Defect Detection of Casting Products Using Convolutional Neural Network

Quality control of industrial products requires a precise and accurate process to provide high-quality products. However, in massive production plants, the control process is time-consuming, and the automated process is vital. Artificial Intelligence (AI) and deep learning models are capable of learning small and ambiguous details in the data and provide fast and accurate classification. Therefore, artificial intelligence and deep learning have gained importance in solving real-life problems and supporting the industry in the last decade. The study implements a light Convolutional Neural Network (CNN) model to detect casting product defects without pooling operation. The model is trained and tested using casting product images, and the evaluation is performed using the sensitivity, specificity, balanced accuracy, and F1 Score. The implemented CNN model achieved 99.11%, 99.61%, 99.45%, and 99.36% sensitivity, specificity, F1 Score, and Balanced Accuracy on the testing set. The obtained results suggest the use of lighter deep models in defect detection by reducing the computation cost and are capable of detecting defects accurately.