How to reduce cost with machine learning: case study

Industrial Manufacturing Falling Behind

When we look at the buzzwords of the Industry 4.0 world, we tend to think of large data centers and modern computer workstations. These are at odds with the realities of manufacturing — especially so in the steel industry. A "modern" steel factory can easily be several decades old. It's dirty, hot, and loud, and doesn't seem to be "4.0" at all at first glance.

But if we look a little bit closer, we find that even in this factory, large amounts of data accumulate on a regular basis. Temperatures and pressures are recorded, chemical compositions are controlled, flows at various points are measured. Steel plants track thousands of variables that affect the quality of their end product.

The tech and finance sector have made a lot of progress in the area of machine learning and artificial intelligence. With machine learning, computers translate complex scientific texts between languages, win in board games against human world champions, and estimate the default risk of loan seekers. However, the same level of adoption is not there with industrial manufacturing. Why is that?

Optimising for Customisation

In this case study, let’s imagine that we are running a hot strip steel mill. Our customer defines the properties of yield and tensile strength, as well as the desired elongation of the product. When it comes to the chemical composition, however, the specifications leave some wiggle room. Alloy costs are a large operational cost item for any mill, especially so for the one that focuses on specialty products. For us, this means that we should try to optimise the exact chemical composition of the steel so that it does meet our customer’s specifications at minimal cost. So even though our benchmark plant is already doing an excellent job at optimising the process and quality using commonly used techniques, maybe there is more that could be done with the mill’s data.

Looking at machine learning solutions, we would expect a software to be able to process all the data, without human bias, in an automatic way. It would not only take those parameters into account that we have been taking into account for decades but instead, identify which parameters influence our production and in which way we should manipulate them.

We would expect it to automatically identify the important patterns. In would allow us to easily find out how each of our parameters has an influence on our KPI. In our case, how alloys influence the mechanical targets yield and tensile strength. We would then expect it to present its results transparently such that our expert engineers, as well as factory floor workers, can act accordingly.

Machine learning is poised to change the way we think about planning, design, maintenance, waste, and energy. The goal of this post is to present a concrete solution from the steel industry. However, use cases lie in manifold verticals and challenges.

Are you taking advantage of the data you’re sitting on right now?

Let Tim know any challenges you’re facing and he’ll be sure to help – here contact Tim on the network.