Chemical Plants Hit Automation Roadblocks on Path to AI

Process plants have pushed legacy automation systems to their limits, shared Jason Urso, chief technology officer for Honeywell Industrial Automation. Decades-old control systems served their purpose of keeping plants running for longer periods without disruptions and bridging the knowledge gap between incoming and outgoing employees.

“But we've reached the maximum ability to solve those problems with the traditional technology,” Urso added.

Major industrial automation providers, such as Honeywell, Rockwell Automation, Schneider Electric and ABB, are integrating AI capabilities into their automation platforms, often working with IT companies like Microsoft to help the chemical industry scale the adoption of smart technologies.

AI-Enabled Automation: The Next Frontier

For example, last year, Honeywell introduced Field PKS, which queries AI to retrieve critical maintenance information, such as job tags, spare parts inventory or piping and instrumentation diagrams. Field PKS and similar systems act as agents or partners that perform tasks or make recommendations on a worker’s behalf, Urso said.

“It's moving from point-solution approaches to one in which AI-based agents are running within your system constantly, monitoring and providing advice to humans to help them perform better and reduce variability,” he explained.

Urso offered two real-world scenarios where AI-enabled agents can help chemical manufacturers optimize operations. The first involves traditional visual inspections of rotating equipment in which operators use their experiential knowledge to determine the root cause. However, ongoing workforce shortages and retirements are leading to a significant loss of this expertise.

“That works great for the 25-year professional, but the three-year professional often needs to call somebody who's an expert and say, ‘Well, what do you think this is?’”

The digital assistant can mine years of service records across multiple sites, looking for commonalities across recurring events.

“If you set up the data model correctly, you get a response that says, ‘Well, these issues have occurred 20 times in the last couple of years; most likely it is a bearing fault. Here's the procedure that you should follow,” Urso shared.

That means newer employees can operate like a 30-year professional, he added.

In October, Honeywell said it would partner with Chevron to develop advanced AI-assisted solutions, including an alarm guidance application.

The AI-enabled solution provides guided and specific actions for alarm response. The system will mine historical data on past actions to identify patterns of alarms and the corresponding operator actions that successfully return the process to normal operation.

Another example involves the use of acoustic and video-based sensors that can detect problems through sound or images. These sensors provide information about what they’re “hearing” or “seeing,” so operators can interpret the data without inspecting the equipment up close, Urso noted.

Overcoming Legacy Infrastructure Challenges

Robert Swim, chemical industry manager for Rockwell Automation's North American business operations, said achieving this level of connectivity remains a challenge for chemical producers because of their legacy technology infrastructure.

Fragmented "islands of automation" on process plant floors remain a major barrier to adoption for chemical producers. Swim said chemical manufacturers often underestimate the age and limitations of their existing equipment.

To address these obstacles, automation providers are creating structured, multiyear digital transformation plans tailored to each facility's specific needs. These plans include assessments of existing infrastructure.

“People want to embrace the new technologies like AI, but they realize they need to build out a broader, robust OT infrastructure that's going to lend itself to the results of AI,” Swim said.

According to Swim, his team might spend a half-day on a customer’s site assessing their automation needs, beginning with a conversation about their business strategy.

“Everyone says [they] want to jump into the AI space, but are they prepared for it?” Swim asked.

Some chemical manufacturers are so far behind on digitalization that they must master the basics before even considering advanced AI applications. As Swim highlighted, some companies still rely on paper-based batch tickets for data collection, and facility assessments often reveal non-functioning instruments on legacy control systems. These manufacturers need to first automate basic tasks and ensure their existing assets are functioning properly before they can pursue more sophisticated technologies, according to Swim.

For example, some assessments reveal old proprietary systems that need to be upgraded with more robust ethernet networks suitable for modern expansion and growth demands, Swim suggested.

Productivity goals, safety, cybersecurity, sustainability objectives and workforce enablement are five key pillars that Rockwell addresses with customers before recommending new automation technologies.

Building a Connected Chemical Plant

The long-term goal, noted Swim, is to help build a connected chemical plant beginning with a robust OT layer that can be integrated with AI for a comprehensive IoT (internet of things) model. He notes that Rockwell partnered with data-modeling software provider Cognite in 2022 to enable batch-performance analytics. The first release of the software, called FactoryTalk DataMosaix, is an industrial DataOps solution that gathers historical and event data from various systems and presents it in a consistent manner for batch-to-batch comparisons.

“It's really a golden batch environment comparison tool,” Swim said.

He added that the company is also building the module to compare overall equipment effectiveness (OEE). The key task is ensuring the solution has the information to feed an OEE model that accurately measures the KPI.

Swim stressed that chemical digital maturity varies across chemical companies, with larger players typically more advanced in their modernization plans. Oftentimes, the discussion with these companies focuses on better utilization of existing systems over major automation upgrades.

Swim said this includes recommendations to rationalize equipment or ensure basic building blocks are in place, such as upgrades to old DCS, PLCs and HMIs.

“Our customers will tell us we're not prepared. We don't have the SCADA, the control platform to embrace that intelligence,” Swim related. “However, we're investing in it now, knowing in the short term we'll be able to leverage that once we do some further modernization efforts. So, it is really kind of a strategy, a plan, looking at it longer term.” ⊕

Don’t Fall for the Hype

“I believe that trends are set by the vendors, not the users,” said R. Russell Rhinehart, emeritus professor in Oklahoma State University’s school of chemical engineering. Rhinehart, who worked in the chemical industry for 13 years before becoming a professor, said vendors are often repackaging old ideas to sell their products.

“The fashion industry is a great example. The sellers decide what is the fashion color each season, so that last year’s clothes are out of fashion, and consumers need to buy new clothes to be fashionable,” he explained. “The fashion industry influences the fashion magazines, who need new stories to print. Consumers like the pretty pictures and want to be a part of those desirable settings that the new fashion promises to do for them. The automation vendors seem to be selling Industry 4.0, autonomy, Big Data and all. They write stories for magazines and make presentations at conferences touting their products. And consumers seem to think that the solutions they need are the new fads. Bosses of control engineers ask them, ‘Is this what we need?’ Control engineers then spend their time understanding Gaussian processes, model-free modeling and such.”

Rather than jumping to the next flashy digital technology, Rhinehart said control engineers can deliver more value by understanding how to implement and troubleshoot classic advanced controls like ratio, cascade and override.

“The classic tools that we have available are much underused because the engineers that move into control jobs have only seen the tip of the iceberg of control and are not aware of the proven tools that are available to them,” he said.

Despite Rhinehart’s skepticism, he sees potential in digital twins and the use of machine learning “to help reveal cause and effect trends that may not be obvious.”

He pointed, specifically, to phenomenological-based dynamic simulation as a promising tool.

Most simulators used for plant design have been static models, also called steady-state simulators, Rhinehart said.

Vendors are now offering simulators that reveal how the processes develop in time, sometimes referred to as dynamic. He noted that dynamic models reveal how quickly events change, such as catalyst fouling or impurity build-up, which indicates how frequently maintenance intervention is needed.

But ever the cynic, Rhinehart added that he believes vendor hype is what’s really driving current digital trends.

“I also believe, as it happened with Neural Networks in the ’80s, that today’s marketing will generate a lot of interest, much investigation but little traction,” he stated.