Response surface analysis, clustering, and random forest regression of pressure in suddenly expanded high-speed aerodynamic flows

Abstract

Experimental analysis of base pressure in suddenly expanded compressible flow from nozzles at different Mach numbers is performed. Intensive experimentation is carried out to investigate the base pressure and wall pressure of flow expanding from the nozzles into the enlarged duct. Microjets to actively control the flow are adopted to increase the base pressure. Experiments were conducted for Mach numbers (one sonic and rest supersonic) from 1 to 3, nozzle pressure ratio (NPR) from 3 to 11. The duct length considered from 10 to 1, and the area ratios tested were from 2.56 to 6.25 are the variables whose effect on base and wall pressure is studied using response surface methodology. The K-means algorithm performs a clustering analysis of this enormous data, which provides useful information and patterns. Regression of both the pressures using a random forest classification algorithm is carried out. The response surface analysis reveals that microjets are efficient when the flow is under the influence of a favorable pressure gradient. The base pressure reduces from maximum to minimum when the flow regime changes from over to correct expansion by increasing the NPR. Lower area ratio and higher duct length have a minimum effect on base pressure. The wall pressure flow field is unaffected due to the presence of the microjets. K-means clustering revealed that a high percentage of base pressure is in the lower range. This necessitates the importance of increasing the base pressure to reduce the base drag. Random forest algorithm has proved to be a handy tool for predicting base pressure and wall pressure and similar highly non-linear data.