节点文献

内燃机一维流动计算方法研究及性能仿真软件设计

Computational Study on One-dimensional Fluid Flow and Design of Performance Simulation Software for Internal Combustion Engine

【作者】 陆霄露

【导师】 邓康耀;

【作者基本信息】 上海交通大学 , 动力机械及工程, 2014, 博士

【摘要】 近二十余年,随着计算机技术、计算数学和计算流体力学的迅速发展,内燃机一维循环模拟程序对于发动机的设计和性能改进发挥着越来越重要的作用,大幅度地降低了新发动机的开发周期和开发成本。已有的商业软件虽然功能强大,由于其源代码的保密性,用户自主开发的新模型、新算法很难和这些软件的内核协调地工作。而且这些软件的费用一般非常昂贵。因此,商用软件无法支持内燃机的前沿研究,必须开发自主的模型和软件。本文采用新的方法,深入研究了相关的一维流动计算模型和性能仿真软件的设计思路,开发了通用的高效高精度内燃机一维循环模拟程序。采用图论中的有向图来定义发动机热流体网络的拓扑结构。有向图中的节点用来描述各种发动机组件,有向边则用来建立这些组件之间的网络拓扑关系。采用这种定义方式,能够对带有任意个气缸、任意进排气管系的发动机进行建模,发动机还可以带有任意串并联的涡轮增压器。采用图论方法对发动机流体网络进行分析。利用图论中的深度优先搜索算法再结合发动机流体网络的特点,可以精确辨别进排气管;通过对有向图中的组件进行拓扑排序,可以自动生成从上游到下游的求解顺序。这些工作为开发通用的一维循环模拟程序奠定了扎实的基础。进排气管道内的流动特性对整机性能有着很大的影响。为了克服现有特征线法流量不守恒的问题,采用基本无振荡有限体积法进行排气管系的流动分析。首先使用有限体积法把每根管子划分为若干个控制容积,然后采用高精度ENO格式计算管道内部控制容积交界面上的通量,管子两端的边界条件则通过特征线法得到。这样做不仅保持了流量守恒、而且还提高了计算精度。采用具有精确解的激波管,比较了不同阶数的ENO格式对计算精度和计算效率的影响。结果表明,ENO格式阶数越高计算精度越高。从1阶到2阶ENO格式,数值解和解析解的误差下降的最快。随着精度的提高,误差下降得越来越慢,但是计算时间却成倍地增长。实际应用时,推荐使用2阶ENO格式,精度高又能保证计算效率。为了进一步提高一维非定常流动的计算效率,提出了自适应当地时间步长推进算法。发动机流体网络中每根管道的计算采用不同的时间步长,不再受其他管道时间步长的约束。时间步长的不同步导致边界条件处理的困难。通过自适应动态调整管道的求解顺序解决了上述问题。实际计算表明:新的时间层推进算法对计算精度没有影响,但是可以节约25%-40%的计算时间。三通接头模型对于进排气质量流量和压力波的模拟精度有比较大的影响,进而对整机性能模拟产生重要影响。通过对柴油机进排气管中常见三通接头在较高空气流速下冷态吹风试验研究,获取流动测试数据,得到了三通接头处的压力损失变化规律。实验所得的总压损失系数和采用Vazsonyi公式计算所得的结果有一定的偏差,但是两者随流量比的变化趋势相同。流速越高,这种偏差越明显。通过改进Vazsonyi公式,同时引入分配函数,建立了考虑较高流速的修正型压力损失模型。对排气管涡前压力波的计算研究表明,修正型压力损失模型比常规压力损失模型的计算精度提高4.3%。在采用有向图定义的流体网络基础上,建立了循环模拟程序的总体框架,设计了有效的数据结构来实现这种框架。通过研究串行程序中逻辑上并行的区域,本文首次采用并行算法把原来的串行求解器改造成并行求解器。在多核PC机上,使用并行计算程序对一维流动及整机性能进行模拟。计算结果表明,并行程序计算所需的时间可以缩短到串行程序的1/3都不到,计算效率显著提高。搭建了D6114增压柴油机实验台架,对排气压力波在全工况范围内的变化规律进行了实验研究。采用前面自主开发的程序,进行了计算和实验的全工况对比。在各个工况下,涡前排气平均压力误差在5%以下。发动机转速低时,误差更小。在一个工作循环内,排气管进出口流量守恒性误差可以控制在0.053%以下。计算所得的各项发动机主要性能参数都和实验结果吻合得很好。

【Abstract】 In recent twenty years, with the rapid development of computer technology, computationalmathematics and computational fluid mechanics, one-dimensional cycle simulation program is playingan increasingly important role on engine design and performance improvement. Development cycle andcosts of new engine can be reduced greatly. Although commercial software is powerful, new model andalgorithm developed by users are difficult to be integrated into kernel of commercial software becausesource code is secret. Therefore, commercial software can’t support frontier research of internalcombustion engine. It is necessary to develop new model and new program for users. In this thesis,some new models about one-dimensional fluid flow are proposed and the design idea of software isalso presented. New general one-dimensional cycle simulation code with high precision andefficiency is developed.Directed graph in graph theory is used to define the engine thermal fluid network topology. Nodesin directed graph are used to describe all kinds of engine components, and directed edges are used toestablish network topological relationship among components. Using this definition, the engine systemwith arbitrary cylinders, intake and exhaust manifold can be modeled, and engine can also be equippedwith arbitrary series or parallel connected turbochargers. Graph theory is adopted to analyze enginefluid network. Intake and exhaust manifold can be accurately distinguished by depth-first algorithmcombining with features of engine fluid network. Numerical solving order from upstream todownstream can be automatically generated by topological sorting on components in directed graph.These work set a solid base for developing general one-dimensional cycle simulation code.The flow features in intake and exhaust manifold have great influence on overall engineperformance. In order to simulate one-dimensional unsteady flow in intake and exhaust manifoldaccurately, the finite volume method was adopted to divide each pipe into a number of control volumes.High resolution ENO scheme is used to calculate the flux of the internal face of control volumes inpipes. Boundary conditions at the ends of the pipes are obtained by the method of characteristics. Inthis way, the flow conservation can be guaranteed and calculation accuracy is improved. By using the theoretical solution of shock tube, it shows how ENO schemes with different orders affect simulationaccuracy and computational efficiency. The results indicate that simulation accuracy increases as orderof ENO scheme increases. The error between numerical and analytical solutions drops fastest as orderof ENO scheme increases from first-order to second-order. With the improvement of precision, errordrops slower and slower but the computation time grows exponentially. For real application, thesecond-order ENO scheme is recommended because it has high resolution and good computationalefficiency.In order to further improve the computational efficiency of one-dimensional unsteady flow,adaptive local time step marching algorithm is proposed. During calculation, each pipe in engine fluidnetwork has its own time step, which is no longer restricted by time step of other pipe. Inconsistenttime step leads to difficulties of dealing with boundary conditions. By adjusting solving order of pipesdynamically, the above problem is overcome. The practical calculation results indicate that new timestep marching algorithm has no effect on calculation precision and computing time can be reduced by25%-40%.Three way junction model has large impact on simulation precision of pressure wave and massflow in intake and exhaust manifold. Furthermore, overall engine performance can also be affected. Acold wind tunnel experiment with higher air velocity for normal junction in exhaust manifold has beencarried out. The variation law of total pressure loss coefficients is obtained. The measured totalpressure loss coefficients have some deviation from calculated results obtained by Vazsonyi equation,but they have the same trend with flow ratio. The higher the air velocity is, the more obvious thedeviation is. By improving Vazsonyi equation and introducing the distribution function, the correctedpressure loss model considering high air velocity is presented. The calculated pressure before turbineshows that the error of the new pressure loss model has been reduced by4.3%, compared with the oldmodel.On the basis of fluid network defined by directed graph, general framework of cycle simulationprogram is proposed. The efficient data structure is designed to implement the framework. By studyinglogical parallel regions in the serial program, it is the first time to transform the original serial solverinto parallel solver by using parallel algorithm in this article. The modified program is adopted tosimulate one-dimensional fluid flow and the whole engine performance on a multi-core computer. Theresults show that the simulation time with the parallel program can be reduced to one third of the serialprogram.A turbocharged D6114diesel engine test bed was established. The exhaust pressure waves were tested under the entire engine operation range. The simulation code developed by the thesis has beenused to simulate fluid flow in d6114. Under all operating conditions, the measured and calculatedresults were compared and analyzed. The error of simulated average pressure before turbine is under5%.When the engine speed is slow, the error becomes smaller. In one working cycle, the error of flowconservation between the inlet and outlet of exhaust manifold can be controlled under0.053%. Themajor engine performance parameters agree well with measured results.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络