Above: Machine learning can help the nuclear industry cut its asset management costs now and over the lifetime of plant

In the past, the nuclear industry has been wary of digital solutions and artificial intelligence. The industry’s concerns have ranged from making the safety case, to the fast pace of digital evolution and the need to ensure that systems do not become obsolete over the long lifetime of nuclear units. But others argue that the industry should take the opposite viewpoint. Sam Stephens says a digital approach can help the industry cut its asset management costs now and over the lifetime of plant.

Stephens should be enthusiastic: he leads on digital transformation for Atkins and its parent company SNC- Lavalin. But he says nuclear’s basis in innovation will enable it to make gains from machine learning and artificial intelligence (ML/AI) because: “The principal factor in the nuclear industry is that we have a commitment to use the best available technology.” He says gains have been made over the last 10 or 20 as digital solutions have matured and been widely adopted in other industries, so “if the nuclear industry doesn’t adopt them it won’t be able to leverage its technology.”

Stephens argues that at the same time the industry has to make sure it is not disadvantaged, technically or economically, as the power sector around it changes. “The energy system we are operating in is becoming increasingly dynamic and increasingly digitalised,” he says, and its nuclear units have to move with it. Meanwhile, the need for low-carbon power supplies makes it still more imperative that nuclear units operate at the optimum and with the minimum of plant downtime.

Improving the safety case

Nuclear’s focus on safety means the benefits outweigh costs already in some activities, says Stephens: “By adopting a digital approach on sites we see opportunities to take people away from the hazard and reduce their exposure to it. [We can] also identify smarter approaches and better strategies to optimise the way we deliver work and what work we deliver.”

What applications can already deliver benefits to the nuclear sector? Stephens gives the example of machine vision: “We are capturing very large quantities of photos and videos from sites. With those datasets you can then use machine learning algorithms to identify any potential or common defects.”

Walkdowns currently carried out by workers who have to enter radiological areas could be entirely automated, using a robot or drone to image the route each day and a machine learning algorithm to bring any changes or anomalies to the attention of human specialists.

Such applications can be implemented without connecting with other plant systems, so the risks of implementation are lower and “that is what we see as being an ideal first use-case,” he says.

For such uses, Stephens argues against excessive caution: “the sooner you start to invest in these types of approaches, the more the value that you can get out of it, given that nuclear plants have a finite life,”he says, noting that the investment case is easier to make now than it will be in 10 years’ time, when plants’ remaining lifetimes are lower.

Other similar ‘self-contained’ options include planning optimisation and using artificial intelligence to optimise programmes and schedules both for outages and general delivery programmes.

Of course, project management software has been around for a long time, but he says, “when you consider project programmes and the high number of potential independent interdependencies and constraints, artificial intelligence opens up the opportunity for you to consider many more different scenarios than a human brain could. Then you can use it to help identify the optimum scenario that will deliver your programme ahead of schedule and ahead of budget.” What is more, it can bring together planning and project management software with other types of information like a 3-D understanding of the plant and the resources deployed.

A third opportunity for easy wins is using existing data pools. Stephens explains, “a lot of plants have been collecting data through enterprise asset management systems. Where we see opportunities is around firstly how we can use AI/ML to spot trends”. That may mean changes at one site or piece of equipment, or it may mean common trends that might occur across a lot of plants.

Finally, Stephens notes that AI/ML can help inform decisions about where additional data might be required. As sensors become cheaper and cheaper, AI/ML can help in planning and prioritising where a company can best place sensors to predict plant issues and reduce downtime.

Replacing lost expertise

Can artificial intelligence replace the all-round experience of long-serving nuclear industry members? Stephens insists that the aim is a partnership. “A lot of our approaches to technology adoption is looking at how we can provide tools that make people in the plants’ lives easier”. He also thinks it is future-proofing the industry against losing individuals’ experience. “When you look at the challenges the industry faces, particularly around skills, increasingly we are going to be resource-constrained. With that shortage of skills we need to find smart ways of doing things more efficiently and differently. In that respect we see technology helping.”

The nuclear industry faces a particular challenge in innovating because new processes have to be qualified so self-contained use-cases are more likely to be able to pay back quickly. In practice, Stephens says that rollout “comes down to the business case and the risk appetite. You have to acknowledge with these new technologies the business case will have to be proved. If it has a return on investment of five or 10 years the current management will take a view of ‘not quite yet’.”

That means ML/AI and digitalisation advocates have to “identify uses that have immediate benefits that set us up for incremental further investment”. Even if plant operators are cautious, one way they can reduce implementation costs at a later date is to consider their data management. SNC Lavalin is making sure that all of the data it captures from nuclear facilities is structured to be readily available to get service insights. He says, “When you have the data structured and available and as you increase the quantity, then it increases the opportunities to use data analytics. That needs to be put in place first of all, before you can leverage some of the smarter data analytics processes.”

Data security

For good reasons, the nuclear industry is conservative and very rigorous about data security. However as the data industry matures nuclear is able to follow its infrastructure peers into more flexible arrangements while retaining that security. For example, Stephens says nuclear will have to follow other industries in shifting towards solutions that are cloud-based. It can do that safely, he says: “That is an inevitable move, as a lot of software providers increasingly put solutions on the cloud, rather than as traditionally on servers. That is a change that the industry needs to embrace. There are arguments that you can improve security with these facilities.”

Attitudes are starting to change. “The pandemic was a big shift, because a lot of operators found that they were reliant on Microsoft Teams and cloud-based solutions to support ongoing work. That’s one area where we found the pandemic really accelerated things.”

Now, he says, “some of the newer technologies that are available enables us to develop solutions much more quickly and in a really agile way. Of course, it has to have a cyber security risk assessment.”

Cloud-based systems provide some future-proofing too, he says: “A lot of clients are increasingly nervous around vendor locking. Signing up to a solution means that essentially switching costs in the future could be prohibitive. We are looking at how they structure and the standards they use around the management of that data, so that it increases portability between systems. With a cloud- based approach it is less likely that you will be locked into any one vendor.”

These types of stepwise move towards better use of data and more flexibility about, for example, using the cloud, needs regulatory acceptance. Stephens says regulatory consideration will also be stepwise. “It is one thing being presented with insights that help humans make better decisions and there is another thing relying upon those in order to assure nuclear safety,” he says.

From a regulatory perspective “there is still quite some time to go before the first line of defence is relying on AI/ML to spot a potential issue. That is a combination of technological of maturity and also maturity of adoption within the industry. And in order to do that you require a really strong foundation of skills, of people understanding and using it and a lot of experience in having used it and some of the pitfalls and constraints.”

The role of digital twins

So-called ‘Digital twins’ are the buzzword of the energy industry at the moment, but Sam Stephens says the phrase means different things to different people. “The digital twin could be anything from a laser scan of the facility that represents the current condition, through to something that is more like what Rolls-Royce or GE will do with their jet engines, where you have a piece of plant that is fully connected and sharing live data with the operators. There is a spectrum.”

In the nuclear sector, he thinks the three key stages of the nuclear life cycle – new build, operation and decommissioning – each gains benefits from a digital twin, “But it will have a different flavour for each part of the life-cycle.”

He explains, “We see low maturity digital twins supporting major programme management of Legacy assets. For new build it is really around how you set the strategy to support operation but also capture the data in design and build to preserve that value for the lifetime. In operational assets it is about strategies to support operation of the plant.”

He adds that “we expect the assets to become even more connected to a virtual representation. This is a long-term trend that we don’t see going away.”

That does not just apply in fission: fusion developers are looking ahead: “We have worked very closely with some of the fusion programmes because in order to ensure that your plant is going to work, more and more of the testing and analysis and simulation is going to need to be done in silica rather than on the plant. As assets become even more complex they require digital solutions to design and build them.”

Author: Janet Wood is an expert author on energy issues