Endüstri 4.0 için derin öğrenme yaklaşımlarıyla zaman serisi analizi

Nowadays, the number of devices connected to the Internet has increased considerably, and this had led to the emergence of the term Internet of Things. Thanks to the Internet of things, devices can now decide and share information between them. Industry 4.0, which has recently become very popular, is a term used for data exchange and automation for production technologies. The manufacturing industry produces a large amount of data due to the many sensors of the devices. These data can be used to improve processes and product quality. For example, by analyzing these data, anomalies that may occur in the lines can be found in advance, the life expectancy of the devices in the factory can be calculated, and as a result, active preventive maintenance can be provided for these devices. This study aims to find the remaining useful life of the components by performing time series analysis with deep learning methods. There are many studies on the estimation of the remaining useful life of the components in the literature. In these studies, various methodologies such as The number of devices connected to the Internet has increased considerably, and the term Internet of Things is emerged. Thanks to the IoT, devices can now decide and share information between them. Industry 4.0 is a term used for data exchange and automation for production technologies. The manufacturing industry produces a large amount of data which can be used to improve processes and product quality. We aim to find the remaining useful life of the components by performing time series analysis with deep learning methods. There are many studies on the estimation of the remaining useful life of the components and they use various methodologies such as artificial neural networks, signal processing and machine learning techniques. The quality of the dataset has great importance in the creation of estimation models. We examined 4 PHM datasets of NASA and we selected Femto-ST Bearing dataset. Because of high-frequency noise in dataset; we applied Discrete Wavelet Transform on the data. Many features were extracted from this data set and health indicator values were calculated. We construct the remaining useful life prediction model, particularly using long-short-term memory neural network. We did various experiments to find the right LSTM parameters and to analyze how the prediction model works with different train and test data. Furthermore, we compared the results of our model with the results of previous studies on this dataset. Our results are underperformed compared to other studies. This dataset is not directly applicable to the LSTM network, and preprocessing needs high effort.