Estimation of Remaining Useful Life for Turbofan Engine Based on Deep Learning Networks

Having accurate prediction on the health of machines in manufacturing can lead to a profitable organization if the operations and maintenance decisions are appropriately performed. This hinges on making well-informed operational and maintenance decisions. Incorporating condition monitoring and predictive maintenance strategies can significantly contribute to achieving this goal. By continuously monitoring the real-time condition of machines, organizations can gather valuable data that offers insights into the performance and health of the equipment. However, dealing with a scarce dataset, which is common in real world applications, makes any prognostics on the maintenance system intricate. This is further exacerbated by the unavailability of failure data within the system which makes degradation model is best suited for the said situation. Since there is no extensive study discussing computational time under similar settings of two different networks for the degradation model in estimating RUL, this study investigates a simple Long Short-Term Model (LSTM) method for prognostics, which is compared to a two-dimensional Convolutional Neural Network (CNN) under the same training options. The networks are trained using the popular Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset from the National Aeronautics and Space Administration (NASA). The aim of this study is to estimate the remaining useful life (RUL) of a turbofan engine in the most effective way. With carefully designed and defined network architectures, better performance can be attained, enabling proper foreseen of the RUL of an engine as soon as it is more likely to be close to failure. Based on the comparison, it is noted that the simple LSTM method for RUL prediction outperforms the two-dimensional CNN with better RUL prediction, Root Mean Square Error (RMSE), and computational time. For future improvement, this study can be further explored for a more sophisticated hybrid model that might produce better prediction in various sectors such as manufacturing, automotive, and military applications.