Machine-learning- based classifiers for the steel industry

What’s the economic added value of machine-learning in the manufacturing sector?

The use of big data analytics techniques and machine learning boosts competitiveness in the EU manufacturing sector.

PROTEUS’ mission is to investigate and develop ready-to-use scalable online machine learning algorithms and real-time interactive visual analytics to deal with extremely large data sets and data streams.

Using these data sets within a steelmaking industrial use case, PROTEUS aims to reduce defects in the steel coil production of ArcelorMittal, a large EU steel manufacturer, using predictive machine learning algorithms and a range of interactive visualisation tools.

In doing so, PROTEUS demonstrates a two-fold function for the research developed:

  • strategically, it bridges the gap in open-data analytics between the EU and the US;
  • economically, it paves the way for the adoption of big data analytics techniques in the EU steel industry, which accounts for the 11% of the global output, thus increasing its competitiveness.

Machine learning classifiers for the steel industry and the manufacturing sector

As part of the PROTEUS project, Trilateral Research is developing the project key performance indicators (KPIs) and benchmarks considering PROTEUS-driven benefits to the steel industry specifically and, more broadly, the EU’s big data capabilities.  KPIs are a set of variables measuring the performances of the PROTEUS tool components, namely data process, visualisation and online machine learning. Benchmarks are reference values against which PROTEUS’ performances are compared.

One of the main KPIs developed by Trilateral’s technical team determines the added value that a predictive classifier, derived from a Machine Learning model, brings about to the steel coil production.

In current production processes, all coils manufactured are deemed as high-quality and of high-value. Nonetheless, because some of the coils produced will be defective, by considering them high-quality and marketing them as such, manufacturers are likely to waste resources in this product’s mishandling (defective coils are still profitable but have a much lower value).

Machine-learning- based classifiers would allow to detect defective coils prior to their mishandling, saving manufacturers resources and time as well as improving the quality of their work.

The creation of added value depends not only on the performance of the classifier itself, i.e. how good it is at detecting defective products (the classifier will detect defective coils as well as high-quality coils erroneously classified as defective), but on producers’ costs and returns.

Hence, a machine-learning- based classifier adds value to a manufacturer when the ratio of A) coils erroneously classified as defective ( false negatives) and B) truly defective coils (true negatives) is lower than the ratio of C) the cost of coil mishandling (treating and marketing a coil as high quality even though it is defective) and D) the value difference between high- quality and  defective coils(how higher is the value of a high-quality coil compared to a defective one).

 Machine learning classifiers for the steel industry

This KPI can be used by manufacturers outside the steel industry as well: those who manufacture products which may be defective but can be still marketed can refer to it.  Additionally, this KPI helps decision making on the adoption of smart-manufacturing. It may not be worth it for manufacturers with relatively low production costs to install sensors and use predictive machine learning tools, such as PROTEUS, to help production. Conversely, smart-manufacturing may be profitable for manufacturers having high production costs.

Through this measurement of the economic added value of machine-learning in the manufacturing sector, TRI aims to support small and medium enterprises (SMEs) characterised by large variations in production costs. Doing so may have promising growth implications as SMEs accounted for 58% of the total employment and 42% of the total added value in the EU28 manufacturing sector in 2016 (European Commission 2017). 

 

Please contact our team for more information:

Graham Hesketh, Data Scientist at Trilateral Research

Graham Hesketh

Giuseppe Maio, Research Assistant at Trilateral Research

Giuseppe Maio


‘Risk Assessment Report and Methodology’

You can view the Executive Summary and Table of contents of the Project Solebay Risk Assessment Methodology Report.

Please sign up to the Solebay mailing list to download the Full Solebay project report.