Software Features Empower Process Optimization

AI, machine learning and predictive analytics encourage faster, data-driven decision making

To help offset the loss of experienced operators and optimize operations, advanced features, such as artificial intelligence (AI), machine learning (ML) and predictive analytics, are being applied to software tools to empower faster decision making, driving efficiency, flexibility and sustainability in the chemical processing industry so processors remain competitive in today’s demanding market.

Overcoming challenges

“Chemical processors face increasingly challenging environments where industrial AI has become critical for competitiveness,” says Bart Moors, general manager, process industries software, with Siemens (Munich, Germany; siemens.com). “Chemical companies face data-related challenges and many struggle to meet customer needs as the industry faces workforce shortages, making the drive toward autonomous capabilities vital.

“Traditional industrial AI excels at structured tasks like predictive maintenance and process optimization, helping to avoid unplanned downtime and ensuring product quality,” Moors continues. “Generative AI transforms operations by making them dialogue-oriented and intuitive, so they function as digital knowledge repositories, enabling instant access to best practices and rapid problem solving.”

He says advanced features address workforce shortages, sustainability requirements and flexible production needs, enabling processors to ensure plant availability, maintain agility and achieve cost and CO ₂ reductions while progressing toward autonomous operations that adapt to changes and operate with minimal human intervention.

Janelle Armstead-English, chemical industry principal, with Seeq (Seattle, Wash.; seeq.com) adds: “Chemical processors are operating in a tough environment, so they are looking for ways to continuously improve, innovate and leverage advances in digital transformation to optimize their processes.

“To address current industry challenges and achieve corporate objectives in today’s manufacturing environment, chemical processors need advanced analytics and AI platforms that integrate disparate data sources to support seamless analytics, visualization and the creation of both ad-hoc and highly polished reports,” she says. “These platforms are intuitive to use, unify multiple analytical methods within a common environment and connect users and data sources across the organization.”

Offsetting workforce shortages

Advanced features such as AI, ML and predictive analytics are being combined to offset the shortage of experienced operators.

“One of the biggest challenges processors face is that staffing on control systems has decreased,” says Bob Rice, VP of engineering, with Control Station (Manchester, Conn.; controlstation.com). “What used to be dozens of operators running facilities is now just two or three people. With fewer eyes on the process, remaining staff need tools that make their job easier and that help find more needles in the haystack.”

Control Station offers process and instrumentation diagram (PID) controller performance monitoring and analytics software that helps to identify the control loops that are driving instability in a plant. “Software makes the identification of issues, isolation of root causes and applications of corrective actions easier and faster through analytics that give value to raw data and apply expertise to find opportunities for process improvement.”

To make it easier for operators, Control Station introduced advanced features. “We added an innovative calculation called Overall Controller Effectiveness (OCE), which allows users to determine if a controller is running, free moving and tracking its targets. OCE provides a singular, easy-to-understand number that corresponds with overall equipment effectiveness (OEE), a performance metric that is widely applied in discrete manufacturing environments.

“And, we’re investing in what we call ‘casual machine learning,’ which analyzes data for the purpose of understanding the relationship between control loops and a plant’s economic drivers. It’s an automatic root-cause analysis capability,” Rice explains. “It addresses an increasingly common challenge, where people who’ve been running a plant and know what to adjust are retiring, taking their experience and know-how with them. Using these new analytics tools makes it easier to uncover those relationships, allowing less-experienced operators to become more efficient at uncovering and correcting controller issues” (Figure 1).

For similar reasons, Honeywell Process Solutions (Charlotte, N.C.; honeywell.com) is adding AI to components of its Experion Process Knowledge System (PKS) suite of process control software. “If you look at a refinery over the last 40 years, the number of people running it has dropped by about 80%. This means a board operator may be looking after a unit with 1,000 control loops and the only way they can manage that is through automation and digitization. Operators need more insight and better understanding due to the shifting workforce,” says Joe Bastone, Honeywell growth initiative leader.

“And the next evolution is AI, which offers guidance to newer operators with less experience, but who also expect to have tools, and are willing to use them to achieve their objectives.”

A recent addition to Honeywell’s Experion PKS lineup is Experion Cognition, which “gets the right information in front of the right people at the right time to help them make decisions,” says Bastone. “For example, if a pump is on the edge of causing failure as indicated by a high bearing temperature, Experion Cognition uses AI, analytics, past events and actions taken to provide a process that the operator can follow, such as shutting off the primary pump and starting the bypass pump. The system provides guidance so operators can be proactive to avoid a problem, rather than being reactive after the pump fails.”

Experion Cognition is based on a rich history of events and collected data about common situations and pulls the information together to develop solutions, so that every operator can respond in a consistent manner. This empowers them to be more responsive to the process, avoiding delays and upsets.

Eschbach GmbH (Boston, Mass.; eschbach.com), the provider of Shiftconnector enterprise manufacturing software, added Shiftconnector Artificial Manufacturer Intelligence (SAMI) to enable chat to change the way operators interact with existing system knowledge, says Andreas Eschbach, founder and CEO.

Operators can use SAMI to troubleshoot issues and find solutions that have been successful in the past and for continuous improvement where users may want to oversee a timespan of years. “The SAMI smart search functionality enables semantic search requests on large amounts of data to quickly find solutions that may be buried under different terms.

“For example, an operator may notice that the viscosity of the raw materials is impacting production, but you’ll very rarely find the word ‘viscosity’ in shift reports; however, you may find ‘creamy’ or ‘milky.’ Our AI-enabled semantic search function can make the connection and uncover problems and solutions that were recorded weeks, months or years before,” he explains.

“This allows users to ask questions and receive answers, simplifying and reducing the time it takes to troubleshoot issues and find solutions,” says Eschbach. “As baby boomers retire, new operators will benefit from the collected knowledge and documentation.”

In a drive to offset workforce shortages and move facilities toward more autonomous systems, Siemens added AI to its Xcelerator Digital Twin platform, “paving the way for autonomous systems,” says Moors. “Our Industrial Copilots function as AI-powered assistants, orchestrating workflows across operations, progressing through three levels: as ‘colleagues,’ providing recommendations; as ‘task executors,’ completing entire engineering tasks; and as ‘orchestrators,’ leading to (semi-) autonomous systems.”

For example, Siemens’ Copilot SIMATIC eaSie now incorporates generative AI for natural language chat or voice interaction for plant data insight, enabling technicians to access plant data through chat interactions and functioning as a digital knowledge repository for instant best-practices access and problem solving (Figure 2). Its gDAP Troubleshooting Assistant processes extensive logs and generates actionable recommendations, reducing troubleshooting time from days to minutes.

And, enhanced gPROMS models now incorporate machine learning for semi-batch control, autonomously optimizing feedrates and temperature while predicting batch outcomes. “This delivers early termination of failing batches, autonomous process adjustments, maximized yield and reduced human intervention,” says Moors. “Advanced applications include multi-agent systems where intelligent software agents collaborate to manage material handling, logistics and production scheduling autonomously, demonstrating the path toward fully autonomous plants.”

Efficiency and optimization

Arul Jothilingam, systems consultant with Yokogawa (Tokyo, Japan; yokogawa.com), which offers a portfolio of software products that span plant operation and control, process optimization, operator training, asset and process monitoring and batch management, agrees that advanced features lay the foundation for autonomous plant operation and says these features enable safer and more optimal plant operation by supporting knowledge capture and retention.

Jothilingam says the integration of AI and ML enables predictive analytics, anomaly detection and self-optimizing control strategies for smarter, more efficient plant operations. “For example, adding AI and machine learning to Yokogawa’s Advanced Process Control (APC) solutions improves quality prediction and adapts models to frequent changes. In refineries, AI-enhanced APC during crude switching predicts product-quality shifts more accurately, minimizing off-specification blends and reducing switching time.

“Intelligent manufacturing features unify data from diverse OT and IT sources to provide key performance indicator (KPI) dashboards for decision-makers, automate reporting with robotic process automation and streamline workflows across departments,” he explains. “A chemical plant might use intelligent manufacturing to combine batch records, quality results and ERP data, streamlining compliance reporting and empowering managers to quickly resolve bottlenecks.”

AI is also helping operators without advanced science degrees leverage the power of advanced analytics so they can do more at a faster pace while expanding their skills, explains Seeq’s Armstead-English. “The Seeq AI Assistant guides users as they build advanced analyses, operationalize machine learning and more,” she says. “The result is accelerated insights, improved decision-making and gains in operational excellence and sustainability” (Figure 3).

She provides an example: A specialty chemicals company used Seeq to address recurring fouling issues in its distillation towers and improve overall operational efficiency. By leveraging the platform’s multivariate modeling capabilities, a wide range of operating parameters, such as composition, temperature, pressure and flow data, were analyzed to detect and predict fouling. This enabled the team to move from reactive responses to a proactive, data-driven approach to equipment performance.

“The platform quickly provided actionable insights, identifying multiple contributing factors to fouling and clarifying how these factors interacted,” she says. “Plant personnel received high-level summary metrics and detailed trend views, along with recommended maintenance actions. Through advanced signal selection and importance ranking, the user pinpointed which variables mattered most and established casual relationships, enabling deeper understanding of the dynamics of fouling events.”

Over time, she continues, the model was refined to forecast fouling conditions up to two months before anomalies occurred. “This predictive capability allowed the company to schedule maintenance in advance, minimize unplanned downtime and extend runtime between cleanings,” she says. “The company also implemented operational changes that boosted efficiency and profitability through longer production runs and reduced maintenance interruptions.”

Flexibility and sustainability

“Fluctuating raw-material availability, tightening environmental regulations and shifting customer expectations make it difficult for processors to maintain efficiency and profitability,” notes Heather Lawson, senior product marketing manager with Emerson’s Aspen Technology business (Bedford, Mass.; emerson.com). “AspenPlus, a simulation software used by chemical processors, empowers engineers to model complex chemical processes with precision so users can rapidly evaluate process alternatives to reduce costs, energy use and emissions — advancing operational efficiency while aligning with sustainability objectives.”

The integration of AI and ML with simulation tools has enhanced engineers’ ability to develop precise models faster, enabling them to shorten design time, troubleshoot bottlenecks and improve process efficiency, she says.

“Aspen Hybrid Models, offered in Aspen Plus, combine the power of data, AI and engineering first principles to create models that reflect plant performance and provide operators with timely, actionable insights. By grounding AI in first principles, such as mass balance, energy balance and heat transfer, process engineers establish guardrails to ensure model reliability, while reducing the volume of plant data needed for analytics algorithms to deliver accurate predictions” (Figure 4).

Lawson continues: “The benefits of Hybrid Models have also been extended into production planning. They help planners respond to changes in feedstock and demand, improving agility and decision-making across the value chain,” she says. “As AI evolves, Hybrid Models will play a critical role in driving smarter, more adaptive operations across process industries.”

And for refiners wishing to evaluate the impact of processing various feedstocks, KBC Advanced Technology’s (Sugar Land, Tex.; kbc.global) Petro-SIM process twin software with Reactor Suite can help, says Rodolfo Tellez-Schmill, technical presales senior consultant.

The company recently launched a renewable diesel model so users can evaluate the impact of processing various bio feedstocks in the refinery, whether as co-processing or a dedicated option. The reactor models include detailed catalyst activity models to account for fresh catalyst properties, feed contaminants, catalyst deactivation and catalyst makeup. Users can track mass balances, heat balance closures and visualize unit KPIs and actual yields versus simulated on dashboards.

The Reactor Suite diversifies from refinery products to olefins, polymers and a wide range of petrochemicals, allowing the creation of completely integrated site-wide models of unlimited complexity.

“Advanced analytics and AI/ML improve planning and operations and de-risk investments,” says Tellez-Schmill. “The technology bridges the knowledge gap for non-technical users to provide a trustworthy indication of model performance as that helps them take risks for quick gains. Machine learning allows models to run faster, which is important because customers expect high-performing tools that deliver fast and accurate results.”

And, when exploring experimental designs to find the optimal process conditions, Intellegens’ (Cambridge, U.K.; intellegens.com) experimental design software with ML can help (Figure 5). “In this role, machine learning teaches computers how to perform the statistics so that process engineers don’t need a deep understanding of them,” says Tom Whitehead, head of machine learning. “This saves process engineers 50 to 80% of the experimental effort in determining the best approach.”

“For example, we had a company working with biopolymers. They wanted to know if they could substitute raw materials, such as chemical-based biopolymers for bio-based biopolymers,” Whitehead explains. “Using our software, they found they could. They were able to specify which performance criteria they must meet and which can flex and that gave them room to optimize and find the best combination of biopolymers and the best way to treat them.

“Computers excel at crunching numbers and machine learning teaches computers to understand the relationships between what’s important and how to optimize it,” continues Whitehead “This saves time and effort and streamlines the design process because process engineers can set the objectives for things like yield and efficiency and the machine-learning approach does the heavy lifting, so users get results as quickly as possible.”

Joy LePree