Potensi Jaringan Saraf Tiruan Backpropagation Dalam Memprediksi Respon Sistem Multi Degree Of Freedom Akibat Pembebanan Dinamik
Abstract
One of the simplification model of structure in the structural dynamic engineering was converting the model of a structure into the system that has mass, stiffness, damping percentage
and number of Degree of Freedom (DOF) whether in single number (Single Degree of Freedom) or multi number (Multi Degree of Freedom) as its components. Yet the application of model SDOF system was used as fundamental analysis and had to be developed for MDOF system. The research of MDOF system had become necessity to be continuously done in order to improve the previous existing analysis method. One of the method that potentially can be used was with Artificial Neural Network (ANN). Thus why this research was aimed to identify the capability of ANN in predicting the system responses. The analysis of system with 4, 6 and 8 DOFs that was subjected to dynamic excitation such as sinusoidal, triangular, rectangular and ramp load was done with Newmark-β method listing program of FORTRAN. Then analysis continued with Backpropagation Neural Network (BP-NN) using MATLAB program. The input data for BP-NN were heights (H), mass, stiffness, damping value, natural period (Tn) and dynamic load factor (DLF) with system responses as target data. The result had shown that variation of dynamic loads and system parameter had affected the value of system responses. While BP-NN training result showed its ability in predicting the system responses was decreasing from displacement to velocity and acceleration. It could be seen within the degradation value of regression (R) from 0.99-0.84, the increase of Mean Squared Error (MSE) from 1.19×10-7-0.7654 and error percentage from 5%-41%. Therefore ANN method was not capable to be used in predicting the responses of MDOF system under dynamic loads.
Keywords: MDOF system, dynamic load, Backpropagation Neural Network, Newmark-β, displacement, velocity, acceleration
and number of Degree of Freedom (DOF) whether in single number (Single Degree of Freedom) or multi number (Multi Degree of Freedom) as its components. Yet the application of model SDOF system was used as fundamental analysis and had to be developed for MDOF system. The research of MDOF system had become necessity to be continuously done in order to improve the previous existing analysis method. One of the method that potentially can be used was with Artificial Neural Network (ANN). Thus why this research was aimed to identify the capability of ANN in predicting the system responses. The analysis of system with 4, 6 and 8 DOFs that was subjected to dynamic excitation such as sinusoidal, triangular, rectangular and ramp load was done with Newmark-β method listing program of FORTRAN. Then analysis continued with Backpropagation Neural Network (BP-NN) using MATLAB program. The input data for BP-NN were heights (H), mass, stiffness, damping value, natural period (Tn) and dynamic load factor (DLF) with system responses as target data. The result had shown that variation of dynamic loads and system parameter had affected the value of system responses. While BP-NN training result showed its ability in predicting the system responses was decreasing from displacement to velocity and acceleration. It could be seen within the degradation value of regression (R) from 0.99-0.84, the increase of Mean Squared Error (MSE) from 1.19×10-7-0.7654 and error percentage from 5%-41%. Therefore ANN method was not capable to be used in predicting the responses of MDOF system under dynamic loads.
Keywords: MDOF system, dynamic load, Backpropagation Neural Network, Newmark-β, displacement, velocity, acceleration
Full Text:
PDFRefbacks
- There are currently no refbacks.