A deep residual sequence autoencoder for future state estimation and aiding prognostics and diagnostics in machines: a case study of mechanical rolling elements

Prognostics and health management (PHM) enables the prediction of faults by condition monitoring and estimation of the future state of engineering systems. By using a predefined end of life in conjunction with the time to start predicting (TSP), remaining useful life (RUL) is estimated. Due to the n...

Celý popis

Uložené v:

Podrobná bibliografia
Vydané v:	Neural computing & applications Ročník 37; číslo 17; s. 10737 - 10756
Hlavní autori:	Ramadhan, Bwambale Rashid, Cahit, Perkgoz
Médium:	Journal Article
Jazyk:	English
Vydavateľské údaje:	London Springer London 01.06.2025 Springer Nature B.V
Predmet:	Accuracy Artificial Intelligence Case studies Classification Computational Biology/Bioinformatics Computational Science and Engineering Computer Science Condition monitoring Data Mining and Knowledge Discovery Deep learning Engineering Image Processing and Computer Vision Preventive maintenance Probability and Statistics in Computer Science Recurrent neural networks S.I.: Timely Advances of Deep Learning with applications and Data Driven Modeling Special Issue on Timely Advances of Deep Learning with applications and Data Driven Modeling State estimation Useful life Time to start predicting LSTM Remaining useful life Prognostics and health management Deep residual sequence-to-sequence model Pearson’s correlation coefficient
ISSN:	0941-0643, 1433-3058
On-line prístup:	Získať plný text
Tagy:	Pridať tag Žiadne tagy, Buďte prvý, kto otaguje tento záznam!

Popis
Shrnutí:	Prognostics and health management (PHM) enables the prediction of faults by condition monitoring and estimation of the future state of engineering systems. By using a predefined end of life in conjunction with the time to start predicting (TSP), remaining useful life (RUL) is estimated. Due to the nature of recurrent neural networks (RNNs) and their variants, they have been used extensively in timeseries-related problems, including PHM. However, the existing impediments known in RNNs that cause nonconvergence during training have created setbacks in their application in fields with data characterized by long dependent series. To overcome these setbacks, models are built shallow, notwithstanding the fact that the depth of a model significantly contributes to the increase in results accuracy. Keeping these problems in mind, this study proposes a deep residual autoencoder comprising LSTM neurons and applies it in PHM to predict the future state of mechanical rolling elements. The study also proposed an objective and scientific technique for detecting TSP contrary to available studies that subjectively perform this task. The model proposed was trained and validated with bearing vibration data available in public repositories. To achieve the desired results, data were normalized using minmax normalization and transformed to positive values, it was then converted into a shape compatible to the model proposed. The model was fed with the latest window after TSP was detected to predict the future state, aiding the prediction of RUL. Finally, the model was evaluated with cumulative relative accuracy as the main evaluation metric together with root mean squared error and mean absolute error; they all demonstrated that the proposed model exhibits reliable results in PHM. Graphical abstract
Bibliografia:	ObjectType-Case Study-2 SourceType-Scholarly Journals-1 content type line 14 ObjectType-Feature-4 ObjectType-Report-1 ObjectType-Article-3
ISSN:	0941-0643 1433-3058
DOI:	10.1007/s00521-024-10756-4