After receiving his bachelor’s and master’s degrees in computer science and engineering, Dr Sourabh Bharti completed a PhD in information technology at the Indian Institute of Information Technology and Management in Gwalior. During his PhD studies, he also worked as a visiting researcher at Anglia Ruskin University in the UK and was awarded a scholarship to pursue partial doctoral studies in Hungary.
He then received a Marie Sklodowska-Curie Actions fellowship in 2020 and joined Confirm, the Science Foundation Ireland research centre for smart manufacturing. Currently, he is based at the Nimbus Research Centre for cyber-physical systems and IoT at Munster Technological University in Cork.
‘Our research is an attempt to encourage organisations to participate in a secure intelligence sharing exercise without compromising on the privacy of their commercial data’
– DR SOURABH BHARTI
What inspired you to become a researcher?
It was back in 2012 when I was working towards my master’s thesis. My supervisor introduced me to the world of research, which gave me a free hand to explore and express independent research ideas.
I got familiar with publications and authorship, which was a perfect gateway for me to connect with the outside world. Moreover, being a researcher does not allow me to settle and thus everyday there is something new to learn and explore – which is the part I love most about research.
Can you tell us about the research you’re currently working on?
The project focuses on realising distributed machine learning on edge devices deployed in the smart manufacturing space to gather various parameters related to industrial assets and operations.
The idea of this project stems from the industrial data silos which remain distributed across locations and are difficult to bring together due to industrial competition and data-privacy issues. This pushes organisations to process data closer to its origin (at local manufacturing sites) but still a single manufacturing site cannot garner all the patterns required for applications such as asset failure prediction and quality assessment.
This gives birth to a collaborative ecosystem where multiple participants belonging to different manufacturing sites of the same organisation or from different organisations agree upon sharing their intelligence without revealing their raw manufacturing data. This is currently being realised by distributed machine learning techniques such as federated learning.
Our project focuses on making federated learning suitable for resource-constrained edge devices by employing lightweight predictive models and enabling privacy-preserving and resource-aware computational offloading whenever required.
The project is in line with EU’s data strategy, which advocates for the shift from centralised cloud-based processing to edge-based processing. We are closely working with our industrial partner IBM to gather information about the current industrial initiatives being taken towards the push for edge computing. We observed and worked on some fascinating use cases of edge computing such as a mobile product quality inspector developed by IBM in collaboration with Apple.
The project can deliver some cutting-edge edge-based solutions which can be adopted by the industry, straightaway! It might also increase the visibility of Confirm as this is still an emerging area and not many efforts have been made in the past to investigate collaborative ecosystems like this for smart manufacturing.
In your opinion, why is your research important?
Traditionally, manufacturing organisations process their data in silos. Reasons include growing industrial competition and reluctance to share the sensitive manufacturing data.
Our research is an attempt to encourage these organisations to participate in a secure intelligence sharing exercise without compromising on the privacy of their commercial data. The encouragements can be drawn from the reduced manufacturing cost generated from the improved organisation-level intelligence and incentives.
What commercial applications do you foresee for your research?
There are numerous commercial aspects of a collaborative ecosystem in manufacturing as the shared intelligence can pertain to different domains such as asset failure, energy efficient manufacturing practices etc. Robust pattern mining from the collaborative data space can minimise the manufacturing cost such as predicting asset failure before its occurrence, which can reduce unplanned downtime.
From the product point of view, edge-based smart manufacturing products such as mobile quality inspectors are in great demand as they remove the dependency on a subject matter expert to inspect the product quality. In addition, they can easily participate in a collaborative learning process without revealing their raw data to each other as it is currently being realised in other applications such as text-based predictions by Google.
What are some of the biggest challenges you face as a researcher in your field?
As the collaborative ecosystem is in its infancy, organisations are reluctant to move away from the traditional practices of data processing. The manufacturing industry is yet to accept such collaborative ecosystems as it requires processing the manufacturing data closer to its origin, which is not practised currently.
From the academic research and development point of view, the biggest challenge is the lack of relevant data sets to test the designed approaches.
What are some of the areas you’d like to see tackled in the years ahead?
I expect for the manufacturing industry to fully adopt edge-based data-processing to garner real-time pattern mining.
This is going to be a huge shift from the centralised data processing, but EU initiatives such as European Data Strategy really put edge computing and collaborative data spaces into perspective.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
The post How data sharing and AI could reduce downtime in smart factories appeared first on Silicon Republic.