A digital twin is a virtual model that accurately reflects a physical object. These are basically sensors that produce aspects of a physical object’s performance like energy output, temperature, etc. This data is then relayed to the processing system and applied to digital copy. In sectors like manufacturing, digital twin is quickly emerging as a game-changer.
To know more about digital twins and data analytics in general, Analytics India Magazine caught up with Vinay Jammu, vice president of physical-digital technologies at GE digital.
Edited excerpts from the interview:
AIM: Tell us about your profession. How did you develop an interest in forming a career in AI and analytics?
Vinay Jammu: Ever since I saw a program on National Geographic on Artificial Intelligence (AI) in my school days in the late 1980s, I was fascinated with how computers can capture and represent intelligence. One-legged robots that could balance themselves by using dynamic balancing and micro-robots that behaved like insects and scurried away from humans and light opened my eyes to the fun world of AI. Here is my professional journey and my top 3 learnings.
After completing my PhD in 1996, I joined Mechanical Technology Inc. and continued my work in bringing domain and data-based knowledge together with AI for industrial equipment health prognostics. I had the opportunity to work with institutions like Rice University, the US Department of Energy, the US Army Space and Strategic Defense Command, and the US Department of Defense.
I moved to the GE Corporate Research & Development Center in 1997 and continued to develop advanced AI-based solutions for predicting failures and life of critical equipment such as gas turbines, jet engines, MRI machines, wind turbines, and locomotives. In 2002, I moved from the US to India and was promoted to lead the Prognostics Lab in GE Research with a dozen researchers developing next-generation technology for machine health prognosis.
In 2020, I moved to GE Digital as Vice President for Physical + Digital technology and continued to develop domain knowledge + AI-based solutions for GE and its customers.
AIM: What are your responsibilities in your current role?
Vinay Jammu: In my current role as VP for Physical + Digital Technologies for GE Digital, my responsibility is to bring Lean principles and Digital Technologies together to impact GE businesses and its customers. My team and I are responsible for building AI-based solutions for inspecting wind turbine blades, image-based damage quantification algorithms for gas turbines, and AI-based data quality improvement algorithms – all of which ultimately helps our customers drive better business outcomes and operational excellence.
Digitally transforming an organisation requires a systematic approach to identifying the right problem statements that matter to the business and finding the right data and AI models to impact these problems. Today most organisations have siloed value statements that address business value on a case-by-case basis and not at the organisation’s financial statement level. This makes it difficult to scale and sustain digital transformation.
What is critical in my current role is how we use Lean principles such as Hoshin Kanri, Value Stream Maps (VSM), and Standard Work to identify transformative business problems and link them to data and model value maps for systematic and sustainable business transformation.
AIM: Digital Twins with AI & ML is believed to transform the future of the industry. What are your comments on this?
Vinay Jammu: Digital Twins are living-learning models of assets, processes, systems, or networks. The concept was originally developed by Professor Michael Grieves and first applied at NASA for continually improving product design and engineering. NASA develops unique first-of-its-kind technologies such as the space probes, the MARS rover, and many more. The challenge they face is that during the design time, it is difficult to anticipate all the operating conditions a technology such as a MARS rover is expected to see. Billions of dollars are spent anticipating operating conditions and testing for these conditions in a lab to eliminate as many issues as possible before launch. However, unseen conditions may still be encountered while on a mission.
The goal of Digital Twins is to learn from new conditions as they occur so that they can be used to respond in real-time, where possible, or improve designs that can handle these situations better in the next iteration. The higher-level goal of Digital Twins is to provide the ability to continuously learn as new data becomes available, so the predictions they make are of the highest accuracy possible.
Digital Twins bring the best of data-based knowledge and domain-based knowledge together using artificial intelligence to build models that produce more accurate results than ever before. These models can be used to solve some of the very critical industrial problems we are facing today. Digital Twins can be used to improve forecasting power production from wind turbines so we can use more renewable energy and move towards net-zero carbon emissions. We can build Digital Twins of patients to drive personalised medicine to improve the treatment of diseases such as cancer. The uses of Digital Twins are unlimited as the technology develops and we evolve the science behind them to solve some of the world’s toughest challenges.
AIM: How do Digital Twins help in bridging the physical and the digital world? What to expect in the near future in terms of this technology?
Vinay Jammu: Digital Twins follow the “Garbage-In Garbage-out” principle. If the right data is not made available, Digital Twins will not perform well. To do this, Digital Twins need to interact with the physical world to collect data.
In summary, physical and digital must come together to solve the challenges we are facing today. In the future, we can expect more data and more computing to be available. Sensors will become even cheaper, and more of them will be embedded in different kinds of assets, increasing their ability to produce better data and become smarter. We are seeing more automation happening now, and it is expected to continue. As this evolves, we can expect “automated systems” to turn into “autonomous systems,” where machines would autonomously handle tasks of increasing complexity and uncertainty (e.g. self-driving cars).
AIM: Tell us about your research on gearbox diagnostics that NASA sponsored?
Vinay Jammu: My PhD work involved developing diagnostic systems to predict and prevent failures in helicopter gearboxes. NASA has a US Army Division at the Glenn Research Center at Cleveland that designs and tests new helicopter gearboxes. A failure in a gearbox can result in loss of human life as well as significant financial loss. Typically, this group runs experiments costing millions of dollars to generate failure data that could be used to predict and prevent these failures and improve the safety and reliability of helicopters.
My research work involved understanding how humans use domain knowledge of gearbox design and available experimental data to perform diagnostics and develop a system that mimics human processes. In addition, the system continuously learns to use new data that comes in as the helicopter starts flying to improve its performance. I built a “neuro-fuzzy” inference system where the domain knowledge of the gearbox vibration signatures from a simple physics model is incorporated as “fuzzy” rules into the inference system. This continuously learns using a neural network-based learning algorithm to improve diagnosis as new flight data comes in. This was in the form of a Digital Twin (though we did not call it that), and NASA recognised this work with two New Technology Innovator Awards.
AIM: Which machine learning and artificial intelligence-related technologies will truly disrupt the industry in the coming years?
Vinay Jammu: Today, we are at ANI – Artificial Narrow Intelligence, where AI can solve specific narrowly-defined tasks such as face recognition. The next evolution of AI is AGI – Artificial General Intelligence, generally considered as a human-level capability that can reason and solve problems on a broad range of tasks and topics. This is still considered a tough challenge with significant complexity and technical challenges. Finally, ASI is Artificial Super Intelligence, where AI can solve problems humans can’t solve today. AI is expected to evolve along these lines in the next few decades.
There are few key areas that accelerate AI along these lines. These include:
- Identification of use cases/problems worth solving for society
- Availability of data and learning algorithms
- Understanding and representing context (including meaning, purpose, right/wrong)
Some of the top problems facing our society today include – the energy transition as we move towards a net carbon zero economy; precision health that provides access to affordable, accessible, and high-quality healthcare for all 7+ billion people on Earth; safety and security, including cyber security; and efficient transportation of goods and people including autonomous vehicles, next-generation transportation, and an optimised supply chain (including food waste reduction, for example). AI is a critical technology that can help solve these problems with the right physical + digital solutions.
AIM: What is your advice to aspiring ML engineers? What can they do to stand out among a large number of applicants?
Vinay Jammu: I would suggest that aspiring ML engineers first get a strong foundation in data science by taking courses in probability and statistics, algebra, linear systems, information theory, and signal processing before they venture to advanced AI and machine learning.
Second, there are many good online machine learning courses that one can take to gain knowledge, but depth can be gained only through taking multiple courses or doing a Master’s degree and applying your knowledge to real-world problems.
Third, there are a significant number of free online data sets available at websites such as Kaggle.com for data scientists to develop themselves. By practising with these data sets, they will learn the importance of domain knowledge in data science.
Finally, it is important to get a mentor/coach who can help guide them – that would be key to ensure they are on the right development path.