engleski

Aluminum microstructure inspection using deep learning: a conventional neural network approach toward secondary dendrite arm spacing determination

The present research investigates the determination of secondary dendrite arm spacing (SDAS) using convolutional neural networks (CNN). The goal is to create a deep learning model for the SDAS prediction with industrially acceptable tolerance. SDAS is predicted from the image taken from the polished sample of EN AC 46000 AlSi9Cu3(Fe) cast aluminum alloy. The Sequential CNN model from the Keras library was trained using Python software. Additionally, image preprocessing methods were used to simplify the training dataset from a full RGB color scale to a black and white color scale. A relatively simple CNN structure could predict different SDAS values on a single cross-section with very high accuracy, an R2score of 99, 2%. However, on an EN AC-42000 AlSi7Mg material cross-section sample which was not used during training, CNN had some lower performances, but still inside the practically acceptable range. Furthermore, a CNN approach towards SDAS determination could be used with industrially acceptable tolerance.

Secondary dendrite arm spacing ; Convolutional neural network ; Casting microstructure inspection ; Deep learning ; Aluminum alloys

nije evidentirano