Hybridizing unsupervised clustering methods for in-cylinder vortex motion analysis under different swirl ratio conditions UM-SJTU JI - Shanghai Jiao Tong Univ

Format:

Book

Conference/Event

Author/Creator:

Zhao, Fengnian, author.

Contributor:

Fan, Weihan

Hung, David

Liu, Mengqi

Wu, Jiajin

Zhang, Junxiang

Conference Name:

SAE WCX Digital Summit (2021-04-13 : Live Online, Pennsylvania, United States)

Language:

English

Physical Description:

1 online resource cm

Place of Publication:

Warrendale, PA SAE International 2021

Summary:

Large-scale vortex motion could enhance the fuel-air mixing and the combustion stability inside a direct-injection engine. For in-cylinder vortex motion analysis, detection of the vortex zones is usually a challenging task because of the large cyclic variations of vortex structures, numbers and locations. In previous study, K-means clustering has been successfully applied to detect the vortex zones and quantify the cyclic variations. However, K-means algorithm is somewhat limited analyzing the complex flow with multiple vortex structures and vortex merging. Therefore, we propose a hybrid clustering method which is composed of K-means clustering and Gaussian mixture model (GMM) clustering to optimize the vortex classification performance under different flow conditions. In this study, in-cylinder flow fields with varying degrees of complexity were recorded by high-speed particle image velocimetry (PIV) under high and low swirl ratio conditions. For flow data during compression stroke, the flow data has a dominant vortex structure. The clustering results show that K-means algorithm can accurately classify the vortex locations into different zones with less computational time. During intake stroke, the in-cylinder flow is complicated with multi-scale vortex structures because of the strong intake air. For these flow fields with multiple vortices, GMM clustering outperforms K-means with a higher clustering accuracy especially for low swirl ratio condition. In summary, this hybrid clustering method blends K-means and GMM for vortex detection during compression stroke and intake stroke, respectively, to improve the clustering results. The temporal evolution and transient features of the in-cylinder vortex motion under different swirl ratio conditions can be accurately revealed by this hybrid clustering method

Notes:

Vendor supplied data

Publisher Number:

2021-01-0425

Access Restriction:

Restricted for use by site license