Fuel injector design has a substantial influence on the performance and emissions of direct injection engines. To date, large eddy simulations coupled with a single-fluid mixture modeling approach have shown great success in evaluating the complex interplay among injector design, fuel properties, and operating conditions on the injector performance. However, this simulation approach is too computationally expensive to be used by industry routinely for injector design due to the fine temporal and spatial resolution required to resolve wall-bounded flow within the injector. The work presented in this paper highlights a potential pathway to addressing this issue. To study the influence of injector design, fuel properties, and operating conditions on injector performance, large eddy simulations were performed to model the turbulent multiphase flow development through a side-oriented single-hole diesel injector. Using Latin hypercube sampling, the design space spanning a range of fuel properties, operating conditions, and needle lifts were explored. Two techniques for dimensionality reduction, namely proper orthogonal decomposition and autoencoders, were compared to evaluate their accuracy in representing the flow in a reduced dimensional space. Based on the findings from this work, recommendations are provided in using machine learning approaches within the context of emulation to construct reduced-order models for internal flow development relevant to automotive applications.

中文翻译：

---

单孔汽车喷油器中多相流的数据驱动模型简化

喷油器设计对直喷发动机的性能和排放有很大影响。迄今为止，大型涡流模拟与单流体混合物建模方法相结合已显示出在评估喷油器设计，燃油特性以及喷油器性能的工作条件之间复杂的相互作用方面取得了巨大的成功。但是，这种模拟方法的计算量太大，以致于无法解决工业上常规用于喷射器设计的问题，因为解决喷射器内壁面受限的流动需要精细的时间和空间分辨率。本文介绍的工作强调了解决此问题的潜在途径。为了研究喷油器设计，燃油特性和操作条件对喷油器性能的影响，进行了大型涡流模拟，以通过侧向单孔柴油喷射器对湍流多相流发展进行建模。使用拉丁文超立方体采样，探索了跨越一系列燃料特性，运行条件​​和针头提升的设计空间。比较了两种用于降维的技术，即适当的正交分解和自动编码器，以评估它们在表示降维空间中的流时的准确性。基于这项工作的发现，提出了在仿真环境下使用机器学习方法构建与汽车应用相关的内部流程开发的降阶模型的建议。探索了跨越一系列燃料特性，运行条件​​和针阀提升的设计空间。比较了两种降维技术，即适当的正交分解和自动编码器，以评估它们在降维空间中表示流的准确性。基于这项工作的发现，提出了在仿真环境下使用机器学习方法构建与汽车应用相关的内部流程开发的降阶模型的建议。探索了跨越一系列燃料特性，运行条件​​和针阀提升的设计空间。比较了两种降维技术，即适当的正交分解和自动编码器，以评估它们在降维空间中表示流的准确性。基于这项工作的发现，提出了在仿真环境下使用机器学习方法构建与汽车应用相关的内部流程开发的降阶模型的建议。