“We have everything in this (Bengaluru) centre, right from the fundamental research to accelerate the design of new systems and new fuels, right through to the deployment of these digital capabilities to monitor our assets globally. And the beauty of that is, as the new capabilities develop in the R&D phase, we have the ability to deploy them within the same centre. So, we can actually take things from digital research to digital deployment right here. And that’s really the ambition, to accelerate digital transformation in Shell by using this hub for the rest of the group. So, India’s a big focus for us in that context.

-Dan Jeavons, VP, Computational Science & Digital Innovation

We are traditionally a chemical engineering and physical sciences company. So we have a lot of insight and knowledge in the way reactors are designed, processes are modelled. What we are trying to do now in this centre is to combine this traditional domain knowledge with digital, with AI, with computational science.

In simulating plant systems, for instance, we use SciML or scientific machine learning. SciML brings the best of both worlds-the speed of AI and the physics of engineering. In our initial tests, we find that with SciML techniques, we can do simulations 100,000x faster than traditional methods, with no compromise in accuracy.

-Suchismita Sanyal, General Manager, Computational Science.

Centre of digital transformation

As Shell works to digitally transform itself, and meet its sustainability goals, the India centre is becoming central to it. Launched in 2017, and hosting about 1,500 people, the centre’s biggest USP is its combination of those with an understanding of the physical sciences and those who know digital technologies, particularly AI/ML. “I think in the Shell world, it’s unique,” says Dan Jeavons, VP of computational science & digital innovation, who moved to Bengaluru last year, after 17 years with Shell in London, to accelerate the centre’s digital capabilities. “I’ve been blown away since coming here by the capability that we can bring to the rest of the Shell group,” he says.

A massive project the centre is leading involves bringing all the data from over three million sensors across Shell’s global asset portfolio – from LNG trains to refineries and chemical parks – into a single data repository in the cloud, building predictive models on top of this, and then monitoring the assets in real-time. Jeavons says the centre monitors over 10,000 pieces of equipment. Based on historical patterns, predictive maintenance algorithms have been developed and deployed, for instance, for valve failures. So, if a valve is about to fail, an alert comes.

While this works well, the system also tends to generate many false positives, because it’s based on historical patterns, and not what is actually going on inside, say, the refinery. Sanyal and Jeavons are now trying to address this. They are working towards developing what are called dynamic digital twins. This is the process of digitising and simulating all of the processes and flows inside the system. And these are becoming possible with technologies like scientific machine learning (SciML).

“Sensors throw numbers at you. So, you need what we call a physics based data reconciliation method. Physics brings the context of the plant. If this amount of oxygen is reacting with this amount of gas to give this amount of product, but the sensor throws something else, that is when an alarm should be raised,” Sanyal says.

Such dynamic digital twins, combined with chemometrics – the science of assigning chemical signatures to data – will also be useful in dealing with, for instance, corrosion inside reactors. Sanyal says the corrosion models the centre is developing will enable operators to understand when corrosion occurs, and exactly where inside the reactor the problem is. “This will help them to significantly reduce unplanned downtime,” she says.

Generative AI

Jeavons and Sanyal are also excited by generative AI. They expect numerous applications, but the most interesting may be in research. “We are providing generative AI the context of molecules and chemistries, and then using LLMs to search through large molecular databases,” Sanyal says. That means identification of molecules, like in the MotoGP fuel case, can be done even faster.

Extract from Times of India