基于多核并行计算的超燃冲压发动机一维模型实时性研究

Real-Time Research of Scramjet One Dimensional Model Simulation Based on Multi-Core Parallel Computation

关键词:  超燃冲压发动机  计算流体力学  空气动力学  并行计算  实时模型

Key words:  Scramjet engine  Computational fluid dynamics  Aerodynamic  Parallel computation  Real-time model

摘要:

在发动机控制系统设计中，为了缩短设计周期、降低研发成本，需要建立面向控制的、较为精确的、实时性高的超燃冲压发动机性能计算模型，以保证模型精度、提高计算速度为研究目标，基于多核高性能计算仿真平台，开展了面向控制的超燃冲压发动机一维模型实时性优化工作。运用简化计算流程、改进C语言程序、开拓缓存区等方法有效提高了一维模型计算速度。创新性地尝试了计算流体力学并行化方法，对隔离段和燃烧室一维模型进行结构分解。计算网格平衡分配至多个中央处理器，并借助核间数据通讯实现多核并行计算。与串行模型计算结果对比，七核并行计算模型性能参数偏差不超过0.1%，全工况仿真时间小于30ms，计算耗时较优化前缩短了75%以上。实时性优化后的多核并行模型计算精度高、速度快、收敛性好，可以作为超燃冲压发动机控制系统设计和半实物仿真验证平台。

Abstract:

In the engine control system design， in order to shorten the design cycle of control system and reduce the cost of research and development， it is necessary to establish a good real-time scramjet engine model oriented to control. In order to ensure the model accuracy and improve the calculation speed， the real-time optimization of one-dimensional scramjet model for control was carried out based on the multi-core high-performance computing simulation platform. Firstly， the calculation speed of one-dimensional model was improved effectively by simplifying the calculation process， improving the C language program and expanding the cache area. Subsequently， the computational fluid dynamics parallelization method was tried innovatively， which decomposed the one-dimensional model of isolation section and combustor. The computing grids distributed multiple central processors in a balanced way and realized multi-core parallel computing with the help of inter-core data communication.Compared with the calculation results of the serial model， the deviation of the performance parameter of the seven-core parallel calculation model is less than 0.1% and the calculation time of the whole working condition is less than 30ms， and the calculation time is shortened by more than 75% compared with that before optimization. After real-time optimization， this parallel model has high computational accuracy， fast computation speed， good convergence， and can be used as a platform for scramjet control system design and hardware-in-the-loop simulation verification.