Gas cyclones have numerous industrial applications. Typically, each cyclone has a dustbin to collect the trapped particles and the dimensions of the dustbin affect the cyclone performance. This paper aims to optimize the dustbin geometry via numerical simulations. The surrogate-based optimization approach has been applied in this study. The Latin-hyper cube sampling plan is used to generate thirty test cases. An artificial neural network with radial basis function has been used as a surrogate model trained by the CFD simulations. Here three design parameters (the dipleg length, the dustbin height, and the dustbin diameter) and two performance parameters, namely, the Euler number and the Stokes number are used. The fitted surrogate model shows that the variations of the dustbin geometry have a larger effect on the Stokes number than that on the Euler number. Both single-objective and bi-objective optimization studies are carried out using the artificial neural network. It is shown that the resulting optimum design of the dustbin and the dipleg leads to better performance than the conventional cyclones.

气旋具有许多工业应用。通常，每个旋风分离器都有一个垃圾桶，用于收集被捕集的颗粒，垃圾桶的尺寸会影响旋风分离器的性能。本文旨在通过数值模拟来优化垃圾箱的几何形状。基于代理的优化方法已在本研究中应用。Latin-hyper cube抽样计划用于生成三十个测试用例。具有径向基函数的人工神经网络已被用作通过CFD模拟训练的替代模型。这里使用了三个设计参数（料腿长度，垃圾箱高度和垃圾箱直径）和两个性能参数，即欧拉数和斯托克斯数。拟合的代理模型表明，垃圾箱几何形状的变化对斯托克斯数的影响大于对欧拉数的影响。单目标和双目标优化研究均使用人工神经网络进行。结果表明，与传统旋风分离器相比，垃圾箱和浸入式支腿的最佳设计可带来更好的性能。