TL;DR
This paper introduces a novel framework for optimizing cabin environmental control in supersonic aircraft, balancing passenger comfort, air quality, and energy efficiency. It compares two fast calculation methods (POD and ANN) and two optimization algorithms (MOGA and Pareto search), finding POD to be more accurate and Pareto search to be more efficient, ultimately guiding the selection of optimal supply air parameters based on specific design priorities.
What Problem Does This Paper Address?
Supersonic aircraft present unique challenges for cabin environmental control systems (ECS). Their high heat flux and dense passenger occupancy can lead to discomfort, poor air quality, and increased energy consumption. Existing optimization methods for cabin environments often rely on computationally expensive simulations, making rapid design iterations difficult. Furthermore, there's a lack of a comprehensive framework specifically tailored to the demanding conditions of supersonic cabins, which necessitates a balance between thermal comfort (PMV), air freshness (air age), and energy efficiency (temperature difference). This paper aims to fill this gap by developing and evaluating a multi-objective optimization framework that leverages fast calculation methods to enable efficient design exploration.
Methods
The study employs a two-pronged approach: fast calculation methods and multi-objective optimization algorithms.
Fast Calculation Methods: * Proper Orthogonal Decomposition (POD): This technique reduces the dimensionality of complex flow fields by identifying dominant spatial modes. It's used here to create a predictive model from CFD data, allowing for rapid estimation of airflow and temperature distributions under various conditions. * Artificial Neural Networks (ANN): A black-box model trained on CFD data to predict cabin environmental parameters.
Multi-Objective Optimization Algorithms: * Multi-Objective Genetic Algorithm (MOGA): A population-based evolutionary algorithm that searches for a set of optimal solutions (Pareto frontier) by simulating natural selection. * Pareto Search Algorithm: An iterative algorithm that directly searches for non-dominated solutions, aiming to find the Pareto frontier efficiently.
The framework involves generating a dataset using CFD simulations, training the POD and ANN models on this dataset, and then using these fast models within the MOGA and Pareto search algorithms to identify optimal supply air parameters (velocity, temperature, and angle). Finally, the entropy weight method is used to objectively assign weights to the optimization objectives to determine the best overall solution.
Datasets
The study utilizes a dataset generated from CFD simulations of a supersonic aircraft cabin model. This dataset, comprising 48 cases, serves as the training and validation set for both the POD and ANN fast calculation methods. The specific boundary conditions for these cases are derived from defined ranges for supply air velocity, temperature, and angle. Additionally, 15 new boundary conditions were randomly selected for validation purposes.
Key Insights and Ideas
- POD's Superior Accuracy: The study highlights that POD generally outperforms ANN in predicting cabin environmental parameters like PMV and air age, exhibiting significantly lower errors. This suggests POD's effectiveness in capturing the underlying physics of airflow in complex geometries like aircraft cabins.
- Pareto Search Efficiency: The Pareto search algorithm demonstrates superior computational efficiency compared to MOGA, finding a larger number of Pareto-optimal solutions in a fraction of the time. This makes it a more practical choice for rapid design space exploration.
- No Single "Best" Solution: The research emphasizes that no single set of supply air parameters achieves optimal performance across all objectives (comfort, air quality, energy). This underscores the inherent trade-offs in multi-objective optimization and the need for decision-makers to prioritize based on specific requirements.
- Entropy Weighting for Objectivity: The use of the entropy weight method provides a data-driven approach to assigning weights to objectives, mitigating subjective biases and offering a more robust method for selecting the final optimal solution.
Key Results
- Fast Calculation Accuracy: POD achieved lower RRMSE values for PMV (0.0315) and air age (0.0174) compared to ANN (0.180 for PMV, 0.124 for air age).
- Optimization Algorithm Performance: Pareto search generated 60 Pareto frontier points in 7 hours, while MOGA found 18 points in 20 hours.
- Optimal Solution Trade-offs:
- Optimal Solution 1 (Pareto Search): Achieved a PMV of 0.73, air age of 152.59 s, and ΔT of 14.27 K, with supply air at 2.5 m/s, 15°C, and 0° angle. This solution was found to be superior for PMV and air age compared to random conditions.
- Optimal Solution 2 (MOGA): Achieved a PMV of 0.72, air age of 158.09 s, and ΔT of 13.73 K, with supply air at 1.68 m/s, 13.2°C, and 15.9° angle. While slightly better in PMV than Solution 1, it was outperformed in air age and ΔT.
- Sensitivity Analysis: Air supply angle was found to have a significant influence on air age, while temperature had a greater influence on PMV. Velocity and temperature showed high total-order indices for air age, indicating coupled effects.
Limitations and Future Work
- Limited Design Variables: The study considered only three design variables (supply air velocity, temperature, and angle). Future work could incorporate additional factors like diffuser geometry, number, and position, especially if different ventilation layouts are used.
- Model Retraining: Changes in ventilation modes or geometric configurations would necessitate retraining of the POD and ANN models.
- Experimental Validation: The authors acknowledge that the proposed methods and conclusions require validation through experiments conducted in actual supersonic aircraft cabins.
- Generalizability: While the framework is presented for supersonic aircraft, its potential application to other optimization problems and fields is suggested, implying further research into its broader applicability.
Summary updated