Prognostics of damage growth in composite materials using machine learning techniques

Composite materials have been adopted and become critical in aerospace industry. However, due to the fatigue under continuous loading, the uncertain in structural integrity still remains an unsolved problem. The assessment of structural damage in composite laminates can be achieved by damage location, classification, and quantification. The growth trend of delamination area is one of the most important factors. In order to predict the delamination size efficiently and accurately, this paper proposes a prognostic method based on machine learning techniques. Prediction models, including linear model, support vector machines, and random forests were investigated. An optimal solution was identified by comparing the test results of different models. In this study, the length of the path across delamination area was selected as the objective value to train the models. The path length measurements augmented the training data sets and avoid the overfitting problem for the models. Moreover, the path length can be used to measure the size of delamination area. The interrogation frequency collected on several composite coupons was adopted as an input variable for the predict model. Experimental results demonstrate the effectiveness of the proposed method.