Deep Learning-based Automatic 3D Printer Anomaly Detection during the Printing Process

Abstract

3D printing is a technology which is expected to be one of the most used technologies in the upcoming time. This technology allows to print out products that are designed using 3D modeling software. Though this is an effective technology, it also has its disadvantages. The disadvantages include anomalies. Anomaly is a defect that is often found when the printer finishes the printing process. Thus, it cannot be rectified when found during the process. To resolve this issue, this study aims in developing a deep learning model using the UNet algorithm. A dataset of pictures of various possible anomalies is gathered from Kaggle. The Kaggle data is then preprocessed using three different methods. The images are initially applied using the target format. The images are then multiplied and shrunk to keep the balance. The UNet method is employed to create a deep learning model. The preprocessed dataset is then used to train the model. To guarantee improved performance, the trained model is subsequently put to the test to assess the model’s final accuracy and loss. In all three instances, the model’s output is determined to be satisfactory. The model produced an accuracy of 98% during the testing and produced a loss value of 0.54%. This loss value is so small that it can be neglected. The model developed is found to be one of the best algorithms that can be used in anomaly detection.