Thesis
Enhancing remanufacturing automation using deep learning approach
Downloadable Content
Download PDF- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2023
- Thesis identifier
- T17021
- Person Identifier (Local)
- 201661303
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- In recent years, remanufacturing has significant interest from researchers and practitioners to improve efficiency through maximum value recovery of products at end-of-life (EoL). It is a process of returning used products, known as EoL products, to as-new condition with matching or higher warranty than the new products. However, these remanufacturing processes are complex and time-consuming to implement manually, causing reduced productivity and posing dangers to personnel. These challenges require automating the various remanufacturing process stages to achieve higher throughput, reduced lead time, cost and environmental impact while maximising economic gains. Besides, as highlighted by various research groups, there is currently a shortage of adequate remanufacturing-specific technologies to achieve full automation. -- This research explores automating remanufacturing processes to improve competitiveness by analysing and developing deep learning-based models for automating different stages of the remanufacturing processes. Analysing deep learning algorithms represents a viable option to investigate and develop technologies with capabilities to overcome the outlined challenges. Deep learning involves using artificial neural networks to learn high-level abstractions in data. Deep learning (DL) models are inspired by human brains and have produced state-of-the-art results in pattern recognition, object detection and other applications. The research further investigates the empirical data of torque converter components recorded from a remanufacturing facility in Glasgow, UK, using the in-case and cross-case analysis to evaluate the remanufacturing inspection, sorting, and process control applications. -- Nevertheless, the developed algorithm helped capture, pre-process, train, deploy and evaluate the performance of the respective processes. The experimental evaluation of the in-case and cross-case analysis using model prediction accuracy, misclassification rate, and model loss highlights that the developed models achieved a high prediction accuracy of above 99.9% across the sorting, inspection and process control applications. Furthermore, a low model loss between 3x10-3 and 1.3x10-5 was obtained alongside a misclassification rate that lies between 0.01% to 0.08% across the three applications investigated, thereby highlighting the capability of the developed deep learning algorithms to perform the sorting, process control and inspection in remanufacturing. The results demonstrate the viability of adopting deep learning-based algorithms in automating remanufacturing processes, achieving safer and more efficient remanufacturing. -- Finally, this research is unique because it is the first to investigate using deep learning and qualitative torque-converter image data for modelling remanufacturing sorting, inspection and process control applications. It also delivers a custom computational model that has the potential to enhance remanufacturing automation when utilised. The findings and publications also benefit both academics and industrial practitioners. Furthermore, the model is easily adaptable to other remanufacturing applications with minor modifications to enhance process efficiency in today's workplaces.
- Advisor / supervisor
- Gachagan, Anthony
- Ijomah, Winifred
- Resource Type
- DOI
- Date Created
- 2022
Relations
Items
Thumbnail | Title | Date Uploaded | Visibility | Actions |
---|---|---|---|---|
PDF of thesis T17021 | 2024-07-25 | Public | Download |