Classification and Inspection of Debonding Defects in Solid Rocket Motor Shells Using Machine Learning Algorithms

Abstract

Debonding problems along the propellant/liner/insulation interface are a critical factor affecting the integrity of solid rocket motors and one of the major causes of their structural failure. Due to the complexity of interface debonding detection and its low accuracy, a method of wavelet packet transform (WPT) combined with machine learning is proposed. In this research, multi-layer structure specimens were prepared to simulate the structure of a solid rocket motor. First, ultrasonic non-destructive testing technology was used to obtain defect data. Then, WPT algorithm was employed to extract characteristic signals of the defect data. Moreover, k-nearest neighbor model, Random Forest model and support vector machine model were applied to the classification. The results showed that the accuracies of the three models were 84.67%, 90.66% and 95.33%, respectively. Positive results indicate that WPT with machine learning model exhibited excellent classification performance. Therefore, WPT combined with machine learning can achieve a precise classification of debonding defects and has the potential to assist or even automate the debonding inspection process of solid rocket motors.