Transfer learning based industrial steel plates fault diagnosis using industrial fault signals

Abstract

Transfer learning (TL) has shown its great advantage to solve small-training-sample issues utilizing information learned from existing large data with deep learning tech- niques. Transfer learning has been e ectively applied in many deep learning net- works where su cient training samples are not accessible; it still experiences es- sential problems for image processing. Image processing technology has become an interesting eld in medics as image processing plays avital role in the discovery of the diseases in the early stages, which facilitates the treatment of these diseases. Image processing divides into numerous scopes. For case, image classi cation, image seg- mentation, image enhancement and image assessment. In this thesis, we will review the existing industrial fault diagnosis models and will propose an image-based deep learning model to detect or predict industrial faults. In order to do that we will convert 1D sensor's fault signals to 2D images. After that, we will extract deep fea- tures using a deep learning model for training and testing the classi er. To validate our model, we will use an industrial fault dataset. As programming tools, we will use Python and MATLAB.