چكيده به لاتين
Considering advances in industries and industrial automation, we can simply figure out that humankind is always looking for a way to make tasks run automatically and eliminate the need for doing some tasks manually. In this thesis, we will study one of the most used elements of the industry, and it is rotary machines. According to the statistics, bearings are one of the most probable parts of the rotary machines to become broken and defected. Accordingly, we advised a way to deal with this problem properly. The proposed method is to use deep neural networks to detect the occurred fault in the rotary machine. This field of study started approximately in 2015 after the great advances in computational capabilities of CPUs and GPUs and availability of the open source libraries. The structure of this thesis is as follows: in the first chapter of the thesis, we will review recent papers on fault detection of rotary machines. Then, in the second chapter, since deep learning is an infant field of study and its been around since 2015 we prepared an introductory information regarding the various structure of Deep Neural networks and optimization methods. In the third chapter, information regarding laboratory set-up is provided and all the details are explained, though the reader of this thesis can easily understand how to deal with the dataset. In the fourth chapter, we will elaborate on algorithms and structures, namely: ANN, DNN, Deep SAE, Deep SSAE, Deep SDAE, and Deep SSDAE and algorithm has been explained step by step. In the last chapter, we will analyze the results and enumerate the pros and cons of each structure and algorithm.
