Statistical Machine Learning with Applications in Industrial Processes (in cooperation with Tata Steel)
Lead Research Organisation:
University of Warwick
Department Name: Statistics
Abstract
Recent innovations in machine learning has been largely driven by the emergence of Big Data that are correlated, high-dimensional, dynamic and often not in the Euclidean space. The PhD project is aimed at developing novel statistical machine learning methodology and
algorithms to address the modelling challenges that arise. The project is motivated by the need for building predictive models to understand steel manufacturing processes. Currently we are exploring possible applications of network science, looking for a new generative network model that incorporates covariate information.
algorithms to address the modelling challenges that arise. The project is motivated by the need for building predictive models to understand steel manufacturing processes. Currently we are exploring possible applications of network science, looking for a new generative network model that incorporates covariate information.
People |
ORCID iD |
Chenlei Leng (Primary Supervisor) | |
Stefan Stein (Student) |
Publications
Stein S
(2021)
A guided analytics tool for feature selection in steel manufacturing with an application to blast furnace top gas efficiency
in Computational Materials Science
Studentship Projects
Project Reference | Relationship | Related To | Start | End | Student Name |
---|---|---|---|---|---|
EP/R51214X/1 | 30/09/2017 | 02/08/2024 | |||
1935144 | Studentship | EP/R51214X/1 | 01/10/2017 | 08/12/2021 | Stefan Stein |
Description | In knowledge intensive industries such as steel manufacturing, application of data analytics to optimise process performance, requires effective knowledge transfer between domain experts and data scientists. This is often an inefficient path to follow, requiring much iteration whilst being suboptimal with regard to organisational knowledge capture for the long term. With the 'initial Guided Analytics for parameter Testing and controlband Extraction (iGATE)' tool we created a feature selection framework that finds influential process parameters and their optimal control bands and which can easily be made available to process operators in the form of guided analytics tool, while allowing them to modify the analysis according to their expertise. The method is embedded in a work flow whereby the extracted parameters and control bands are verified by the domain expert and a report of the analysis is automatically generated. The approach allows us to combine the power of suitable statistical analysis with process-expertise, whilst dramatically reducing the time needed for conducting the feature selection. We regard this application as a stepping stone to gain user confidence in advance of introduction of more autonomous analytics approaches. We present the statistical foundations of iGATE and illustrate its effectiveness in the form of a case study of Tata Steel blast furnace data. We have made the iGATE core functionality freely available in the igate package for the R programming language. |
Exploitation Route | The iGATE code base is modular, meaning that the method can easily be extended and adapted by industry research team to their personal preferences and needs. Also, its applicability readily extends beyond the immediate steel manufacturing context. |
Sectors | Manufacturing including Industrial Biotechology |
Description | We have created a novel guided analytics platform and framework for steel manufacturing processes that is actively being employed by my industry sponsor. |
First Year Of Impact | 2019 |
Sector | Manufacturing, including Industrial Biotechology |
Impact Types | Economic |
Title | iGATE: A guided analytics tool for feature selection in steel manufacturing |
Description | With the 'initial Guided Analytics for parameter Testing and controlband Extraction (iGATE)' tool we created a feature selection framework that finds influential process parameters and their optimal control bands and which can easily be made available to process operators in the form of guided analytics tool, while allowing them to modify the analysis according to their expertise. The method is embedded in a work flow whereby the extracted parameters and control bands are verified by the domain expert and a report of the analysis is automatically generated. The approach allows us to combine the power of suitable statistical analysis with process-expertise, whilst dramatically reducing the time needed for conducting the feature selection. We regard this application as a stepping stone to gain user confidence in advance of introduction of more autonomous analytics approaches. We have made the iGATE core functionality freely available in the igate package for the R programming language. |
Type Of Material | Data analysis technique |
Year Produced | 2020 |
Provided To Others? | Yes |
Impact | We have created a freely available R package with more than 7000 downloads to date. Please see our publication in Computational Materials Science for the full details of the methodology. |
URL | https://cran.r-project.org/web/packages/igate/index.html |