“…Specifically, three types of RNNs are studied: simple RNN, Gated Recurrent Unit (GRU), and LSTM, with GRU and LSTM obtaining a similar performance in all cases. Attending to multiclass problems, different LSTM configurations are studied in Nguyen and Medjaher (2019), Fernandes et al (2020), Zhang, Zhang, and Li (2019), Chen, Zhu, et al (2021) to obtain the probability of future failure from time‐series sorted operating records. Specifically, in Nguyen and Medjaher (2019), the LSTM predicts the failure probability in different time‐windows, while in Fernandes et al (2020), Zhang, Zhang, and Li (2019), Chen, Zhu, et al (2021), it is predicted the state of degradation of the system.…”