Deep Learning for Detecting Manufacturing Defects Using Convolutional Neural Networks

This research paper applies Convolutional Neural Networks (CNN) and supervised learning algorithms to predict faults in a casting manufacturing dataset. The goal is to create an efficient deep learning model balancing accuracy and trainable parameters. Data preparation includes loading, resizing images, data augmentation, and splitting the dataset. Baseline performance is established using Logistic Regression, Decision Trees, and Gradient Boosting. The study implements CNN models, aiming to find the simplest model with high accuracy. CNNs outperform traditional models, with a VGG-like model showing high accuracy. A CNN with 4 layers achieves higher accuracy than models with fewer layers. The paper proposes a final model with three convolutional layers, balancing accuracy, and training time. The study recommends a VGG-like model for the casting manufacturing fault detection problem, emphasizing the balance between model complexity and performance.