Natural Frequencies Prediction of Hybrid Composite Laminate based on Finite Element Simulation and Artificial Neural Network

• Published in: 2024 IEEE 22nd Student Conference on Research and Development, SCOReD 2024
• Main Author: 2-s2.0-85219506859
• Format: Conference paper
• Language: English
• Published: Institute of Electrical and Electronics Engineers Inc. 2024
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85219506859&doi=10.1109%2fSCOReD64708.2024.10872682&partnerID=40&md5=3a69d0cd56e887fef4a36c89279cb8b4

The analysis of structural vibrations is essential for comprehending failure mechanisms. Traditionally, physical tests have been the prevalent method; however, they are costly, time-consuming, and labour-intensive. Simulation is an alternative to improve the efficiency of vibration analysis, offering a more cost-effective, faster, and flexible approach. Through computational tools such as finite element analysis (FEA) and other numerical methods, engineers can simulate structural responses under various conditions without the need for extensive physical prototypes. Thus, prediction tools are becoming increasingly valuable in structural analysis. By training model, this tool can predict natural frequencies and mode shapes with high accuracy. This research assessed the natural frequency response of hybrid composite laminates during free vibration and established an accurate prediction model utilising Artificial Neural Networks (ANN) in MATLAB/Simulink. The model considered variations in plate thickness, volume fractions, and orientation angle, facilitating precise predictions of natural frequencies based on these parameters. Finite element models were constructed by using ANSYS APDL to accurately describe the natural frequencies of hybrid composite laminates made of Glass/Epoxy and Carbon/Epoxy under free vibration of an eight-layered configuration with angle orientation of [θ° /-θ° / θ° /-θ°]s. The study conducted 17 case studies generated by Design of Experiment, and the ANN result's prediction was compared to FEA results. The highest error between them was only 10.78 %. The prediction tool utilises an Artificial Neural Network (ANN) with a two-layer feed forward algorithm, and ten hidden layers, using Levenberg-Marquardt as the training algorithm. The ANN's adequacy in predicting natural frequencies was verified, with an R2 exceeding 0.99712 and an MSE of 35.396. © 2024 IEEE.