On the future of industrial machine intelligence

Interview with
Dr. Duane S. Boning, Electrical Engineering, and Computer Science in the EECS

Dr. Duane S. Boning is a Clarence J. LeBel Professor of Electrical Engineering and Computer Science at MIT. He is also an Associate Director of the MIT Microsystems Technology Laboratories (MTL), as well as a Co-Director of the Leaders for Global Operations (LGO) and Machine Intelligence for Manufacturing and Operations (MIMO) Programs. The tec.News staff recently had an opportunity to interview Professor Boning and ask about exciting new developments in machine learning, specifically for industrial applications.

INDUSTRIAL COMPETITIVENESS

tec.news: Could you describe the progression of machine intelligence in the context of operations and manufacturing transformations?

Professor Boning: Let’s start with some history of operations and manufacturing methods. 30 years ago, Lean Manufacturing methods transformed the industrial sector. These methods focused on eliminating excess inventory by receiving goods only when they were needed, reducing excess costs for storage and increasing productivity and profit.

Then, 15 years ago, there was a significant shift of the supply chain from local sourcing to global sourcing with a heavy dependence on Asia. The COVID-19 pandemic revealed inherent weaknesses in this type of system with global supply chain disruptions resulting in massive shortages.

Digitalization and machine intelligence offer a counterbalance to these issues with the possibility of accelerating manufacturing competitiveness and making supply chains more resilient.

In addition, recent advances in sensing, low-cost powerful phenome computing, and software tools provide the opportunity for new innovative manufacturing applications.

tec.news: Could you outline what you see as the main challenges to applying machine intelligence for industrial applications?

Professor Boning: My research and experience led me to the conclusion that bringing machine intelligence into manufacturing and operations poses challenges for the average manufacturer, one that needs to be surpassed in order to achieve the promised benefits. Due to custom processes, off-the-shelf machine learning is hard to apply in most manufacturing situations. Machines must be taught by a user who completely understands the problem

to create an effective solution. My research focuses on developing people that are “translators” with domain knowledge about machine learning and production processes to improve success. I work on teaching undergrad students the fundamentals of machine learning as a foundation so that they can bring this necessary skill to future employers.

The first step in machine learning is having the appropriate data as a starting point. However, just having data is not necessarily enough for a particular machine learning application. You need to have relevant data. There are a number of machine learning techniques including statistical methods, decision trees, random forests, neural networks, time series analytics, etc. that should be applied using practical engineering and business judgment. Other factors to be considered include methods for dealing with outlier data, data structuring, anomaly and fault detection, and model drift. Model drift is the phenomenon where data correlations made at one point in time change over weeks, months, or years as the process equipment ages or the manufacturing processes themselves subtly change over time.

In short, successfully developing a machine learning model requires a broad understanding of data modeling and access to the right data.

tec.news: What do you see the role of industry – academia partnerships in helping to solve these problems?

Professor Boning: Industry – Academia partnerships are critical to providing real-world platforms and case studies for investigating and testing new approaches and hypotheses. At the end of the day, both are needed to address real world needs.

MIT has a long history of such partnerships, and through organizations such as the MTL as well as other centers across MIT, we are able to leverage the strengths of both the industrial and academic environments and live up to the spirit of our motto, MENS ET MANUS –or the merging of the theoretical and the practical.

tec.news: What do you see as the future of machine learning and the outcome of these programs?

Professor Boning: Machine Learning is critical for the next evolution of manufacturing processes. The work we are doing today helps lay the foundation for the future. These programs teach necessary concepts and skills that these students will bring to future employers, who in turn, will benefit from competitive advantage.­