Mukund Subramaniyan is a recent Ph.D. graduate from the Division of Production Systems, Department of Industrial and Materials Science at Chalmers University.
Mukund works at the intersection and forefront of AI and manufacturing. His research has been published in leading academic-industrial engineering journals and conferences. His research outcomes have also been implemented by leading automotive manufacturing companies in Sweden and Germany. His Ph.D. was advised by Dr. Anders Skoogh and Dr. Johan Stahre at Chalmers University, where he also received his Masters in Production Engineering.
Congratulations on your dissertation from Chalmers. The title is ”“Data-driven throughput bottleneck analysis in production systems.
What was one (or two) of your big “aha” moment when doing research and writing this?
Thank you very much for your wishes!
There were two aha moments in my research:
- We all talk about big data. It is a word that is trending in the industry as well as in research. Everyone believes that big data means big value and can lead to big decisions. Though this might be true, I found the opposite of this common belief in my research. Small data created more value in analyzing bottlenecks in manufacturing than big data. It was an aha moment not only for me but for manufacturing companies and the research community.
- Researchers go behind in achieving 100% accuracy of AI. Manufacturing companies do not appreciate the value of AI if it is not 100% accurate. I was also having the same belief before I started my Ph.D. studies. During my Ph.D., I found something interesting. AI can add tremendous value in bottleneck analysis if its accuracy beats the naïve accuracy. Let me give you an example. In manufacturing, engineers must predict the upcoming bottlenecks in the production system. Today, engineers make this prediction in a naïve way. They assume that today's bottlenecks will be the bottlenecks of tomorrow. In my research, I found this to be true only 70% of the instances. The AI that I developed to predict the bottlenecks had an accuracy of 80%. Now compare the difference. AI accuracy is more than a naïve approach accuracy! In other words, AI is more valuable than a naïve approach to improve practice, even though it is not 100% accurate. Explaining the comparison raised not only my academic colleagues' eyebrows but also engineers from Swedish manufacturing companies. Though the comparison is simple, the academic community and the manufacturing companies did not think that way before. It was an aha moment!
How can our members learn from your dissertation? What can you tell them?
I have created a new and fast AI that can save many millions of Swedish Krona by removing bottlenecks in factories. Today Swedish manufacturing companies' productivity is alarmingly low at 50%. They need to improve the productivity significantly to remain profitable and competitive in the future. Removing bottlenecks will help companies to increase productivity. And my AI can search for bottlenecks, analyze them, and give an action list that suggests how to remove the bottleneck throughputs. All these happen in less than a second. To make a simple analogy, you may say that this tool is like a GPS that tells the car driver how to avoid blind alleys. Research partners find my AI very useful. Some have already implemented this tool. For example, an automotive company in Germany implemented AI to identify short-term throughput bottlenecks. Now, they can identify the bottlenecks in real-time by sitting in their office rooms. A car manufacturer in Sweden has implemented another AI to identify long-term throughput bottlenecks. They are now using it to plan maintenance actions on bottlenecks. Apart from these auto-manufacturers, a global leading manufacturing analytics software provider has included my AI to identify long-term historical throughput bottlenecks into their software packages. Any manufacturing company that buys their software can now use my AI to identify bottlenecks.
From my experience working with bottlenecks, I can recommend the engineers three points:
- Start the AI journey in small steps. For example, if your factory is not digitalized, then deploy data collection technologies and collect machine activities data. Once you collect the data, you can use AI to monitor real-time bottlenecks and take actions in real-time to remove them. This simple way of tracking real-time bottlenecks will help to reduce variations in the factory throughput. Once you collect sufficient data over time, then you can use AI to deeply analyze and diagnose historical bottlenecks and take actions to eliminate them which will help to significantly improve throughput. By accumulating data over time, patterns might start to emerge. You can then use AI to learn these patterns and use them to predict future bottlenecks and also prescribe proactive elimination actions on them.
- Start with the available data. Do not wait for the perfect data. Having perfect data is a moving target. In my research, I have shown that simple ANDON lights data can be used to analyze bottlenecks.
- Augment the AI insights. Though AI accuracy improves over time as it learns from data, do not expect it has 100% accuracy from the start. You need to augment the AI results with your domain knowledge on factory dynamics to take effective elimination actions.
Tell us, where you are from.
I am from Chennai, India: A small city with 7 million people 😉. I was born and lived there until the age of 23. I moved to Sweden to do my master-level studies in production engineering at Chalmers in 2013.
I chose Sweden to pursue my higher studies as I saw Sweden was strong in manufacturing education.
Also, many Swedish manufacturing companies were massively expanding in India at that time. Seeing these trends, I decided to move to Sweden. After my master-level studies, I started my Ph.D. studies at Chalmers. So, it is 7.5 years now since I moved to Sweden. Time flies!
I enjoy taking a walk in the forest, be with nature, and get inspired by nature. I am also an Indian Carnatic singer. So, I practice singing when I am not working.
After many years of work you are finally done. What is your next dream project? (or just project)
My research is driven by a passion for building reliable AI technologies to tackle real-world problems in manufacturing. I approach manufacturing problems with a computational lens and develop algorithms to analyze large sets of machine data to support decision-making. In my Ph.D., I solved an important manufacturing problem of bottleneck analysis to help companies improve productivity. In the future, I would like to continue building AI technologies for routine diagnosis and prognosis of factory dynamics to achieve higher shop floor productivity.
THANK YOU Mukund!
Good links:
Chalmers Production Area of Advance
https://www.smukund.com/ Link to some of the You Tube research videos:
Identification of bottlenecks
Prediction of bottlenecks
DAIMP Research project (in which I developed the AI algorithms for bottleneck analysis)