Prediksi Frekuensi Alami Struktur Kapal Trimaran Menggunakan Jaringan Saraf Tiruan

Putra, Dio Mukti Wibowo Kusno (2026) Prediksi Frekuensi Alami Struktur Kapal Trimaran Menggunakan Jaringan Saraf Tiruan. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 04111940000058-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Prediksi frekuensi alami merupakan aspek penting dalam analisis getaran struktur kapal karena berkaitan dengan keselamatan struktur dan pencegahan resonansi. Penelitian ini bertujuan mengembangkan model prediksi frekuensi alami menggunakan Artificial Neural Network (ANN) berdasarkan parameter respons modal berupa Structure ID, displacement, velocity, dan acceleration. Dataset diperoleh dari hasil analisis modal penelitian sebelumnya, kemudian diperluas melalui pembentukan data sintetis dan augmentasi Gaussian noise. Model dikembangkan menggunakan arsitektur Feedforward Neural Network dengan algoritma Bayesian Regularization Backpropagation (trainbr). Evaluasi dilakukan menggunakan koefisien determinasi (R²), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Five-Fold Cross Validation, serta perbandingan dengan regresi linear. Model ANN menghasilkan R² sebesar 0,9334, RMSE sebesar 1,5926 Hz, MAE sebesar 1,1824 Hz, dan MAPE sebesar 2,9652%. Hasil Five-Fold Cross Validation menunjukkan Mean R² sebesar 0,9232 dan Mean RMSE sebesar 1,7291 Hz, sedangkan regresi linear menghasilkan R² sebesar 0,7915, RMSE sebesar 2,8186 Hz, MAE sebesar 2,2251 Hz, dan MAPE sebesar 5,7146%. Hasil penelitian menunjukkan bahwa ANN lebih akurat dibandingkan regresi linear dalam memprediksi frekuensi alami. Namun, model yang dikembangkan masih bergantung pada hasil analisis modal sehingga belum dapat menggantikan analisis modal berbasis metode elemen hingga.
===========================================================================================================================
Prediction of natural frequency is an important aspect of ship structural vibration analysis because it is closely related to structural safety and resonance prevention. This study aims to develop a natural frequency prediction model using an Artificial Neural Network (ANN) based on modal response parameters consisting of Structure ID, displacement, velocity, and acceleration. The dataset was obtained from previous modal analysis results and expanded through synthetic data generation and Gaussian noise augmentation. A Feedforward Neural Network trained using the Bayesian Regularization Backpropagation (trainbr) algorithm was employed. Model performance was evaluated using the coefficient of determination (R²), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Five-Fold Cross Validation, and comparison with a linear regression model. The ANN achieved an R² of 0.9334, RMSE of 1.5926 Hz, MAE of 1.1824 Hz, and MAPE of 2.9652%. Five-Fold Cross Validation produced a mean R² of 0.9232 and a mean RMSE of 1.7291 Hz, indicating consistent predictive performance. Compared with linear regression (R² = 0.7915, RMSE = 2.8186 Hz, MAE = 2.2251 Hz, MAPE = 5.7146%), the ANN demonstrated superior prediction accuracy. The developed model is effective as a data-driven approach for predicting natural frequencies within the scope of the dataset used. However, it cannot replace finite element-based modal analysis because its input variables are derived from modal analysis results.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Artificial Neural Network, Bayesian Regularization, Frekuensi Alami, Struktur Kapal, Analisis Modal.
Subjects:	V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering > VM311 Catamarans.
Divisions:	Faculty of Marine Technology (MARTECH) > Marine Engineering
Depositing User:	Dio Mukti Wibowo Kusno Putra
Date Deposited:	04 Aug 2026 08:37
Last Modified:	04 Aug 2026 08:37
URI:	http://repository.its.ac.id/id/eprint/143581

Actions (login required)

View Item