A new ensemble residual convolutional neural network for remaining useful life estimation

Remaining useful life (RUL) estimation is one of the most important component in prognostic health management (PHM) system in modern industry. It defined as the length from the current time to the end of the useful life. With the rapid development of the smart manufacturing, the data-driven RUL approaches have been widely investigated in both academic and engineering fields. Deep learning, which is a new paradigm in machine learning, has been applied in the RUL related fields, and has achieved remarkable results. However, classical deep learning algorithms also encounter the vanishing/exploding gradient problem found in artificial neural network with gradient-based learning methods and backpropagation. In this research, a new residual convolutional neural network (ResCNN) is proposed. ResCNN applies the residual block which skips several blocks of convolutional layers by using shortcut connections, and can help to overcome vanishing/exploding gradient problem. What's more, the ResCNN is enhanced by using the k-fold ensemble method. The proposed ensemble ResCNN is conducted on the C-MAPSS data provided by NASA. The results show that the proposed ensemble ResCNN has achieved significant improvement in both the mean and the standard deviation of the prediction RUL values. The proposed ensemble ResCNN has also compared with other famous machine learning and deep learning methods, including Multilayer Perceptron, Support Vector Machines, Deep Belief Networks, Long Short-Term Memory Model, Convolutional Neural Network and many other methods in literatures. The comparison results show that ensemble ResCNN achieved the start-of-the-art results, and outperform almost all of them.