Machine learning approach for delamination detection with feature missing and noise polluted vibration characteristics

We have developed a machine learning (ML) approach that can precisely detect the delamination of laminated composites using the noise polluted and feature missing vibration characteristics. Here, the input features of ML model are the first ten natural vibration frequencies of laminates with certain delamination, and the outputs are the delamination parameters. In order to improve the prediction precision for the noise polluted input, the principal component analysis (PCA) method is introduced to transform the ten frequencies into the principal components. Then the alternative principal components are used to the training and prediction procedure. For the missing feature problems, a similar ML model is developed to predict the missing frequency, and then the predicted value is used for the delamination detection. The proposed ML approach shows excellent performance in assessing the delamination of laminated composites with prediction error less than 10% even for 10% noise of the input signals.