Research And Implementation Of Classification Of Hypersonic Inlet Start/Unstart

Hypersonic inlet is one of the key components of the scramjet engine,which usually occurs unstart because of the backpressure,the internal contraction ratio(ICR)and the Mach number,etc.The unstart phenomenon is the most frequent issue of working hypersonic inlet,which causes the exception of force and moment and leads to the change of the vehicle performance.Therefore,the recognition and control of hypersonic inlet unstart is a problem which needs to be solved urgently.Aiming at the problem of inlet unstart caused by backpressure,the unstart phenomenon and reasons of the inlet unstart was explored,and the recognition method of the inlet unstart was researched.Firstly,Numerical method was used to simulate the inlet unstart phenomenon caused by backpressure disturbance.A 2D hypersonic inlet / isolator model was simulated in different freestream conditions.The internal flow field structure of hypersonic inlet in the influence of backpressure was analyzed,the two kinds of working mode was defined by the location of shock train and the surface pressures on inner runner.Then the wall static pressures were recorded to produce the training dataset.Secondly,because the inlet unstart can be discriminated by the signal of change surface pressure,and the classification of inlet start/unstart is a binary classification problem,so the pattern recognition was used to solve the inlet classification.First of all,two optimal pressure points were selected by the feature selection algorithm.The improved Support Vector Machine-Recursive Feature Elimination was proved better than the original SVMRFE,and the combined algorithms which were based on the Relief and improved SVMRFE or correlation coefficient with high efficiency and high accuracy were proved.Then the support vector machine algorithm was used to train the classification plane,the fisher linear discriminant and logistic regression algorithm was used to validate the result.Finally,all experimental result shows that the feature selection performance was improved,the optimal classification criterion has strong robust performance and generalization performance.And the change of pressures in two optimal pressure points and changes of flow characteristics conform to the actual physical law,so the result shows that the criterion is valid.