2020/11/20 by Tobias Schlagenhauf, Schlagenhauf, Tobias, Chenwei Sun +3
Engineering · #68T10 #Advanced Machining and Optimization Techniques #Advanced machining processes and optimization #FOS: Computer and information sciences #FOS: Electrical engineering #I.2.10 #Image and Video Processing (eess.IV) #Industrial Vision Systems and Defect Detection #Machine Learning (cs.LG) #Mineral Processing and Grinding #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2011.10235
openalex publication_date 2020/11/20 · openalex created_date 2022/10/01 · openalex updated_date 2026/07/28
The lack of reliable large datasets is one of the biggest difficulties of\nusing modern machine learning methods in the field of failure detection in the\nmanufacturing industry. In order to develop the function of failure\nclassification for ball screw surface, sufficient image data of surface\nfailures is necessary. When training a neural network model based on a small\ndataset, the trained model may lack the generalization ability and may perform\npoorly in practice. The main goal of this paper is to generate synthetic images\nbased on the generative adversarial network (GAN) to enlarge the image dataset\nof ball screw surface failures. Pitting failure and rust failure are two\npossible failure types on ball screw surface chosen in this paper to represent\nthe surface failure classes. The quality and diversity of generated images are\nevaluated afterwards using qualitative methods including expert observation,\nt-SNE visualization and the quantitative method of FID score. To verify whether\nthe GAN based generated images can increase failure classification performance,\nthe real image dataset was augmented and replaced by GAN based generated images\nto do the classification task. The authors successfully created GAN based\nimages of ball screw surface failures which showed positive effect on\nclassification test performance.\n