Nur Najiha Kamarulzaman, Nor Salwa Damanhuri, Noor Azlina Mohd Salleh, Nor Azlan Othman, Belinda Chong Chiew Meng, Anik Nur Handayani
A defect classification system that aligns with the Fourth Industrial Revolution (IR 4.0) improves the inspection process since previous methods of inspecting additive manufacturing (AM) products heavily rely on operators' experience. However, the defects may appear similar to a no defect product because a small defect is difficult to differentiate. Hence, it is important to have a system that is capable in differentiating the features between defect and no-defect AM product. Thus, the aim of this study is to classify the AM product according to defect or no defect classes by using You Only Look Once version 8 (YOLOv8). A total of 1080 images of AM products from the Malaysia Automotive, Robotics & IoT Institute (MARii) are being used in this study. The features of the input images of the AM product are extracted when passing through the feature learning phase of YOLOv8. The input images are filtered with kernel, stride, and padding when they pass through the convolutional blocks. Then, the features are extracted, producing a feature map. These features are utilized at the classification phase for classification purposes. In this research, two types of activation functions are used at the feature learning phase which are SiLU and GELU. Results show that with training data at 70%, validating data at 20% and testing data at 10% with SiLU activation function, the defect classification system produced the highest accuracy of 93.39% within the 3 epochs used. A Graphical User Interface (GUI) has been successfully developed to classify the AM product into defect and no defect. Hence, this proposed classification system can be potentially applied to automatically classify other manufacturing products. © 2025 IEEE.
Universiti Teknologi MARA, Cawangan Pulau Pinang, Electrical Engineering Studies, Pulau, Pinang, Malaysia; Universiti Teknologi MARA, Faculty of Mechanical Engineering, Selangor, Shah Alam, Malaysia; Universitas Negeri Malang, Department of Electrical Engineering and Informatic, Malang, Indonesia
