Most manufacturers still are not effectively making sense of the data they gather
Published on : Monday 03-05-2021
Berk Birand, Chief Executive Officer, Fero Labs.
How effectively is the industry dealing with the data deluge?
In the past few years, the industry has gotten better at storing and managing large quantities of data (‘Big Data’). However, most manufacturers still aren't effectively making sense of the data they gather. As a result, they're missing out on strategic insights that could help them save money and make production more efficient.
We see lots of companies still using tools like Excel and traditional statistics packages like Minitab to analyse data. Classical stats is not intended for large, complex data sets. These tools are also not intended to be deployed alongside real-time control systems. In contrast, newer approaches like machine learning are specifically designed for big, complex data sets, like what's produced in a factory.
What is the extent of scope for improving productivity within existing resources with better data management?
Manufacturers can see dramatic productivity improvements simply by using better tools for analysing large data sets, such as machine learning.
For example, one of our customers used Fero software to optimise chemical formulas so they could minimise the amount of raw materials they needed to achieve their desired result. As a result, the company saw a 9% reduction in raw material costs. Another customer in the chemicals space achieved an 11% increase in yield.
There are countless vendors today offering machine learning solutions to industry. What is the USP of Fero Labs?
Many vendors try to blindly apply ML systems designed for the tech world to the industrial world. However, what works for Google or Facebook doesn't work in a factory. It's critical for industrial ML models to be explainable and transparent – in other words, for process engineers and domain experts to understand how the models work, be able to adjust them using their domain expertise, and most critically, be confident in the algorithm's predictions. At Fero, explainability and transparency are a core part of our technology. We've heard from many customers that this sets us apart from other ‘AI’ companies they've worked with in the past.
What are the different ways in which Fero Labs helps companies optimise their production?
Traditionally, engineers must balance competing priorities including cost, quality, and emissions reduction. But using Fero, they can easily build sophisticated ML models that help them optimise for all these objectives at the same time--without any data science experience.
Using these models, companies can build a virtual factory (i.e., digital twin) and use it to efficiently optimise production, from running plant tests to testing what-if scenarios. They can learn what formulas and processes work best and implement those, without wasting valuable production time and resources.
How does it perform on the Return of Investment yardstick?
Extremely high. One of the metrics that we promote with industrial executives is that of the ‘model RoI’. Basically, even with Excel, you can achieve some RoI. But ML has a higher RoI, and explainable ML has an even higher one.
You can get the highest RoI, however, through explainable ML that's deployed in real time. When the software is connected to the plant, the models are able to update themselves based on live, real-time data, allowing for the most accurate predictions and analysis.
Left to itself, will the industry follow the sustainability path without an effective regulatory mechanism?
I believe regulations and technology can work hand in hand. While regulations set up the goalposts, new technology like ML helps manufacturers meet those goals without compromising their other objectives.
If companies see tangible proof that improving sustainability can drive cost savings and efficiency improvements, they'll be more likely to follow the sustainability path moving forward.
Berk Birand is the founder and CEO of Fero Labs, the only industrial process optimisation software that uses Explainable Machine Learning to help factories reduce emissions, optimise quality and increase profits. He holds a PhD in electrical engineering and computer science from Columbia University. His academic research focused on optimising wireless and optical networks with efficient cross-layer algorithms. He holds a patent in IoT systems for resilient fibre-optic networks.