Abstract:In this study, a novel method for detecting the growth of delamination in sandwich structures has been proposed. To this end, we suggested hybridizing the Deep Learning techniques (DL) and Finite Element Method (FEM) for predicting the growth of delamination in this structures. A dataset of simulated delamination growth under different delamination sizes has been produced using the FEM method. Then, a DL model has been trained using this dataset to precisely predict the growth of delamination. This study focused on predicting delamination growth using a tuned and optimized deep learning based regressor. Therefore, to find the ideal set of hyperparameters, the Bayesian optimization algorithm has been used for selecting the best structure and enhancing the regressor performance. Afterward, the model was evaluated and multiple processes were conducted to improve its behavior and solve its stability and overfitting issues. Particularly, an inconsistency between validation loss and training loss has been initially detected in the behaviour of the model, which may indicate overfitting. To tackle this issue, dropout regularization has been added, which improved the consistency between the loss functions but results in less smooth convergence from the expectations. So, in a third study, dropout and L1 regularization has been combined to improve the stability of the model. This combination achieved a consistent and smooth convergence between the validation and training loss functions. The findings highlight the importance of hyperparameter optimization and regularization techniques in improving regression model performance. The study shows the efficiency of Bayesian optimization in hyperparameter tuning and the iterative optimization of a regression model. Furthermore, the outcomes show that the suggested method can identify and predict delamination growth with high accuracy.

A Spatiotemporal Prediction Model for Rapid Prediction of Delamination Growth in Open-Hole Composite Laminates

Prediction Model of Residual Load-Bearing Capacity of Composite Laminates Using Deep Learning

A Deep Learning Method for Damage Prognostics of Fiber-Reinforced Composite Laminates Using Acoustic Emission

Dynamic Data Driven Load-Carrying Capacity Prediction Method for Composite Laminates with Delamination

Predicting delamination of composite laminates using an imaging approach

Predicting delamination in composite laminates through semi-analytical dynamic analysis and vibration-based quantitative assessment

Data-Driven Approaches for Characterization of Delamination Damage in Composite Materials

A novel version of hierarchical genetic algorithm and its application for hyperparameters optimization in CNN models for structural delamination identification

Modeling delamination growth in laminated composites

Numerical Prediction of Damage Interaction during L-Shaped Laminated Composite Structures Unfolding

A highly efficient prediction of delamination migration in laminated composites using the extended cohesive damage model

Deep learning-based prediction of delamination growth in composite structures: Bayesian optimization and hyperparameter refinement

Delamination prediction in composite laminates under low-velocity impact

Prediction of delamination growth in laminated composites using acoustic emission and Cohesive Zone Modeling techniques

A novel method for predicting delamination of carbon fiber reinforced plastic (CFRP) based on multi-sensor data

A machine learning-enabled prediction of damage properties for fiber-reinforced polymer composites under out-of-plane loading

Artificial neural network based delamination prediction in laminated composites

In-situ Damage Mechanism Investigation and a Prediction Model for Delamination with Fibre Bridging in Composites

An efficient surrogate model for damage forecasting of composite laminates based on deep learning

Progressive failure prediction of three-dimensional woven composites using a generic multi-scale analytical model

Prediction of Delamination Growth in Laminated Structures Loaded by Quasi-static and Cyclic Loads