Lessons from the Manufacturing Industry

Using the right analytics tools is essential in getting value from today’s data. This article outlines some ways in which artificial intelligence and machine learning can help turn data, in this case manufacturing data, into business insight. It also offers tips and inspiration for those wanting to learn more about advanced analytics and what it can do.


There’s a lot of discussion around artificial intelligence (AI) and machine learning (ML) and the ways in which advanced analytics can enhance business performance and outcomes. However, exactly how you use it and which outcomes you aim for depends a lot on the culture of your organization and the industry you work in.

Let’s take manufacturing as an example. There’s a lot of data flying around a typical auto production line or steel foundry but capturing and making sense of it all can be challenging.

Many manufacturers carry out business intelligence (BI) reporting using spreadsheets to answer specific, backwards-looking questions such as ‘how many units did we make last quarter?’. While there’s data available to answer more complex questions and even predict future outcomes, it is difficult to integrate this data so that it can be analyzed.

Existing approaches like Six Sigma are well adapted for handling small sets of structured data and independent projects, but lack the real-time, predictive capabilities that AI and ML can offer.

Machine learning in manufacturing

Committed to helping manufacturers address this challenge, Fero Labs has developed an ML solution to automate and enhance decision making. The company has developed a set of algorithms that capture the full spectrum of structured and unstructured data and identify idiosyncrasies that can predict potential issues and enable employees to take preventative action. The solution also indicates how probable a predicted outcome is, enabling workers to make informed, confident decisions more quickly

“Our aim was to use the last fifteen years of machine learning development to help augment the familiar Six Sigma processes,” explains Berk Birand, CEO of Fero Labs. “For example, a typical Six Sigma workflow would involve producing say, several tons of steel, then changing some parameters and producing a few tons more, then comparing the two batches to see which was most effective. This wastes a lot of time and money. With machine learning though, you can build models to predict the results of experiments before you run them physically, so you can make your decisions before committing resources.”

Solving real-life challenges

This type of predictive analytics can be applied to almost any industry, adding value from small-scale tasks to entire production chains. Take an oil refinery, for example. It may produce a lot of hot oil. As the oil is moved around the site, it leaves a residue in the pipes, which builds up over time. When this happens, the oil doesn’t heat as well and so more energy is needed to bring it up to the right temperature: bad news for the environment, and the refinery’s power bill. However, without cutting into the pipe and disrupting production, it’s hard to know where the build-ups are occurring. Using sensor data and ML algorithms like Fero Labs’, the refinery can predict where and at what rate this residue will build up, and so plan its maintenance schedule to keep the pipes clean.

Getting started

With cloud-based solutions like Fero Labs’, manufacturers can implement more advanced analytics capabilities into their production environments fairly quickly,  using existing hardware infrastructure.

Fero works with its customers to run a proof of concept using a single set of data first, helping to demonstrate ROI up front and train the algorithms to work smoothly with the organization’s available data. Once this is done, manufacturers can migrate their real-time production data to a secure cloud environment. Here, Fero Labs’ algorithms, running on the Intel® Xeon® processor E5 family in a Microsoft Azure* environment, process the data and deliver predictions, along with interpretable explanations of how they were made, to decision makers through a dashboard.

“We’re sometimes asked why we chose to base our solution on CPUs, rather than GPUs,” says Birand. “GPUs are very well adapted for doing a lot of small operations, like simple computations on large numbers of values. Our algorithms have a lot of cool benefits like being able to provide confidence with the predictions but this means that they are more complicated. As a result, they are much better suited to run on CPUs. It’s about transparency into why the algorithm made the decision. While GPU-based deep learning capabilities are very powerful for certain kinds of fuzzy outcomes, they’re not as appropriate when you need levels of control and understanding.”

What’s next?

Implementing AI and ML to address specific production line challenges like some of those described here is a good first step for manufacturers on the advanced analytics journey. For those that have already seen success in such distinct projects, the next goal could be applying and integrating AI across the supply chain or multiple factories. For example, a single piece of steel will go from the mill where it is produced by an auto manufacturer and eventually becomes part of a car on a showroom floor. Tracking that entire process as one journey can help identify possibilities for improvements that can impact many stages in the supply chain.

And of course, these types of predictive analytics algorithms can make a huge impact beyond manufacturing. For example, in healthcare, they could track patient behavior and influence when they are away from the hospital to help predict an individual’s care needs and pre-empt a hospital visit.

You can learn more about how to get started with advanced analytics in your own organization from this planning guide or find out about Fero Labs’ solution for manufacturing in this solution brief.