Modeling spatial distribution of particles in transportation systems using computational fluid dynamics and machine learning approaches

Zeinab Bahman Zadeh

doi:10.17918/00010886

This research presents a framework for modeling and predicting the spatial distribution of particle concentrations in public transportation systems, specifically buses and trains, using a combination of Computational Fluid Dynamics (CFD) simulations and Machine Learning (ML) techniques. By integrating these methodologies, the study provides valuable insights into improving air quality and ensuring passenger safety through enhanced ventilation strategies, emitter configurations, and data-driven risk assessments. The parametric analysis revealed significant differences in particle behavior between buses and trains. Probability Density Functions (PDFs) and Cumulative Distribution Functions (CDFs) were used to identify high-risk scenarios, assess variability, and analyze the impact of ventilation rates (ACH), emitter configurations, and thermal conditions. Buses exhibited broader Relative Concentration distributions with higher particle concentrations, whereas trains demonstrated more consistent particle dispersion. Higher ACH levels were found to stabilize air quality, reducing RC values and variability, while emitters placed at the front of vehicles provided the most efficient particle dispersion. Thermal conditions had a minimal influence on RC; therefore, it can be concluded with certainty that thermal conditions did not significantly impact particle distribution behavior. CFD simulations further highlighted the critical role of ACH in reducing particle concentrations, with mean RC values decreasing by up to 73% in buses and 62% in trains as ACH increased. The number and location of emitters significantly influenced particle accumulation, with back-located emitters producing the highest RC values. Spatial analysis showed that certain seating areas consistently experienced higher RC values, such as seats 14-19 in buses and seats 30-54 in trains, indicating particle "hotspots." 3D mapping revealed that the front zones in buses and the middle zones in trains exhibited the highest probabilities of RC > 1, underscoring ventilation inefficiencies in these areas. Machine learning models were employed to predict RC values, with Neural Networks, Random Forest, and Gradient Boosting models achieving high accuracy (R² = 0.87-0.90) and low error rates. These models outperformed linear regression and support vector regressors, which struggled with the nonlinear nature of particle distribution. However, model accuracy declined for higher RC thresholds, highlighting the challenges posed by data imbalance in extreme cases. This study offers a robust framework for analyzing particle distribution in public transportation, identifying high-risk zones, and developing effective mitigation strategies. By leveraging CFD and ML techniques, the research provides actionable insights to enhance ventilation design and passenger safety, paving the way for safer and healthier public transportation systems. Keywords: Particle Distribution, Computational Fluid Dynamics (CFD), Machine Learning (ML), Transportation System, Relative Concentration (RC), Parametric and Statistical analysis, Air Changes per Hour (ACH), Emitters, 3D Mapping, Exposure Risks

Modeling spatial distribution of particles in transportation systems using computational fluid dynamics and machine learning approaches

Files and links (1)

Abstract

Metrics

Details

Modeling spatial distribution of particles in transportation systems using computational fluid dynamics and machine learning approaches

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media