Locating delamination in a composite laminate using machine learning and recurrent deep neural networks based on vibration response

Abstract

In this article, different types of machine learning (ML) methods have been used to detect the delamination location in laminated composite plates. A finite element model is developed to extract the first ten natural vibration frequencies of composite laminates with various delamination locations as input for training the ML models. Support vector machine (SVM), Gaussian process regression (GPR), tree-based and deep neural network methods are utilized in this study. The main novelty of this research is to present a multilayer perception and gated recurrent unit (MLP-GRU) hybrid model that learns in several stages and detects the delamination location. The proposed hybrid network shows higher accuracy for identifying the delamination in laminated composites than classical ML methods, such that the value of the root mean squared error (RMSE) with this method is 0.0959 and the coefficient of determination (R-squared) is 0.8084. © 2024 Institution of Structural Engineers