Casting Product Image Data for Quality Inspection with Xception and Data Augmentation

Abstract

Casting defects encompass a broad spectrum of imperfections, such as blow holes, pinholes, burrs, shrinkage defects, and various metallurgical anomalies. Detecting these defects manually requires a trained eye, and even the most diligent inspectors can inadvertently overlook subtle irregularities. To address these challenges, there is a growing movement toward automation in quality control. Deep learning models, including the Xception model, are being harnessed to create a robust classification system. Such models have the capacity to analyze thousands of product images with precision, identifying defects that may elude human inspectors. Furthermore, data augmentation techniques are applied to enhance the dataset, allowing the model to generalize more effectively and improve its defect recognition capabilities.