperature zones, is transferable to thermal management systems for other enclosed power equipment, including diesel engine enclosures, generator compartments, and industrial gas turbine packages. For enclosures with different geometric dimensions or heat source characteristics, the method only requires scaling parameter ranges (e.g., adjusting inlet diameter thresholds) and redefining evaluation criteria (e.g., temperature limits) to match specific operational demands, ensuring broad applicability beyond marine gas turbine enclosures.

2. Physical Model and Methodology

2.1. Physical Model

This study establishes a physical model of the cooling ventilation system for marine gas turbine enclosures, as depicted in Figure 1 [19,23]. The enclosure measures 6400 mm by 2800 mm by 2670 mm. The specific dimensions appear in Figure 1a.

Figure 1b illustrates the distribution of research parameters and monitoring points. The cooling air inlet tube comprises three sections: the inlet section, the transition section, and the section connecting to the enclosure. The research parameter D (800 mm ≤ D ≤ 1250 mm) represents the enclosure inlet diameter. In practical engineering applications, the cooling air inlet tube’s front end connects to upstream equipment, with its diameter fixed by existing equipment interface constraints. Thus, diameter modifications are limited to the tube section near the enclosure inlet to meet engineering requirements. The research parameter L (1600 mm ≤ L ≤ 4000 mm) indicates the cooling air inlet position, while parameter P_in (0 Pa ≤ P_in ≤ 300 Pa) denotes the cooling air inlet pressure. As shown in Figure 1b, five temperature monitoring points (P₁–P₅) were strategically positioned at critical locations to assess the thermal state and potential hot spots inside the enclosure: P₁ was located directly below the compressor outlet flange, which monitors the outlet temperature of the compressor; P₂ was positioned just above the combustion chamber outlet flange, capturing the outlet temperature of the combustor; P₃ and P₄ were arranged circumferentially 180° apart on the turbine outlet flange (top and bottom, respectively) to assess the uniformity of the internal temperature distribution; and P₅ was placed near the bend of the exhaust diffuser to monitor the local airflow temperature, reflecting the overall cooling effectiveness of the ventilation system. The temperatures measured at these points are directly defined as the temperature evaluation indicators T₁–T₅ in the Methodology, establishing a direct correspondence between physical monitoring and performance assessment.

2.2. Modelling Assumptions

To balance computational accuracy, efficiency, and alignment with the research objectives (optimizing inlet parameters for cooling ventilation performance), the following key modeling assumptions were adopted, with brief justifications provided:

(1)

Steady-state RANS equations as the governing framework: The marine gas turbine operates in a long-term stable state during normal navigation, and transient fluctuations have negligible impacts on the core relationship between inlet parameters and cooling performance. Steady-state simulation avoids the high computational cost of transient calculations while adequately capturing the average flow and temperature fields critical to the study.

(2)

Air treated as an ideal gas: The operating temperature within the enclosure ranges from 300 K to 355 K (27 °C to 82 °C) and pressure is near atmospheric pressure, which falls within the applicable range of the ideal gas equation of state. This assumption simplifies the calculation of fluid density without compromising accuracy for the targeted operating conditions.

(3)

Surface-to-Surface (S2S) radiation model: The enclosure contains multiple high-temperature components (combustion chamber, turbine) whose radiative heat transfer significantly affects internal temperature distribution. The S2S model is suitable for complex geometries with multiple surfaces, enabling the accurate calculation of radiative heat flux between components—consistent with the need to characterize comprehensive thermal behavior.

(4)

Non-slip wall conditions with fixed emissivity (0.9): The gas turbine enclosure walls are made of metal materials with low surface slip potential, justifying the non-slip assumption. An emissivity of 0.9 was selected based on typical values for industrial metal surfaces [19,24], ensuring alignment with practical material thermal properties.

(5)

Neglect of conjugate heat transfer: The research focus is on the influence of inlet parameters on internal flow and temperature fields, rather than heat transfer within wall materials or internal auxiliary components. Ignoring conjugate heat transfer reduces computational complexity while preserving the core physical mechanism that governs cooling performance.

(6)

Simplified geometric model of internal auxiliary equipment: The geometric model omits detailed configurations of internal tubing and minor auxiliary components. These components occupy a small volume and have minimal impact on the overall flow field induced by inlet parameters. Simplification ensures computational efficiency without altering the key flow characteristics relevant to inlet parameter optimization.

2.3. Mathematical Model

This study was based on the Reynolds-Averaged Navier–Stokes (RANS) equations and conducted under steady-state conditions. The corresponding governing equations are given below [19,25]:

The mass conservation equation:

$$\mathsf{\nabla }\cdot \left(\rho \overrightarrow{u}\right)=0$$

(1)

where ρ represents the fluid density in kg/m³; ∇ denotes the Hamilton operator; u represents the velocity vector.

The momentum conservation equation:

$$\mathsf{\nabla }\cdot \left(\rho \overrightarrow{u}\overrightarrow{u}\right)=\mathsf{\nabla }\cdot \left(-pI+\Gamma \right)$$

(2)

where p represents the static pressure, Pa; $I=\left({\delta }_{ij}\right)$ is the identity matrix; $\Gamma =\left({\tau }_{ij}\right)$ denotes the viscous stress tensor.

The energy conservation equation:

$$\mathsf{\nabla }\cdot \left(\rho \overrightarrow{u}E\right)=\mathsf{\nabla }\cdot \left[\left(-pI+\Gamma \right)\cdot \overrightarrow{u}\right]-\mathsf{\nabla }\cdot \overrightarrow{q}$$

(3)

where $\overrightarrow{q}$ represents the heat flux vector, and E denotes the total energy per unit mass of the fluid.

To complete the Navier–Stokes equation set, an equation of state relating density (ρ) and pressure (p) is introduced, expressed in the following mathematical form:

$${\displaystyle \frac{p}{\rho }}=RT$$

(4)

The mathematical formulations of the Realizable k-ε turbulence model are presented as follows:

Turbulent kinetic energy:

$$\mathsf{\nabla }\cdot \left(\rho k\overrightarrow{u}\right)=\mathsf{\nabla }\cdot \left[\left(\mu +{\displaystyle \frac{{\mu }_t}{{\sigma }_k}}\right)\mathsf{\nabla }k\right]+G_k+G_b-\rho \epsilon -Y_M$$

(5)

where $\overrightarrow{u}$ denotes the velocity components of the fluid; σ_k represents the realizability correction coefficient for the turbulent kinetic energy dissipation rate.

Turbulent kinetic energy dissipation rate:

$$\mathsf{\nabla }\cdot \left(\rho \epsilon \overrightarrow{u}\right)=\mathsf{\nabla }\cdot \left[\left(\mu +{\displaystyle \frac{{\mu }_t}{{\sigma }_{\epsilon }}}\right)\mathsf{\nabla }\epsilon \right]+\rho C_{1}S\epsilon -\rho C_2{\displaystyle \frac{{\epsilon }^2}{k+\sqrt{\nu \epsilon }}}+C_{1\epsilon }{C}_{3\epsilon }{C}_b$$

(6)

where μ_t represents the turbulent viscosity; σ_ε denotes the correction coefficient for the turbulent dissipation rate.

The Surface-to-Surface (S2S) radiation model is designed for analyzing radiative heat transfer in systems with complex geometries and multiple surfaces. Within the gas turbine enclosure, multiple high-temperature surfaces, including the combustion chamber and turbine, generate significant radiative heat transfer that influences the internal temperature distribution. The S2S model enables accurate calculation of radiative heat flux between these surfaces, providing precise simulation of the thermal environment within the enclosure [19]. The mathematical formulation of the S2S model is presented as follows:

The total radiative heat flux emitted is expressed by:

$$q_{out,k}={\epsilon }_k\sigma T_k^4+{\rho }_k{q}_{in,k}$$

(7)

where q_out,k represents the total radiative heat flux emanating from the surface; ε_k denotes the emissivity of the surface; σ represents the Stefan–Boltzmann constant; q_in,k indicates the incident radiative heat flux from surrounding objects.

The incident radiative heat flux on surface k can be expressed as a function of the radiative heat flux emanating from the other surfaces:

$$A_k{q}_{in,k}={\displaystyle \sum _{j=1}^N}{A}_j{q}_{out,j}{F}_{jk}$$

(8)

where $A_k$ represents the surface area; $F_{jk}$ denotes the view factor between the surfaces.

2.4. Boundary Conditions

This study employed ANSYS 2021 for numerical simulations using an ideal gas as the working fluid. The SIMPLE algorithm is consistently implemented to address the coupled velocity and pressure fields within the simulations. The boundary conditions are specified in Figure 2a. The surfaces of the high-temperature gas turbine components are designated as non-slip walls with an emissivity of 0.9 [19,24], and the temperature configurations for each component are illustrated in Figure 2b.

2.5. Numerical Solution Setup and Convergence

The commercial finite-volume solver ANSYS Fluent (2021 R1) was employed for all simulations. The governing equations were discretized and solved using the following numerical schemes:

(1)

Pressure–Velocity Coupling: The Semi-Implicit Method for Pressure-Linked Equations (SIMPLE) algorithm was used for its robustness in solving incompressible and weakly compressible flows.

(2)

Spatial Discretization: A second-order upwind scheme was applied for the discretization of momentum, energy, and turbulence equations to achieve a balance between accuracy and computational stability. The pressure interpolation was handled using the PRESTO! (PREssure STaggering Option) scheme, which is recommended for flows involving strong pressure gradients or swirl.

(3)

Gradient Treatment: The cell-based least squares method was used for computing gradients.

The solution convergence was rigorously monitored based on the following criteria:

(1)

Residuals: The iterative computation was considered converged when the scaled residuals for continuity, momentum, and turbulence equations dropped below 1 × 10⁻⁴, and the residual for the energy equation fell below 1 × 10⁻⁶.

(2)

Physical Quantity Monitoring: In addition to residuals, the mass flow rate at the cooling air inlet, the entrainment ratio (η), and the area-weighted average temperature at the outlet of the mixing tube were monitored. Convergence was confirmed when these key global quantities exhibited no observable change (variation less than 0.1%) over a minimum of 200 successive iterations.

For the turbulence boundary conditions at the cooling air inlet, a medium turbulence intensity of 5% was specified, along with a hydraulic diameter equal to the inlet diameter, which is a standard practice for fully-developed internal flows entering a large enclosure.

2.6. Evaluation Indicators

Prior to analyzing the effects of various parameter modifications on marine gas turbine enclosure performance, it is crucial to establish performance evaluation criteria. Three key indicators were introduced: Entrainment ratio (η), Pressure loss coefficient (Pressure loss coefficient (PLC)), and Temperature indicators (T_s). These metrics enable a comprehensive and integrated assessment of the enclosure’s performance. The definitions and calculation methods for these indicators are presented below:

(1)

Entrainment coefficient (η)

The entrainment coefficient represents a dimensionless parameter that quantifies the entrainment capacity of the ejection cooling system, and is defined as follows [26]:

$$\eta ={\displaystyle \frac{G_2}{G_1}}$$

(9)

where G₁ represents the mass flow rate of the mainstream, and G₂ denotes the mass flow rate of the secondary stream, respectively, kg/s.

(2)

Pressure Loss Coefficient (PLC)

The pressure loss coefficient is a dimensionless parameter that quantifies the flow loss in the ejection cooling system and is defined as follows [24]:

$$PLC={\displaystyle \frac{P_1-P_2}{q}}$$

(10)

where P₁ represents the total pressure at the outlet of the exhaust plenum, P₂ denotes the total pressure at the outlet of the mixing tube, and q indicates the dynamic pressure at the outlet of the exhaust plenum, respectively, Pa.

(3)

Temperature Indicators (T₁–T₅)

T₁–T₅ are formally defined as the point temperatures measured at the five corresponding monitoring points P₁–P₅. No spatial averaging is applied, and each value directly reflects the instantaneous temperature at the specific monitoring location. The temperature indicator T₁–T₅ refers to the maximum allowable temperature for safe operation of the enclosure, set at 355 K (82 °C) in this study [16]. This 355 K limit applies specifically to the air temperature in a typical working plane of the enclosure, which is representative of the thermal environment surrounding non-high-temperature core components (e.g., electronic control modules, auxiliary pipelines). It is important to note that the high-temperature surfaces of the gas turbine (e.g., combustion chamber wall, turbine casing) inherently generate extreme local temperatures, and the section immediately adjacent to these core components is excluded from the 355 K limit because ejection cooling alone cannot substantially reduce temperatures in this region, and such exclusion aligns with practical engineering operating requirements.

2.7. Multi-Objective Optimization Framework Based on Entropy Weight-TOPSIS Method

The entropy weight-TOPSIS method is a comprehensive evaluation approach that combines the entropy weight method and the TOPSIS method. The entropy weight method determines the weights of evaluation indicators by calculating the information entropy of each indicator: the smaller the information entropy, the greater the information content of the indicator, and thus the higher the weight [27]. The TOPSIS method is a distance-based evaluation method, which ranks various alternative schemes by calculating the distance between each alternative scheme and the positive ideal solution as well as the negative ideal solution. Figure 3 systematically illustrates the methodology of this method and its specific operational process.

3. Algorithm Validation

3.1. Mesh Independence

This study employed Fluent Meshing in ANSYS 2021 for mesh generation. Figure 4 illustrates the mesh arrangement within the computational domain. Non-structured polyhedral grids were implemented for mesh generation. The grids underwent refinement in complex regions, including the inlet, outlet, nozzles, and ejectors. To ensure result independence from mesh density, five distinct mesh sizes were generated based on the baseline geometric model (P = 0 Pa, L = 1.6 m, D = 800 mm) and utilized for system performance analysis. Figure 5 demonstrates the influence of different mesh sizes on system performance, primarily assessed through the entrainment ratio (η) and the average temperature at the outlet of the mixing tube (T_out,avg).

As illustrated in Figure 5, when the mesh count reaches 6.54 million, additional increases in mesh count produce negligible changes in the entrainment ratio (η) and the weighted average temperature at the outlet of the mixing tube (T_out,avg). Balancing simulation speed and result accuracy, this study adopted a mesh count of 6.54 million for subsequent simulations.

3.2. Turbulence Model Validation

The experimental test system, illustrated in Figure 6a, was a scaled-down and geometrically simplified model designed to capture the core ejector-driven mixing process relevant to the full-scale enclosure. Key dimensions included a motive nozzle diameter of 140 mm and a mixing tube diameter of 240 mm. During validation tests, the motive nozzle inlet velocity was varied between 20 m/s and 35 m/s, corresponding to a Reynolds number range (based on nozzle diameter) of approximately 1.86 × 10⁵ to 3.26 × 10⁵, which ensured fully turbulent flow. Velocity measurements were obtained using a calibrated Pitot tube connected to a digital micro-manometer with an accuracy of ±0.5 Pa, leading to an estimated velocity measurement uncertainty of less than ±3%.

Furthermore, a numerical simulation model with identical dimensions and spatial configurations to the experimental test system was developed to verify the accuracy of the numerical simulation algorithm, as illustrated in Figure 6b. A high-precision Pitot tube measured the wind speed at the outlet of the convergent nozzle and the mixing tube outlet. Each test point represents the arithmetic mean derived from three independent measurements. Figure 7 presents a detailed comparison between the experimental and numerical simulation results.

As shown in Figure 7, the numerical simulation results demonstrated high consistency with the experimental results in terms of trends, with a maximum deviation of only 4.88%. Thus, the Realizable k-ε turbulence model proves suitable for further application in this study.

This validation setup replicates the fundamental jet entrainment and mixing dynamics central to the passive cooling mechanism. Its main limitations are: (1) the isothermal experimental conditions, which validate the flow field and turbulence model but not the heat transfer or radiation effects; (2) the simplified geometry, which omits the full-scale enclosure’s internal components. Thus, the validation confirms the CFD model’s accuracy for flow predictions, while the thermal results depend on the combined application of the validated turbulence model with the prescribed boundary and radiation conditions.

4. Results and Discussion

4.1. Effect of Cooling Air Inlet Pressure

This section analyzes the flow field and temperature field characteristics under four different inlet pressures (0 Pa, 100 Pa, 200 Pa, and 300 Pa) to address insufficient ventilation in the passive cooling ventilation system of marine gas turbines, particularly inadequate cooling air inlet under harsh operating conditions. The analysis aims to comprehensively elucidate the effects of varying inlet pressures on cooling system performance and efficiency. Figure 8 demonstrates the impact of varying cooling air inlet pressure on the entrainment ratio (η) and pressure loss coefficient (PLC).

Figure 8 demonstrates that both the η and the PLC exhibit significant sensitivity to cooling air inlet pressure. As the cooling air inlet pressure increased from 0 Pa to 300 Pa, the η improved from 30.32% to 39.87%, while the PLC increased by 2.03% relative to its initial value. Figure 9 reveals that increased cooling air inlet pressure enhances inlet velocity, enabling greater cold air entry into the enclosure and strengthening ejector capability. Additionally, the increased inlet velocity results in a higher pressure loss coefficient.

The variation in cooling air inlet pressure significantly influences the temperature distribution and flow characteristics within the enclosure. Figure 9 and Figure 10 present the streamline patterns and temperature contours at different cooling air inlet pressures.

Figure 9 demonstrates that increasing the cooling air inlet pressure caused the streamlines in the inlet section to transition from yellow to red, indicating an overall velocity increase within the enclosure. The velocity in area A of the enclosure showed an increasing trend with higher cooling air inlet pressures, evidenced by streamline concentration. Additionally, increased inlet pressure reduced the vortex stagnation phenomenon in area B of the enclosure. However, the heightened inlet velocity intensified disturbance effects at the enclosure’s bottom, potentially amplifying local vortex formation, as illustrated in Figure 9b.

Figure 10 illustrates that increasing cooling air inlet pressure transformed Area A within the enclosure from orange-green to blue-green, resulting in an overall temperature reduction of approximately 16 K and diminished high-temperature zones. At a cooling air inlet pressure of 100 Pa, localized high-temperature phenomena emerged at the enclosure’s bottom due to swirling vortices, though these remained within acceptable limits. Subsequent pressure increases eliminate bottom vortices, effectively reducing high-temperature regions. Figure 11 depicts temperature variations at monitoring points under different cooling air inlet pressures.

Figure 11 reveals that temperatures at monitoring points P₁ and P₅ initially increased then decreased with rising cooling air inlet pressure, reaching maximum values at 100 Pa. In contrast, temperatures at monitoring points P₂–P₄ demonstrated an initial decrease followed by an increase, reaching minimum values at 100 Pa. While increasing cooling air inlet pressure enhances the entrainment ratio, it simultaneously increases pressure loss. Notably, elevated inlet flow velocity may generate local vortices within the enclosure, necessitating comprehensive consideration of multiple factors in practical applications.

The observed enhancement in entrainment ratio with increased inlet pressure aligns with the fundamental principle that higher motive flow momentum enhances secondary flow entrainment in ejector systems, as noted in prior ejector studies [21,28]. However, the concomitant increase in pressure loss is a trade-off that must be managed. Our finding that a 300 Pa inlet pressure boosts η by 9.55% while increasing PLC by only 2.03% suggests a favorable efficiency gain relative to the added loss. This trend is consistent with the optimization goal of Shi et al. [19], who also sought to improve entrainment, though their focus was on nozzle geometry rather than inlet pressure. In contrast, studies on complex ejector geometries with swirl vanes [29] often report higher pressure losses for similar entrainment gains, highlighting the advantage of our simpler inlet-based approach for retrofit applications.

4.2. Effect of Enclosure Inlet Diameter

The cooling air inlet diameter significantly influences the enclosure’s ejector performance by directly affecting the jet flow area. In practical engineering applications, while the inlet tube diameter remains fixed, the diameter near the enclosure air inlet can be modified. This section analyzes four different enclosure air inlet diameters (800 mm, 950 mm, 1100 mm, and 1250 mm), examining their impact on system evaluation indicators. Figure 12 illustrates how variations in enclosure inlet diameter affect the entrainment ratio (η) and pressure loss coefficient (PLC).

Figure 12 reveals that both the entrainment ratio (η) and pressure loss coefficient (PLC) initially increased then decreased with enlarging enclosure inlet diameter. In Figure 12, as the diameter increased from 800 mm to 1100 mm, the expanded jet flow area enables more cooling air entrainment into the enclosure, resulting in increased η. However, a further diameter increase to 1250 mm raised the expansion ratio in the inlet section, exacerbating flow separation at the expansion, as shown in Figure 13d. This reduced effective jet area led to decreased η. The PLC varies with the effective jet area at the enclosure inlet; larger jet areas increase direct jet flow impact on gas turbine walls, elevating dynamic pressure loss. The PLC increases as the diameter expands from 800 mm to 1100 mm due to increasing effective jet area. However, at 1250 mm, increased flow separation in the inlet section reduces the effective jet area, causing decreased PLC.

After analyzing the enclosure inlet diameter’s effects on system performance indicators, attention turns to its influence on internal temperature distribution and flow field patterns within the enclosure. Figure 13 and Figure 14 present the streamline patterns and temperature contours at various enclosure inlet diameters, respectively.

Figure 13 shows that in area A within the enclosure, velocity increased and streamlines became more concentrated as the enclosure inlet diameter increased from D = 800 mm to D = 1100 mm. However, a further diameter increase to D = 1250 mm resulted in sparse streamline distribution in this area. This occurs because excessive inlet diameter causes flow separation within the inlet section, reducing cooling air velocity entering the enclosure. Area B, positioned left and below the inlet, exhibited an unstable vortex stagnation zone.

Figure 14 demonstrates that as the enclosure inlet diameter increased from 800 mm to 1100 mm, the low-temperature region (300 K–316 K) progressively expanded, resulting in enhanced high-temperature conditions within the enclosure. Specifically in area A, the temperature contour plot shifted from green to blue, indicating a substantial reduction in high-temperature zones, with a decrease of approximately 17 K. However, increasing the enclosure inlet diameter further to 1250 mm led to temperature degradation within the enclosure, attributed to flow separation in the inlet section, causing temperatures to rise predominantly within the range of 316 K–333 K. Local overheating zones emerged at the enclosure’s base. To facilitate a more direct observation of temperature patterns within the enclosure, temperature monitoring points were established at strategic locations. Figure 15 displays the temperature variations at these monitoring points across different inlet diameters.

Figure 15 shows that the temperature at all monitoring points demonstrated a pattern of initial decrease followed by increase with the expansion of the enclosure inlet diameter. Specifically, the minimum temperature at monitoring point P₄ occurred at an inlet diameter of 950 mm, while the other monitoring points recorded their lowest temperatures at a diameter of 1100 mm. The data reveals that at an enclosure inlet diameter of 1100 mm, temperatures at all monitoring points fall within 307 K to 315 K, representing the lowest overall temperature in the enclosure. In conclusion, a moderate increase in the enclosure inlet diameter effectively reduces internal temperature and improves ejector efficiency, and despite increased pressure loss coefficient, it remains acceptable. However, excessive enlargement of the enclosure inlet diameter may reduce the entrainment ratio and elevate internal temperatures.

The non-monotonic relationship between enclosure inlet diameter (D) and entrainment efficiency, with an optimum at D = 1100 mm, echoes the findings of ejector studies that identify an optimal area ratio for maximum entrainment [20]. The deterioration in performance at excessive diameters (1250 mm) due to flow separation is analogous to the performance degradation observed by Maqsood and Birk [21] when increasing the curvature of a mixing tube. Our result that a moderate diameter increase reduces internal temperatures is consistent with the goal of uniform cooling emphasized by Bagheri and Vahidi [16]. However, while they focused on active ventilation layouts, we demonstrate that a simple geometric adjustment in a passive system can similarly improve thermal uniformity without additional energy input.

4.3. Effect of Different Cooling Air Inlet Positions

This section examines the effects of four distinct cooling air inlet positions (1.6 m, 2.4 m, 3.2 m, 4 m) on overall system performance. Figure 16 presents the relationship between these varying cooling air inlet positions and their impact on the entrainment ratio and pressure loss coefficient.

Figure 16 reveals that the entrainment ratio decreased as the cooling air inlet position approached the mixing tube outlet, though the overall reduction remained modest at approximately 1.5% compared to the initial position at L = 1.6 m. The pressure loss coefficient displayed a pattern of initial decrease followed by an increase as the cooling air inlet position nears the mixing tube outlet, with maximum variation approximately 1% from the initial position at L = 1.6 m. The data indicate that both the η and the PLC remain relatively stable regardless of cooling air inlet position. Additional analysis of the cooling air inlet position’s influence on system performance will be conducted through examination of the internal flow and temperature fields within the enclosure. Figure 17 and Figure 18 illustrate the streamline patterns and temperature contour plots, respectively, for various cooling air inlet positions.

Figure 17 demonstrates that rightward movement of the cooling air inlet position significantly intensifies vortex disturbance within the enclosure. The areas adjacent to and below the inlet, as well as those near and beneath the gas turbine, showed increased vortex stagnation zones as the cooling air inlet shifted rightward. Velocity in these regions progressively decreased, particularly when the cooling air inlet position reached L = 4.0 m, generating extensive vortex stagnation areas with average velocities below 4 m/s, substantially impeding enclosure ventilation. These vortex stagnation zones significantly influence internal temperature distribution. Figure 18 indicates that rightward movement of the cooling air inlet position causes the temperature contour plot to transition from blue through green to red, signifying a marked increase in enclosure internal temperature. This primarily results from internal vortex formation impeding ventilation and cooling processes. At cooling air inlet position L = 4.0 m, substantial overheating occurred within the enclosure, potentially compromising gas turbine and electronic equipment performance and efficiency. This condition clearly fails to meet the operational requirements. Figure 19 presents the temperature variations at monitoring points for different cooling air inlet positions, enabling direct observation of internal temperature trends.

Figure 19 indicates that cooling air inlet position significantly affects temperature distribution within the enclosure. At monitoring point P₁, temperature steadily increases with rightward movement of the cooling air inlet position. Monitoring points P₂, P₃, and P₅ exhibited a pattern of initial increase, followed by decrease, then subsequent increase as the cooling air inlet shifted rightward. Conversely, monitoring point P₄ showed a decrease before increasing. In conclusion, while cooling air inlet position minimally affects entrainment ratio and pressure loss coefficient, it substantially influences internal flow and temperature fields. Optimal enclosure inlet positioning requires maximum distance from the gas turbine exhaust side to ensure smooth airflow, facilitating effective ventilation and cooling.

The relative insensitivity of η and PLC to inlet position (L) contrasts with the strong influence of position on internal flow patterns and temperature distribution. This underscores that global performance metrics may not fully capture local thermal management challenges. The finding that positioning the inlet far from the turbine exhaust (L = 1.6 m) yields better cooling aligns with the general design principle of separating cold air intake from hot exhaust streams to minimize thermal short-circuiting. This principle is implicitly applied in the unconventional manifold design of Dumitrescu et al. [30]. However, our systematic quantification of the position’s effect on internal vortices and temperature gradients provides new insight beyond previous studies that primarily focused on global ejector performance [19,28].

5. Multi-Objective Optimization of Enclosure Inlet Structure

It can be clearly observed from the above research trends that changes in the enclosure inlet parameters do not uniformly affect performance indicators. This is because an adjustment beneficial to one criterion may have an adverse impact on other criteria, which could potentially compromise the overall system efficiency. Therefore, optimizing the enclosure inlet design requires comprehensive consideration to achieve multi-dimensional balance, rather than pursuing isolated parameter improvements. In this section, a multi-objective coordinated optimization framework integrating the genetic aggregation prediction method and the entropy weight-TOPSIS method is adopted to perform multi-objective optimization on 7500 predicted design schemes, ensuring the fairness of the solutions.

5.1. Genetic Aggregation Prediction Method

The hybrid modeling strategy of genetic aggregation prediction combines the Genetic Algorithm (GA) with data aggregation technology and improves prediction accuracy through the collaborative optimization of data feature extraction and model parameter calibration [31]. Compared with the traditional response surface method, this method exhibits superior stability in terms of the evaluation index (goodness-of-fit) for the current dataset. The goodness-of-fit index quantifies the degree of alignment between the predicted values and observed values at design points, and is intuitively expressed as the deviation between data points and the regression line—larger deviations indicate the inadequacy of the model for specific problems. As shown in Figure 20, the genetic aggregation prediction method achieved excellent goodness-of-fit performance, making it more suitable for data characterization in this study.

The genetic aggregation prediction model in this study was constructed within the ANSYS Workbench optimization framework using the 10 initial design points (see

Table 1. The experimental design points.

Design Points	L (m)	Pin (Pa)	D (mm)	η	PLC	T1	T2	T3	T4	T5
1	1.6	0	800	0.303	14.255	311.141	316.595	315.886	329.546	317.292
2	2.4	0	800	0.303	14.217	313.615	319.847	321.419	324.250	327.349
3	3.2	0	800	0.297	14.116	325.785	299.973	317.146	323.605	314.831
4	4	0	800	0.289	14.161	344.002	301.846	320.852	328.307	321.117
5	1.6	100	800	0.327	14.341	320.987	314.683	314.980	317.187	320.100
6	1.6	200	800	0.366	14.444	311.817	314.730	314.728	320.732	319.466
7	1.6	300	800	0.399	14.549	310.254	315.070	315.752	323.274	318.451
8	1.6	0	950	0.326	14.417	309.095	314.920	321.376	310.569	312.617
9	1.6	0	1100	0.331	14.454	307.832	309.653	313.098	313.200	312.500
10	1.6	0	1250	0.315	14.254	309.437	315.333	314.266	321.928	318.508

). A Kriging surrogate model was employed to approximate the system responses. The integrated Genetic Algorithm (GA) was configured with a population size of 50, a maximum of 100 generations, a crossover rate of 0.9, and a mutation rate of 0.01 to drive design space exploration and optimize the model’s hyperparameters. The model’s goodness-of-fit was evaluated using leave-one-out cross-validation. The results, summarized in

Table 2. Accuracy assessment of the surrogate model predictions.

Performance Indicator	R2	MAE
Entrainment ratio (η)	0.992	0.0042
Pressure loss coefficient (PLC)	0.986	0.032
T1	0.981	0.78 K
T2	0.978	0.83 K
T3	0.985	0.68 K
T4	0.99	0.61 K
T5	0.983	0.75 K

, present the coefficient of determination (R²) and the Mean Absolute Error (MAE) for each key response metric. The high R² values (all exceeding 0.97) and low MAE values collectively confirm the high predictive accuracy and reliability of the constructed surrogate model.

Therefore, based on the aforementioned single-factor studies, this study systematically generated 10 experimental design points and constructed a predictive response surface using the genetic aggregation model. The detailed design points are presented in Table 1, and the response surface containing 7500 sets of sample data is shown in Figure 21.

As shown in Figure 21, to facilitate further verification of the accuracy of the predicted results, the variation trends of each evaluation index with respect to the three parameters were projected onto a two-dimensional plane. These predictions are highly consistent with the calculation results introduced in Section 4.1, Section 4.2 and Section 4.3, thereby verifying the reliability of the constructed predictive response surface. It is worth noting that compared with the inherent discrete limitations of traditional orthogonal experimental design, the 7500 sets of sample data generated through response surface prediction established a comprehensive initial sample library for the subsequent global optimization search.

5.2. Multi-Objective Optimization Results

The entropy weight-TOPSIS method comprehensively evaluated 7500 predicted schemes to determine the optimal structural design.

Table 3. Weight table of evaluation indicators.

Indicator	Information Entropy Value (e)	Information Utility Value (d)	Weight (%)
η	0.9888	0.0112	19.14%
PLC	0.9943	0.0057	9.63%
T1	0.9952	0.0048	8.19%
T2	0.9921	0.0079	13.51%
T3	0.9955	0.0045	7.65%
T4	0.9868	0.0132	22.44%
T5	0.9886	0.0114	19.45%

lists the calculated weights of each evaluation index.

The weight analysis results in Table 3 show that T₄ had the highest weight in the comprehensive evaluation system, and the remaining parameters were ranked in descending order of weight as follows: T₅, η, T₂, PLC, T₁, T₃, and ΔP. Based on this evaluation criterion,

Table 4. Comprehensive rating table.

D (mm)	Pin (Pa)	L (m)	Ideal Solutions (D+)	Negative Ideal Solutions (D−)	Relative Closeness (C)	Rank
800.00	300.00	1.60	0.1514	0.2902	0.6571	1
800.00	293.88	1.60	0.1533	0.2886	0.6531	2
800.00	287.76	1.60	0.1552	0.2870	0.6491	3
800.00	281.63	1.60	0.1570	0.2854	0.6451	4
800.00	275.51	1.60	0.1514	0.2902	0.6411	5
⋮	⋮	⋮	⋮	⋮	⋮	⋮
800.00	36.73	3.17	0.3157	0.1116	0.2611	7496
800.00	42.86	3.17	0.3171	0.1121	0.2611	7497
800.00	48.98	3.22	0.3171	0.1118	0.2606	7498
800.00	36.73	3.22	0.3155	0.1101	0.2587	7499
800.00	42.86	3.22	0.3169	0.1106	0.2586	7500

lists the comprehensive ranking of all configurations as well as the top 5 and bottom 5 configurations. The results indicate that the device achieved the optimal comprehensive performance when the cooling air inlet diameter D was 800 mm, the inlet pressure was 300 Pa, and the inlet position was 1.6 m.

The apparent discrepancy between the single-factor and multi-objective optimal results, specifically the selection of D = 800 mm over the diameter D = 1100 mm that maximized entrainment in isolation, is reconciled by the comprehensive weighting of all performance criteria. The entropy weight-TOPSIS optimization does not solely maximize the entrainment ratio (η); it seeks the best compromise among η, pressure loss (PLC), and the critical temperatures at all monitoring points (T₁–T₅). As indicated by the weights in Table 3, temperature indicators collectively hold over 50% of the decision weight, underscoring that thermal safety is the paramount design driver. While D = 1100 mm yielded a higher η, it also induced greater pressure loss and, more importantly, failed to minimize temperatures at all key locations as effectively as the D = 800 mm configuration under the optimal P_in and L. The chosen optimum (D = 800 mm, P_in = 300 Pa, L = 1.6 m) therefore represents the most balanced solution, prioritizing thermal safety with an adequate margin (all T < 355 K) while still achieving a 31.7% improvement in η over the baseline. Consequently, the robustness of this optimal design stems directly from the overwhelming dominance of the temperature constraints in the decision-making process, ensuring its stability against minor variations in the weighting of flow performance metrics (η and PLC).

From the above analysis, it can be concluded that the predictive modeling and optimization methods presented in this study provide strong decision support for the practical engineering design of complex thermal fluid systems.

5.3. Limitation and Future Work

While this study establishes a validated framework for optimizing the cooling air inlet and provides actionable design insights, its findings are bounded by specific modelling assumptions and scope. Acknowledging these limitations is crucial for interpreting the results and delineating clear pathways for future research.

The main limitations of the current work are summarized as follows:

(1)

Model Simplifications: The steady-state RANS approach with prescribed wall temperatures neglects transient effects and conjugate heat transfer, simplifying thermal inertia and solid conduction.

(2)

Geometric Abstraction: The omission of internal equipment and piping focuses on global flow but may not capture localized obstructions or heat accumulation.

(3)

System Boundary: The model isolates the enclosure, excluding external heat influx from the engine room and the system-level impact of cooling air pressure drop on overall plant efficiency.

(4)

Ambient Conditions: Simulations assumed constant ambient properties, leaving the influence of variable temperature and humidity on cooling performance unexamined.

(5)

Evaluation Method Limitation: The entropy weight-TOPSIS method is data-dependent, with weights derived from sample statistics rather than physical mechanisms. It quantifies trade-offs between heat exchange (T₁–T₅) and aerodynamic performance (η, PLC) but cannot explain the underlying fluid–thermal coupling mechanisms.

To address these limitations, future studies should: (1) employ transient simulations and conjugate heat transfer modeling; (2) incorporate high-fidelity internal geometries; (3) integrate the enclosure model with the gas turbine cycle for system efficiency analysis; (4) investigate the sensitivity of cooling performance to ambient conditions; (5) pursue comprehensive experimental validation under thermal loads; and (6) optimize the evaluation framework by integrating mechanism-driven weight methods to reduce data dependency.

6. Conclusions

This paper utilized CFD methodologies to perform comprehensive numerical simulation and optimization analysis of the cooling ventilation system in marine gas turbine enclosures. The investigation examined the effects of cooling air inlet pressure, enclosure inlet diameter, and cooling air inlet section positioning on ventilation cooling performance. Through analysis of entrainment ratio and pressure loss coefficient under various operating conditions, coupled with an examination of flow field and temperature distribution within the enclosure, the following conclusions emerged:

(1)

Validated a CFD numerical model integrated with the Realizable k-ε turbulence model and S2S radiation model for marine gas turbine enclosures, with experimental data showing a maximum deviation of only 4.88%, and conducted the first systematic investigation on the coupled effects of three key inlet parameters (cooling air inlet pressure (P_in), enclosure inlet diameter (D), and inlet position (L)) on cooling ventilation performance.

(2)

Systematically quantified the single-factor effects of key inlet parameters on enclosure cooling performance: elevating cooling air inlet pressure (P_in) to 300 Pa improves the entrainment ratio (η) by 9.55% (with a 2.03% increase in pressure loss coefficient); an enclosure inlet diameter D of 1100 mm optimizes η to 0.33 and minimizes internal temperatures, achieving a temperature reduction of approximately 17 K compared to the baseline diameter of 800 mm; positioning the inlet away from the turbine outlet (minimum L = 1.6 m) suppresses vortex stagnation and enhances cooling uniformity.

(3)

Combined the genetic aggregation prediction method with the entropy weight-TOPSIS method for multi-objective optimization: the integrated framework constructs a high-fidelity response surface (goodness-of-fit close to 1.0, 7500 sets of sample data), objectively weights indicators (e.g., T₄: 22.44%, η: 19.14%), and screens the optimal structure (D = 800 mm), P_in = 300 Pa, L = 1.6 m), which improved η by 9.55% and reduced the maximum internal temperature by approximately 16 K compared to the baseline.

(4)

Developed a low-cost, mechanically non-intrusive optimization solution: the solution requires no structural restructuring, avoids high costs of complex ejector modifications, and is generalizable to diesel engine enclosures or generator compartments; for marine use, it maintains all monitoring point temperatures at 307–315 K, well below the 355 K (82 °C) operational limit.