|
- 2018
Thermal Stress Control in Functionally Graded Plates with Artificial Neural NetworkKeywords: Functionally graded plates,artificial neural network,single layer model,Levenberg-Marquardt algorithm,finite difference method,thermal stress analysis. Abstract: In this study, trained models were obtained by using Artificial Neural Network (ANN) in order to determine the equivalent stress levels of one dimensional functionally graded rectangular plates. In this training set, a single layer sensor model was used according to our linear problem. With ANN, the models were trained by changing parameters the number of different iterations, number of neurons and learning algorithms. and the trained model was tested and its performance was measured. In our study, thermal stress analyses were performed for different compositional gradient exponents using finite difference method to constitute data sets. The data sets were constructed for the smallest value of the largest value of the equivalent stress levels, the greatest value of the greatest value of the equivalent stress levels, the greatest value of the smallest value of the equivalent stress levels, and the smallest value of the smallest value of the equivalent stress levels. Five different training algorithms were used in our training network: Levenberg-Marquardt, Back Propagation Algorithm, Momentum Coefficient Back Propagation Algorithm, Adaptive Back Propagation Algorithm and Momentive Adaptive Back Propagation Algorithm. The Levenberg-Marquardt algorithm is found to be more efficient than the other algorithms. With this study, trained models have been developed to provide time and job savings to determine equivalent stress levels in functionally graded plates, which are very important for high temperature applications. These educated models will provide important contributions to the literature and will be a source for the work to be done in this regard
|