Effective Latent Representation for Prediction of Remaining Useful Life

Abstract

AI approaches have been introduced to predict the remaining useful life (RUL) of a machine in modern industrial areas. To apply them well, challenges regarding the high dimension of the data space and noisy data should be met to improve model efficiency and accuracy. In this study, we propose an end-to-end model, termed ACB, for RUL predictions; it combines an autoencoder, convolutional neural network (CNN), and bidirectional long short-term memory. A new penalized root mean square error loss function is included to avoid an overestimation of the RUL. With the CNN-based autoencoder, a high-dimensional data space can be mapped into a lower-dimensional latent space, and the noisy data can be greatly reduced. We compared ACB with five state-of-the-art models on the Commercial Modular Aero-Propulsion System Simulation dataset. Our model achieved the lowest score value on all four sub-datasets. The robustness of our model to noise is also supported by the experiments.