Remaining Useful Life (RUL) Prediction of Industrial Equipment Using a Hybrid Deep Learning Framework: Autoencoders, GANs, and Bayesian Ridge Regression

Industrial operations need extreme dependence on machinery reliability and efficiency not to experience unplanned equipment breakdown, thereby maximizing maintenance schedules. Since traditional approaches toward Remaining Useful Lifetime (RUL) estimation rely on prior knowledge and are incapable of dealing with the complexity and variability of an industrial environment, they often fall short. In this paper, a new hybrid model is proposed, where the integration of auto-encoders, Generative Adversarial Networks (GANs), and Bayesian Ridge Regression (BRR) is expected to realize high accuracy in RUL prediction. The auto-encoders will be applied to extract essential features from the high-dimensional sensor data in the approach. Whereas, GANs alleviate the scarcity of data by generating realistic synthetic data. BRR quantifies uncertainty through probabilistic RUL predictions. The hybrid model proposes mitigating the effect of the limitations of existing techniques to provide a more reliable and extended approach for accurate RUL estimation. This model was tested against the NASA C-MAPSS dataset and arrived at very promising results in terms of RUL prediction accuracy, attaining RMSE values of 32.10 for FD001, 34.20 for FD002, 54.32 for FD003, and 48.63 for FD004. These results show the power of the model to treat a large diversity of operating situations and fault mechanisms. In the paper, predictive maintenance procedures are proposed, which could improve dependability in industrial machinery.Funding: The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.