Research on Structural Health Monitoring of Composite Material Structure Based on Machine Learning Method

Abstract

With the wide application of composite materials in aerospace, construction, automobile, and other fields, the damage identification and structural health monitoring of composite materials have become a research hotspot. In this paper, based on machine learning theory, the damage of composite structure caused by tensile fatigue load is identified. The main research contents are as follows: obtaining the time domain characteristics of the material health state from the measured Lamb waves, which is better than the traditional time-frequency domain methods EMD, EEMD and LMD. The extracted features are selected, the importance of features is sorted, and the optimal feature set suitable for model training is selected. Four features with higher importance are used as the input of LightGBM training. Compared with the traditional machine learning methods SVM, RF and GBDT, LightGBM model has a better classification effect. LightGBM is used to realize the high efficiency and high accuracy diagnosis of composite health state, which provides a new method for non-destructive and on-line monitoring of composite fatigue damage. This paper’s test data comes from the joint injury test of NASA and Stanford University.