# Numerical simulation and parallel implementation of freight train air brake system

## TL;DR

This paper presents a parallel computing model to efficiently simulate the air brake system of long freight trains, improving simulation speed and accuracy.

## Contribution

The novelty lies in a parallel computing approach for simulating train air brakes, using C# with ThreadPool and Parallel libraries to enhance efficiency.

## Key findings

- Parallel computing outperforms serial computing in simulation efficiency, with speedup increasing with more CPU threads.
- ThreadPool implementation is more efficient than Parallel library due to reduced scheduling overhead.
- Simulation efficiency stabilizes when workload distribution is fully optimized, even with more vehicles.

## Abstract

Numerical simulations of railway air brake systems are becoming increasingly computationally demanding due to the growing complexity and length of trains. This study introduces a parallel computing-enhanced model for simulating pressure dynamics in heavy-haul trains. The proposed approach assigns dedicated threads to each vehicle, with additional threads managing the interface fluxes of brake pipe connections, and is implemented in C# using the ThreadPool and Parallel libraries. A train configuration consisting of one locomotive and 116 vehicles was simulated under various brake pipe pressure reduction conditions. The validity of the air braking system simulation model was confirmed through comparisons with experimental data. Furthermore, the impact of parallel computing on simulation efficiency was investigated. The results indicate that the efficiency of serial computing is primarily influenced by the CPU Boost Clock frequency. Parallel computing consistently outperforms serial computing, with speedup ratios increasing as the number of CPU threads grows. Although the simulation partitions tasks by vehicle units, the computations remain sufficiently fine-grained that the ThreadPool implementation outperforms the Parallel library, as it eliminates the dynamic scheduling overhead inherent to Parallel’s workload distribution mechanism. Additionally, parallel computing efficiency improves as the number of vehicles increases; however, beyond a certain threshold, efficiency stabilizes as workload distribution becomes fully optimized.