IRPC Process On-Demand: Harnessing Cognitive Technology in Refining to Accelerate Digitization

 

Beyond Limits Refinery & Process Manufacturing Product Manager, Leslie Rittenberg joined a stellar International Refining and Petroleum Conference (IRPC) line-up with industry experts from BASF, Shell, Dow, ENI, Tupras, Schneider Electric, and more to discuss Harnessing Cognitive Technology in Refining to Accelerate Digitization.

 

Check out the talk to learn how refiners like bp are using AI to close the profitability gap and operate closer to plan. Watch this recording or read the transcript below.

 

Welcome, I’m Leslie Rittenberg, the Refinery and Process Manufacturing Product Manager from Beyond Limits. Today, we’re going to be talking about harnessing cognitive technology in refining to accelerate digitization.

 

I think we’ve been on some great digital journeys over the last couple of years and improved in many spaces; everybody would like to know how to continue to use AI as we operate closer to plan. 

 

We’ve used machine learning and AI applications as they relate to availability and optimization; this talk is going to be focused on a discussion in the optimization space as it relates to refining.

 

When optimizing, what do we strive for? We certainly strive for better raw material yields, wanting faster and more confident decision-making by our operators. We want to always improve our product quality without too much product giveaway. We want to be able to respond to the constraint management challenges and bottlenecks we face in our plants. We postulate that, if we use AI, it’s going to help us start to significantly improve in all of these areas.

 

What have been some of our challenges? 

 

I know many of us have done a significant amount of proof of concepts, trying to deploy new software products. In all of those journeys, what we’re trying to do is capture our work and knowledge then encode and digitize that information as fast as possible. This is a race to accelerate and digitize across a business. We want to accomplish this as effectively as possible without facing an extreme amount of programming.

 

We have also found that we don’t want a black box solution. We want to understand how our models, AI, and advanced analytics come to their conclusions and give us the recommendations that we can trust. 

 

Finally, humans still need to be kept in the loop with the ability to continue making context-rich decisions. If we do not have consistent and trustworthy results from our AI models, then we have difficulty in deploying and using them.

 

Here’s a diagram that shows a little bit of the journey we’ve been on. We certainly have improved our process data. We’ve brought a lot of that data into the cloud and spent a lot of time cleaning up data quality. In many cases, we’ve deployed new sensors for even stronger confidence in our numeric solutions.

 

We’ve also deployed numeric AI – which has helped us see things faster or understand certain relationships that we didn’t previously understand, training them on a lot of historical data. We are now alerting and visualizing messages to be more proactive in our plants, as opposed to reactive.

 

However, many numeric AI programs still give us false positives in several cases. Again, this goes back to the level of trust that we have and the amount of work needed to dismiss the false positives; we have a lot of work, in terms of model breakdown, and in sustaining those models, to give us the best solutions.

 

With anything, because of the safety-critical nature of all of our businesses, we really want our operators to buy into the solution, so they can trust it and take the optimal course of action.

 

So, what can we do about this?

 

We can start to insert what Beyond Limits is recommending: symbolic AI. It’s a more effective hybrid – or composite solution – that is going to start closing the gaps and deficiencies we’ve been experiencing with solely numeric AI solutions. Symbolic AI has human-like reasoning and solving capabilities that can help us with the deficiencies we’ve been seeing in numeric AI.

 

What do we want it to do? 

 

We wanted to improve that decision support package for the humans in the loop, predominantly our operators. The solution can recommend classifications, give us confidence scores, and provide recommendations. Most noteworthy, the solution can act as a glass box, providing cognitive traces that explain how it came to its conclusions. We are also able to interact with the solution, increasing operator trust and reducing uncertainty; ultimately, arriving at a faster time-to-decision and upscaling our workforce simultaneously.

 

Now let’s get deeper into refinery-specific challenges.

 

Over the years, especially in the space of optimization, we’ve been innovative – and with most of us utilizing advantage feedstock programs. We’ve improved our profitability by buying cheaper feedstocks and crudes, and they generally have more challenging needs when processing them in our refinery; having higher sulfur and are heavier crudes by far.

 

Also, especially with last year’s pandemic, we’ve seen a large amount of our experienced workforce depart. Many of the major oil and gas companies offered severance packages with the challenging P&L operations, and a lot of experience was lost. 

 

Finally, with product demand challenges that we saw result from the pandemic (and cyberattacks), large volatility in our refinery margins is a persistent problem. We’ve responded well over the years by incrementally improving our margins. Many of us have invested in the front end of our kits, modifying our crude units, and began the process of major capital investments. 

 

It’s not unusual to be faced with new bottlenecks and constraints that show up downstream of your refinery. When faced with those constraints, they present new operational challenges when it comes to executing a commercial plan. They can create volatility in plant operations, often changing the predictability of catalyst life, as well as instigate unplanned maintenance and downtime. We’ve all seen these challenges, knowing that our operators always need assistance and support from engineering experts.

 

Here’s a diagram as we start to think about how we execute our refinery commercial plans. As you can see, on the bottom level, we have some very strong systems and control processes. We’ve invested heavily over the years to automate and to have a high level of confidence in the performance of that equipment. 

 

On the top level, we’ve clearly invested in our planning and scheduling processes. We’ve improved our models in that space, we have very capable staff in that space, and we have high expectations that we’re going to deliver a very executable and profitable plan when we process our different crude feedstocks.

 

Now, if we look at the middle section, we have improved our operations staff over the years through simulator training and obviously have turnover in that space, providing good training when we bring on new staff. 

 

But if we think about it in all honesty, how we hand over the plan that we want them to execute flawlessly each time, has not changed tremendously over the years. We still experience, even though we have definitely improved levels one, two and four, we still have a gap between our plan and actual execution when we carry out a commercial plan. 

 

Beyond Limits felt we could place an advisor in this space to close that profitability gap, serving as an assistant to operations in real-time and identifying plan versus actual performance with full transparency as one source of the truth. This would provide the ability to execute closer to plan. 

 

The advisor also works to encode expert engineering knowledge, actually helping operators solve the gap problem in real-time. It deploys an AI reasoner that looks at process constraints and recognizes the best commercial options for an operator to utilize in order to adjust the plant and achieve a higher level of profitability. It iterates and tells the planning and scheduling organization whether their plan was as achievable as they thought it would and/or where the gaps for improvement exist. 

 

Over time, we are seeing an improvement in adherence-to-plan as well as the increased ability to iterate across an entire organization and fundamentally improve the capacity to deliver solutions that shrink plan gaps.

 

We have deployed our solution at a very large complex refinery in Indiana; bp’s Whiting refinery is within the top 10 of the largest, most complex refineries in the US. Before implementing our solution, they were struggling in a couple of places.

 

For example, when they were doing feedstock campaign changes, they would have inconsistent operations and a larger than average unsteady state during those changes. They were faced with competing priorities in their constraints and, at times, operators did not fully understand the instructions or the priorities that were coming from the commercial group. Due to this, they would have uncertainty in their decision-making leading to flawed commercial execution.

 

The solution we deployed had a no-code SaaS application in the cloud. It allowed bp to optimize according to the plan, closer to 100% of the time. The unsteady state that they were experiencing during campaign changes was reduced, and they upgraded to higher-value products. Overall, operator agility was accelerated with increased confidence leading to faster decision-making.

 

Another incredible thing around the solution’s deployment was the adoption time. I’ve been involved in software deployment packages over the years in many different plants; bp’s Whiting team was successful in adopting this program in one month. They were able to engage with their entire process and optimization engineers and create over 200 complex refinery objectives to represent their commercial plan. They represented this with no code-in language, with which operators were very familiar.

 

After deployment, we monitored their ability to meet the plan and we saw a 17% improvement in operating to plan. We also saw that they used over 2000 real-time sensors to help the operators evaluate and prioritize.

 

Beyond Limits is excited to bring these AI products to many different industries and within the oil and gas sector.

 

The example I just gave you was just one of our solutions – our Refinery Operations Advisor package. We also have offers in the well production space with well infill and well health solutions. Well infill is focused on improving overall production and helping production teams better understand where to drill while well health manages the health and asset monitoring around wells. 

 

Our Cognitive Formulation Advisor is also helping the lubrication portion of our business create more economical formulations, faster.

 

In power and water, we also have sensor placement solutions – while in healthcare, we have a smart patient monitoring solution. We offer investment solutions to the finance industry as well.

 

Thank you for your time; I’m looking forward to answering any questions you may have.