Soft-sensor development for milling processes guided by discrete element method Modelling

Download files
Access & Terms of Use
open access
Copyright: McElroy, Luke Patrick
Altmetric
Abstract
This work establishes a new approach 10 soft-sensor development for monitoring of variables for which there is no means of measurement. The study of unmeasurable variables is commonly performed with lengthy computer simulations, using methods such as the discrete element method (OEM) of modelling, which is used to simulate many particulate systems. In this thesis, rotating and stirred drums are simulated over a wide variety of operating conditions using DEM modelling, Microscopic scale data are obtained from the simulations and useful characteristics of particle interaction and motion determined for each operating condition. These characteristics are partide flow regime and impact intensity, which are very difficult or impossible 10 obtain experimentally. Particle-wall (p-w) impact event data are also collected from the simulations and processed to reflect the real collection of surface vibrations with accelerometers mounted on a drum shell. This data processing is informed by knowledge that the majority of surface vibrations measured by an accelerometer mounted on a drum shell are caused by p-w impact events adjacent to the accelerometer, and knowledge that these surface vibration signals can be decomposed to identify individual p-w impact events. Soft-sensors are then trained to predict partide flow regime and impact intensity using input variables extracted from p-w impact data. The soft-sensor structures are given by the multivariate methods of Fisher discriminant analysis (FDA) and principal component regression (peR), which are used for qualitative prediction of flow regime and quantitative prediction of impact intensity respectively. Using new input data the soft-sensors are found to satisfactorily predict the variables of interest, thus proving the concept of a soft-sensor that infers unmeasurable OEM data from surface vibration data. This work provides an approach to online monitoring of important process variables that are generally studied offline with lengthy DEM simulations.
Persistent link to this record
Link to Publisher Version
Link to Open Access Version
Additional Link
Author(s)
McElroy, Luke Patrick
Supervisor(s)
Creator(s)
Editor(s)
Translator(s)
Curator(s)
Designer(s)
Arranger(s)
Composer(s)
Recordist(s)
Conference Proceedings Editor(s)
Other Contributor(s)
Corporate/Industry Contributor(s)
Publication Year
2010
Resource Type
Thesis
Degree Type
PhD Doctorate
UNSW Faculty
Files
download Mcelroy-014954915.pdf 9.57 MB Adobe Portable Document Format
Related dataset(s)