Track: Manufacturing and Design

Abstract

For years, manufacturing industry faced challenges in conserving high cost and efficiency in research and development of products especially in relation to welding structure. To handle this problem, this research conducts an effort to develop methods to determine tensile strength properties of Friction Stir Welding (FSW) in Aluminum alloy using prediction model. An artificial intelligent technique, i.e. Artificial Neural Network (ANNs) is used to develop the prediction model. Initial work involves FSW welding experiment and testing of subsequent result to analyze the tensile strength properties of the welding structure based on process parameter input. Prediction model is developed based on Back Propagation (BP) of error to predict tensile strength of the FSW structure. Input parameters for the model is tool rotational speed and travel speed while output of the model is the tensile strength of FSW welded structures. Proposed prediction model is then trained with experiment data. Testing of the proposed prediction model was then conducted using unused experiment data from the training. Research result shows that proposed prediction model is aligned with experiment data. This shows that average error value from the training and testing is 0.010286 or very small (close to zero) which means that the desired output of the prediction model for the training and testing is close to each other. Result from regression graph for training and testing shows matching linear regression between output and target when compared to dash line of the ideal result. This shows an absolute linear relationship where R equals to 0.9 or close to 1. This proves good compatibility the prediction model.