Optimising complex clean-up operations

The challenge

A long-term clean-up operation

Washington River Protection Solutions (WRPS) manages one of the US government’s most challenging nuclear waste clean-up projects. It’s responsible for the Hanford Site, a 586-square-mile area in Washington state which currently houses more than 56 million gallons of radioactive and chemical waste.

Since taking over the site in 2009, the WRPS team has been looking for ways to reduce the time it will take to complete the clean-up, which has already lasted 30 years and cost billions of dollars. 

The team was already using predictive models, but those models didn’t account for delay, failure, repairs or maintenance – which meant they lacked the accuracy needed to make informed decisions about how to most effectively invest WRPS’s time and budget.

“We carry out highly complex processes, manage large data volumes and have long lead and repair times for specialty machinery,” says Douglas Hendrickson, Mission Analysis Engineering Team Leader at WRPS.

“Because the models we were using didn’t account for this real-world variation, they weren’t helping us understand how processes and projects actually affected timescales. We wanted a stronger evidence base that would help us make more informed decisions.”

Using its initial approach, it could take as long as a month to answer a single query and apply the result. Preparing input files took hours of work, while the tasks of running as many as 80 scenarios and analysing the output could each take days.

The approach

Introducing AI-powered digital twins

Since 2016, WRPS has been working with Royal HaskoningDHV Digital to introduce AI on top of their predictive simulation process. This adds a crucial layer of intelligence and real-time, real-world data to its models that feeds into digital twins of key assets and processes within their operations.

“Our objective has been to bring reality into the models, providing consistent insight into how scenarios affect mission life – and therefore cost,” says Douglas. 

“We needed the ability to analyse ideas and what-if scenarios, so we could understand the facilities, personnel, environmental impact, spares and instrumentation needed to achieve our mission in the shortest time possible.”

Using the WITNESS Horizon solution from Royal HaskoningDHV Digital, WRPS now has twins of thousands of components throughout the site, each using data from multiple sources to model different scenarios for the operation.

To boost the efficiency of its models, WRPS created a cloud-based app which uses Royal HaskoningDHV’s compute resources to test different scenarios simultaneously. This approach isn’t just faster – it’s also scalable and flexible, which means WRPS can adapt the platform as its digital twin requirements continue to grow. 

Each simulation the app runs produces real-time log data – just like a real production process – which feeds an algorithm that identifies bottlenecks within WRPS’s system. The team can also dig into the algorithm to understand how it makes predictions, which helps highlight which factors influence optimum output. 

This new ecosystem of digital twins, consisting of numerous simulations and constantly-improving algorithms is providing more detailed, accurate insight and speeding up critical decision making for the WRPS team.

The impact

Investing millions more efficiently

By replacing part of its existing simulation model with AI-based digital twins, WRPS has dramatically reduced the run-time of each simulation. Queries that could take close to a month can now be completed in as little as two days.

Each twin pulls in thousands of data points from multiple sources to give the simulations the real-world basis it needs to produce accurate predictions.

As the project continues, the team will enhance the digital twins by training and testing more AI-based models to integrate into the site’s everyday production processes.

Individual projects have also seen major process improvements using the new twinning approach. For example, by modelling the site’s effluent treatment facility – creating digital twins of over 2,800 different elements – WRPS identified a filtration bottleneck which would impact a planned $30 million of upgrades.

Even with the upgrades, the bottleneck would have made the team miss its throughput target by as much as 75%. Using evidence from the digital twin, Douglas and his team gained rapid approval for $6 million in filter improvements which ensured the project would exceed its target.

Similarly, the team used a digital twin to analyse the impact of installing a secondary waste pump into the site’s double-shell storage tanks.

When the single pumps within the tanks broke down, the highly technical nature of the equipment and the risks from dealing with radioactive waste meant repair frequently took more than six months to complete.

The digital twin model showed that a redundant pump could reduce that by 45 days each year – resulting in millions in cost savings.

“These projects garnered high levels of recognition internally and made significant contributions to operations,” says Douglas. “Other models provide ongoing impact and value by supporting predictive maintenance and challenging process assumptions. Predictive simulation is helping us remove blind spots, pre-empt issues and proceed with our mission more efficiently.”