Developing Novel Deep Learning Models for Automated Quality Inspection in Casting

Ensuring the attainment of high-quality casting products presents a formidable challenge within manufacturing industries, primarily due to the susceptibility of traditional manual inspection methods to human error and inefficiencies. This study endeavours to address this challenge by developing deep learning models tailored for automated quality inspection in casting operations. Specifically, the research entails formulating a novel custom convolutional neural network (CNN) model designed to detect and classify defects prevalent in casting products. The performance of the custom model is rigorously evaluated against established pre-trained architectures such as ResNet and EfficientNet. The evaluation encompasses a comprehensive case study utilizing an industrial dataset. Notably, the novel custom model demonstrates superior performance, attaining higher accuracy, F1, recall, and precision scores. The research offers valuable insights into the efficacy and applicability of diverse CNN models for real-world applications in casting quality inspection, garnered through extensive experimentation and analytical scrutiny. The findings thus contribute to advancing deep learning methodologies in optimizing inspection processes, curtailing costs, and augmenting overall efficiency within casting production contexts.