To achieve early detection and prediction of high-frequency unstable combustion in model rocket engines,a data-driven prediction framework is established.This framework extracts recurrence matrices for nonlinear feature extraction of combustion noise,and utilizes deep learning models for training and prediction.Based on the dynamic pressure signal measured from the hot-fire tests of a single-injectors rocket combustor,the prediction of combustion instability is carried out,which can predict the occurrence of combustion instability approximately 35 ms in advance.Altogether 25 groups of dynamic pressure datasets are used,including experiments with different combustion chamber geometries.The results of cross validation upon the prediction framework show that the prediction accuracy of the framework was above 95%,which indicates the effectiveness and robustness of the prediction framework.
维普
万方数据
