Abstract
Machine maintenance is a major part of total operating costs for manufacturing and production plants, hence, having a cost-effective maintenance strategy is vital to success. Most standard solutions operate on a time-driven approach where repairs or replacements are carried out after an allotted amount of time. This results in healthy machines being replaced unnecessarily incurring losses in time, resource, and parts. Predictive maintenance utilises sensors capable of collecting real-time operating data and analytical techniques to predict when a machine will require maintenance.This research, in partnership with TATA Steel, focuses on the development of a low-cost condition monitoring tool capable of detecting variations in machine behaviour, known as operating conditions, to provide deeper insight into the monitored machines. Additionally, the condition monitoring tool should be flexible in design, capable of monitoring a variety of machines in the same manner, without changing the design structure. This research conducts three different case studies, verifying the developed generalised condition monitoring tool known as AutoCM. AutoCM is validated by testing a number of machines. The mold oscillator is one such machine, which is responsible for supporting the mold in its ability to cool the molten steel. Additionally, AutoCM is used to monitor the behaviour of electric motors, used to power motors that drive steel into coilers.
In order to achieve this, an Auto-Encoding Convolutional Neural Network takes processed Short Time Fourier Transform images as input and reduces their dimensionality into a smaller set of features. A Self-Organising Map then fits a two dimensional plane to the feature-set, where the images can be separated into clusters. The proposed case studies show promising results and are able to discriminate between varying operating conditions. The proposed system will allow TATA Steel to identify to the current and historical operating conditions for each machine, with further developments allowing for user interaction through a web-based application known as AutoCM Live.
The benefits of this research being TATA Steel have a low-cost generalised monitoring solution that can be used in conjunction with existing machine-specific data to provide context to the identified machine behaviour patterns.
| Date of Award | 2024 |
|---|---|
| Original language | English |
| Sponsors | KESS 2 PhD Student, University of South Wales |
| Supervisor | Andrew Ware (Supervisor), Penny Holborn (Supervisor) & CK Tan (Supervisor) |