Session: CIE-24-01 - AMS-CAPPD-SEIKEM: Artificial Intelligence and Machine Learning in Design and Manufacturing
Paper Number: 90084
90084 - Synthetic Image Assisted Deep Learning Framework for Detecting Defects During Composite Sheet Layup
Automation of high-performance manufacturing processes such as the prepreg composite layup has been gaining a lot of interest lately. Automation of such processes require accurate defect detection methods to maintain the desired quality. Composite prepreg layup involves manipulation of a carbon fiber sheet, for such deformable components traditional machine vision based defect detection techniques do not work. Advanced defect detection techniques enabled by deep learning are the key for such applications. However, Deep learning usually requires enormous amount of physical images of the process which is infeasible in high-mix manufacturing applications. In this paper, we present an approach where we use finite element based simulation for emulating the defects accurately. We employ advanced graphics techniques to generate realistic synthetic images of the actual defects from the simulated 3D meshes. Approximately, 10000 synthetic images are generated and combined with around 1000 images of real sheets to train a ResNeSt based deep learning model. We present an efficient 2-stage methodology for training the deep learning network. With the proposed modelling methods and data augmentation techniques our method can achieve a mean Average Precision (mAP) of 98% on actual production data for detecting defects. The code and the entire dataset are available at: https://github.com/omeym/Physics-Informed-Deep-Learning-Based-Defect-Detection.git.
Presenting Author: Omey Manyar University of Southern California
Presenting Author Biography: Ph.D. Student at the University of Southern California
Authors:
Omey Manyar University of Southern CaliforniaJunyan Cheng University of Southern California
Reuben Levine University of Southern California
Vihan Krishnan University of Southern California
Jernej Barbic University of Southern California
Satyandra Gupta University of Southern California
Synthetic Image Assisted Deep Learning Framework for Detecting Defects During Composite Sheet Layup
Paper Type
Technical Paper Publication