| Summary: | 碩士 === 國立中央大學 === 資訊工程學系 === 106 === Influenced by the revolutionary concept of German Industry 4.0, major manufacturing industries have been moving from automatic production into smart production for maintaining their competitiveness. Industry 4.0 advocates smart factories that use Internet-enabled assembly lines to collect large amounts of data and then through data analysis to achieve the goals of smartly adjusting production processes, intelligently saving energy, precisely forecasting customer demands, and accurately predicting mechanical failures. In general, smart factories can yield flexible and even customized products in the most efficient way. This paper focuses on estimating machine remaining useful life (RUL), which is a kind of the machine condition pre-diagnosis. By accurate RUL estimation, we can perform predictive maintenance, instead of preventive maintenance, to avoid sudden breakdown of machines/components during the operation process to prevent huge loss.
This paper proposes a Recurrent Neural Network (RNN) deep learning method to estimate the remaining useful life of machines, especially the aero-propulsion engines. The proposed method uses the Long Short-Term Memory (LSTM) model with the attention-based (AB) mechanism. The LSTM model is useful for extracting relationship between time-series data items that are far separated, and the AB mechanism can help emphasize different factors that affect the RUL in different time. The NASA C-MAPSS (Commercial Modular Aero-Propulsion System Simulation) dataset is taken to evaluate the URL estimation accuracy of the propose method. The evaluated results are compared with those of related methods, namely the MLP, SVR, RVR, CNN, Stack LSTM methods. Comparisons show that the proposed method is superior to the others in terms of the scoring function value defined by the C-MAPSS dataset, and the Root Mean Squared Error (RMSE) .
|
|---|
