Unsupervised Remaining Useful Life Prediction through Long Range Health Index Estimation based on Encoders-Decoders

The prediction of the Remaining Useful Life (RUL) is a critical step in Prognostics and Health Management (PHM) systems under degradation. For efficient RUL predictions, most Artificial Intelligence (AI-based) methods perform direct mapping between raw sensor data input as output targets for supervised learning. However, majority real-life cases, available are either incomplete or unlabeled, which calls unsupervised methods. This paper proposes such an method. Firstly, this method uses autoencoder model to extract Virtual Index (VHI) from sensors readings. Secondly, LSTM-based (Long Short-Term Memory) encoder-decoder achieves VHI future predictions. Once exceeds pre-determined threshold, recursively inferred. Such thus allows obtain predictions without using RUL-labeled data. tested on C-MAPSS dataset. results obtained encouraging offer new perspectives real industrial applications