Classification of surface roughness for milled A6061 aluminum alloy based on depth map models with convolutional neural networks

Milling is a traditional machining method and is crucial in the manufacturing industry. It allows shaping and refining mechanical details, from flat surfaces to complex shapes. To assess the quality of a component processed using milling, determining the surface roughness is extremely important and also the easiest way to evaluate its quality. In this study, a classification model of surface roughness using a convolutional neural network (CNN) with 16 different features extracted from surfaces created through milling was developed. The dataset was generated based on images with various features, which were then divided into training, validation, and testing sets. Utilizing support from optical microscopy with different magnifications (50×, 100×, and 200×), image capturing was performed at a size of 1600 pixels. The data collection process, image processing techniques, especially depth map technique for analysis, were presented. The accuracy in classifying surface roughness was evaluated with the highest accuracy achieved at 50 × magnification (97.05%). However, at larger magnifications, the performance decreased, with an accuracy rate of 95.62% at 100 × magnification and 86.36% at 200 × magnification. Results from various experiments demonstrated the effectiveness and potential of applying deep learning methods in detecting and classifying surface roughness features of machined blanks through milling processes.