Back to search

University of Birmingham

Deep learning for remaining useful life prediction in a remanufacturing system

Abstract

dc:description.abstract

This thesis investigates Remaining Useful Life (RUL) prediction, which is central to Predictive Maintenance (PdM), as part of efforts to address uncertainties related to the return of products to a remanufacturing system. By accurately predicting the remaining useful life of critical components in a product, uncertainties associated with its return for remanufacturing are reduced. Remanufacturing planners can determine the optimal timing for product retrieval and establish well-structured schedules for subsequent processes. The central focus of the research is utilising Deep Learning (DL) algorithms to predict the remaining useful life of components, a critical factor in determining the ideal timing for sending a product to the remanufacturing process. DL techniques, namely, Long Short-Term Memory (LSTM), Gated Recurrent Units (GRUs), Convolutional Neural Networks (CNNs) combined with LSTM, and CNNs combined with GRUs, have been selected for prediction of the RUL of bearings in rotating machinery. Additionally, a swarm-based optimisation tool, the Bees Algorithm (BA), and one its latest variants, the Two-Parameter Bees Algorithm (BA2), have been introduced to optimise DL learnable parameters and tune DL hyperparameters, respectively. The performance of these predictive models was evaluated through comprehensive benchmarking utilising prediction scores. Three studies were conducted. In the first study, the integrated CNN with LSTM model attained a maximum score of 0.49, with the possible values ranging from 0.0 to 1.0, where the optimal value is 1.0. However, in the second study, which offers an automated alternative to manual hyperparameter tuning using BA2, the best score achieved by the CNN-GRU model was 0.48. The results of the final study, which incorporates the Adaptive Moment (ADAM) optimisation algorithm into the BA, demonstrate that the highest score obtained by the three models - LSTM, CNN with LSTM, and CNN with GRU - was 0.42. Thus, this study revealed that integrating BA with ADAM optimisation leads to reduced efficiency and inferior outcomes across all the models compared to previous studies. Nevertheless, the study underscores the superiority of combined DL models compared to individual DL models. This thesis has shown that DL models can be developed for predicting the remaining useful life of critical components, such as bearings in rotating machinery, and that the best DL for the bearing RUL prediction problem is a combination of CNN and LSTM. The three studies conducted in this research have also demonstrated the possibility of the Bees Algorithm as a tool for optimising DL models.

Degree

thesis:*
Name dc:type.qualificationname
d_ph
Level dc:type.qualificationlevel
d_ph
Grantor dc:publisher.institution
University of Birmingham
Year dc:date.issued
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Theinnoi, Nathinee

Subjects

dc:subject × 1

Rights

Language dc:language.dcterms_iso639_2
eng

Chain of custody

source
Harvested from
University of Birmingham
Base URL
etheses.bham.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Theinnoi, Nathinee. Deep learning for remaining useful life prediction in a remanufacturing system. d_ph thesis, University of Birmingham, 2024.