Development of a novel natural frequencies prediction tool for laminated composite plates using integrated artificial neural network (ANN) - simulink MATLAB / Mohd Arif Mat Norman

Vibration analysis of structures is crucial for understanding failure mechanisms. Typically, physical tests have been the common approach, but they are expensive, timeconsuming, and labor-intensive. Finite Element (FE) programming is an alternative, but it requires strong theoretical and mathematical knowledge, along with programming skills. Analytical methods are also viable but become complex with composite materials, requiring extensive mathematical computation. Thus, there is a need for an accurate yet user-friendly natural frequencies prediction tool. This study aims to develop a novel MATLAB®/Simulink® program called AJ Natural Frequency Predictor (AJNatFreP) based on the Classical Laminate Plate Theory (CLPT) integrated with Artificial Neural Networks (ANN). The program provides an efficient means to calculate natural frequencies for laminated composite plates, and it includes a userfriendly Graphical User Interface (GUI). The program is trained to predict natural frequencies with acceptable accuracy, utilizing various input configurations. The model's accuracy was assessed through comparisons with 3D elasticity solutions, finite element simulations, and other published literature references. The study conducted 16 case studies, and the program's predictions were validated using analytical methods and compared to FE modal results. The highest error observed between the FE modal and AJNatFreP was only 2.15%. Convergence analysis and numerical verification were performed to establish accurate FE models for free vibration analysis of laminated composite plates. The study explored various factors influencing natural frequencies, including aspect ratio, principal moduli ratio, anti-symmetry plies, volume fiber fractions, and skew angle of laminated composite plates. 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 value exceeding 0.9999 and an MSE of 107.527. Several case studies evaluated the performance of the prediction tool, showing good agreement with other lamination theories, such as FSDT (less than 3%) and HSDT (less than 7%). Comparison with experimental results demonstrated errors of less than 5%. A paired t-test confirmed the significant improvement of the prediction tool compared to other laminated composite theories, with the t-test value exceeding the 95% confidence statistical t-value. In conclusion, this study introduces a novel, user-friendly prediction tool (AJNatFreP) that accurately calculates natural frequencies of laminated composite plates. This tool offers a valuable contribution to the field, aligning with the goal of simplifying methods while producing accurate results in the context of free vibrations of laminated composite plates.