Data science in manufacturing: Unlocking digital characteristics (DC) of metal forming processes
File(s)
Author(s)
Liu, Heli
Type
Thesis
Abstract
As one of the most important manufacturing crafts, the metal forming industry accounts for 15-20% of the national GDP of modern industrialised nations. Over 160,000 engineering materials and nearly 90% (wt. %) of products made of metals are manufactured by metal forming processes. Approximately 90% of steel products4 and 67% of aluminium products5 are manufactured by at least one metal forming process. Meanwhile, voluminous data, proliferated daily at an ever-greater scale during production, research, and development activities, have driven scientific advances in the manufacturing sector towards the era of digital manufacturing. These data are essential to unlock insightful features that can facilitate comprehensive scientific understanding of metal forming technologies. To date, there has been limited research in approaching metal forming from the perspective of data, not to mention that most collected data lack critical information to fully characterise a metal forming process. This is mainly due to limitations of data collection capabilities and data privacy, which hinders the extraction of insightful digital information from underlying scientific patterns.
In this thesis, a cloud-based data repository was established that currently contains 201 functional data modules of diverse manufacturing processes. Each data module consists of experimentally verified manufacturing metadata and essential information with a globally unique and permanent ID which provides open access for scientific communities. Tackling the long-standing challenges on processing information absent data requires an emerging research field combining knowledge of metal forming and data science. Here, a physical-based Evolutionary Binary (EB) algorithm was developed for enabling information absent data processing. Results demonstrate that the EB algorithm yields a highly efficient recognition of missing features for a metal formed product with nearly 95% of overall accuracy with sparsely labelled data points (≤1%)...
In this thesis, a cloud-based data repository was established that currently contains 201 functional data modules of diverse manufacturing processes. Each data module consists of experimentally verified manufacturing metadata and essential information with a globally unique and permanent ID which provides open access for scientific communities. Tackling the long-standing challenges on processing information absent data requires an emerging research field combining knowledge of metal forming and data science. Here, a physical-based Evolutionary Binary (EB) algorithm was developed for enabling information absent data processing. Results demonstrate that the EB algorithm yields a highly efficient recognition of missing features for a metal formed product with nearly 95% of overall accuracy with sparsely labelled data points (≤1%)...
Version
Open Access
Date Issued
2023-07-27
Date Awarded
01/12/2023
Advisor
Wang, Liliang
Publisher Department
Mechanical Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)