An Artificial Neural Network Approach to Predict Strain Gauge Results of Unidirectional Laminated Composites' Tensile Test

In this research, the artificial neural network (ANN) approach was investigated for predicting strain gauge results of unidirectional laminated composites' tensile tests. This approach involves training an ANN with a dataset of known strain gauge readings and their corresponding tensile test results. The required data to train the network was generated by using 15 different tensile test data created by MTS series 322 test frame. Strain values of MTS device were used as an input in ANN formation to estimate strain gauge results. The dataset was rearranged by applying normalization and linearization processes. Strain results were predicted approximately above 99% accuracy. In conclusion, a highly trained ANN system is a reasonable approach to approximate strain gauge results from MTS device test results. As a future goal the well-trained ANN system can be the option for obtaining materials stress-strain curves without testing by using machine learning and deep learning algorithms.

I. Introduction

Material science in aviation and space technologies needs time and complex test processes for demanding material standards in the sector. Tensile test-a destructive testing method-need specimens for every circle of testing, and it requires special tools like strain gauges for examining results. These types of instruments need measurement sensitivity and expertise in usage. Applications with the high accuracy is used instead of mechanical designs.