Tim Eschert’s new myth buster series on Machine Learning addresses the differences between Six Sigma, Big Data and Machine Learning and why it’s not yet widely adopted by manufacturing industry.
Industrial manufacturing has always been at the forefront of technological advances. Manufacturing companies were among the first to use supercomputers (outside of aerospace and academia), e.g. for finite element analysis. PLCs have essentially replaced fixed relais circuits and are today the heart of modern distributed process control systems.
However, when we look at Machine Learning as a technological advance, it has mostly been adopted by the finance and technology sectors. Companies in these industries have used Machine Learning to:
- translate complex texts between languages
- recommend items to purchase on e-commerce websites
- estimate the default risk of loan seekers
But the manufacturing sector hasn’t adopted machine learning with the same enthusiasm. (Yet.) One of the reasons is that we have always been using data in various ways within manufacturing, e.g. Data Analytics and Six Sigma. So what exactly is the difference between Data Analytics, Six Sigma, and Machine Learning - and what do they have in common? Ultimately, they are three different tools that are useful in different situations.
- Data Analytics is exploratory. It involves setting up an automated process of collecting data and then manually analyzing a dataset to search for patterns.
- Six Sigma relies on hand-picked values/parameters. It relies on a defined process, the “DMAIC cycle”. This cycle must be followed in a Six Sigma improvement project. The amount of data going into a Six Sigma problem is a lot smaller. Values are hand-picked in the “define” phase and then measured in the second phase. This means that a Six Sigma process is often biased by the human experts conducting the studies.
- What makes Machine Learning unique is a holistic, detached view on manufacturing. Machine Learning automatically recognizes hidden patterns in data. It is explanatory, which goes beyond the exploratory approach of data analytics. It is exhaustive, which goes beyond the limited and potentially biased approach of Six Sigma.
In conclusion, Machine Learning is just another screwdriver in our process optimisation toolkit. It does not necessarily want to replace the other tools - it is a screwdriver for a different screw. But maybe in the past, we have sometimes been using the wrong screwdriver for some of them.