Research On Flaw Characteristic Extraction Of UT Echo Signal And Neural Network Identification For Solid Propellant Of Rocket Engine

During the solid propellant of rocket engine manufacturing process, cracks, hollows, debondings, inclusions and other defects appear sometimes, which seriously affects the safety of solid rocket motor. Propellant ultrasonic testing is very important for the propellants'internal quality evaluation, and it is also significative to assure propellant security and reliability in the manufacture process and use process.This paper focus on solid propellant of rocket engine defects characteristics automatic extraction and identification technology, which is helpful for propellant automatic ultrasonic NDT and automatic defect identification. Propose a method of flaw classification which takes each spectrum's energy varieties to be the characteristics based on the wavelet packet transform. Take the energy varieties characters and the time-domain characters to be the inputs of the artificial neural network which is used for defect identification. This method achieved the requirement that solid propellant of rocket engines defects can not be omitted. At the same time, the average value of separability measure is as high as 97%, which indicates that this method is quite effective in the feature extraction of ultrasonic flaw echoes.In this paper, RBF neural network and improved BP neural network are used for training. The results have both reached good effect and accuracy. RBF neural network has a strong clustering and fault-tolerant, and the training speed of RBF network is faster than the BP network, so RBF neural network is better on the whole.