What Does It Take to Deploy Autonomous Control in a Chemical Plant?

The only certainty in the chemical industry today is the pressure to do more with less.

Global instability has shaken markets that were already contending with tightening regulations, rising raw materials costs and fewer skilled workers.  

Increasingly, company leaders view autonomous control as a viable solution.

In a global survey published by Schneider Electric in March, 44% of senior executives in the energy and chemicals sector said implementing autonomous operations will be a critical priority for their organizations over the next decade. Achieving autonomy is especially urgent in North America, where nearly 90% of executives called it a high priority for the next five years.

Standing still isn’t an option. In the survey, many executives warned that delaying adoption of autonomy could ratchet up operating costs, exacerbate talent shortages, undermine workplace safety and put their companies at risk of falling behind more advanced operators.

“Autonomous technologies are increasingly viewed as enablers of operational gains on multiple fronts, underpinned by electrification and digitalization,” Gaurav Sharama, an independent energy market analyst, asserted in the Schneider Electric report. “Long-established industrial platforms are fast becoming truly autonomous — driven by artificial intelligence, machine learning, advanced analytics, predictive maintenance and digital twins.”

As the industry moves toward autonomous operations, most chemical plants aren’t starting from scratch. They’re building on decades of investment in distributed control systems, advanced process control and other technologies that provide a solid foundation for more autonomous control.

“Advanced process control is an important enabler for more autonomous operations,” said Randhir Shetty, product marketing director for Emerson’s Aspen Technology business. “It allows operations to sustain peak performance despite constantly evolving plant conditions.”

Digital twins — which McKinsey & Company calls “the next frontier of factory optimization” — are becoming another key enabler of autonomous operations. While modeling and simulation are nothing new in manufacturing, today’s digital twins are more dynamic than previous iterations, incorporating real-time data, IoT connectivity and advanced analytics.

“Digital twins play a crucial role because autonomy depends not only on access to data, but on understanding that data in the right operational context,” said Axel Lorenz, CEO of process automation for Siemens. “They support more accurate predictions, more robust decision-making and increasingly closed-loop optimization within defined boundaries.”

Digital twins can be applied in different ways to support the shift toward more autonomous operations, Lorenz explained. For example, a “decision twin” combines real-time data and predictive insights to guide operators and enable increasingly autonomous decision logic, while an “asset twin” supports predictive maintenance and asset-health monitoring. A “production twin” provides a clear view of how processes and workflows connect across the plant.

“Together, these complementary perspectives create a holistic digital representation of the plant that supports more informed decisions and progressively more autonomous operation,” Lorenz said.

A Layered Approach to Autonomy

Evonik, an Essen, Germany-based chemical manufacturer, recently announced plans to collaborate with Siemens on next-generation autonomous technologies for chemical facilities. Rather than building autonomous operations from the ground up, the companies are developing what they call a “digital plant technology stack” that layers AI-enabled capabilities onto existing control systems.

At the center of that model is a shared data environment designed to bring control systems, digital twins and other applications into a common framework. The goal is to ensure that all systems operate from a unified view of plant operations — and that personnel have access to actionable, real-time data.

“This shared foundation enables existing systems to benefit from new capabilities, including AI-supported analytics, advanced process control, operator-assistance functions and agent-based AI systems,” Lorenz explained.

Building on that foundation, AI-supported analytics can be used to detect anomalies, generate predictions and optimize performance, according to Lorenz. At the same time, more structured forms of automation — such as state-based control and procedural workflows — can help manage plant operations across different conditions, from startup and steady-state production to grade changes and shutdown.

Lorenz also noted that edge computing plays a key role by enabling data processing and the execution of analytics and automation services closer to the process.

While many of these technologies are already in use, applying them at scale across entire facilities remains a work in progress. Evonik and Siemens are validating “integration patterns” in pilot environments before scaling to broader production systems, according to Henrik Hahn, chief digital transformation officer at Evonik. In this approach, the companies are activating new autonomy capabilities “in observe/advisory modes, then graduating to closed‑loop actions within defined safety boundaries.”

Hahn noted that the “preferred validation route is a major production site that represents a plant type we operate in multiple, comparable installations worldwide.”

“Lessons learned harden the reference architecture and operating playbooks before broader scaling,” he added.

Validating Autonomy in Real-World Operations

When Yokogawa in 2020 approached ENEOS Materials Corp. about validating an AI-driven control protocol it had been developing, ENEOS production managers quickly recommended a butadiene distillation column at the ENEOS plant in Yokkaichi, Japan. The plant’s control systems were struggling to regulate two critical valves that stabilize liquid levels in the column, shifting much of the burden to operators and making the site an ideal candidate for this type of pilot project, according to Yokogawa.

The goal of the field test was to demonstrate that the protocol – a reinforcement learning-based AI algorithm called factorial kernel dynamic policy programming (FKDPP) – could directly control operations at a plant. To execute the pilot project, Yokogawa and ENEOS used a three-step process:

1. Develop a plant model and train the AI. Engineers built a digital model of the plant using a simulator, allowing FKDPP to self-learn and generate an initial AI control model.
2. Validate and refine the model. The AI recommendations were evaluated using historical and real-time operating data, with operators applying the suggested control actions through the DCS to test performance and stability.
3. Deploy in live operations under safety constraints. After confirming reliability and safety, the AI model was integrated into the plant and used to control operations, with safeguards and fallback procedures in place.

For 35 consecutive days in January and February 2022, FKDPP successfully controlled the distillation column, which Yokogawa touted as a “world’s first.” A few weeks after the initial deployment, ENEOS and Yokogawa took advantage of a scheduled maintenance shutdown to refine the AI control model, using operational data and feedback from plant personnel.

Over the following year, the AI system autonomously managed the two critical valves, helping stabilize butadiene distillation while maintaining product quality and reducing energy consumption. Today, ENEOS continues to use the AI-based autonomous control system as a routine part of operations at the Yokkaichi facility, and plant engineers are exploring opportunities to apply the technology to other areas at the site.

From Pilot to Production: Where Autonomy Hits a Wall

Even as pilot projects demonstrate the viability of autonomous control, scaling those capabilities across multiple sites can bring on a new set of challenges.

In the Schneider Electric survey, executives in the energy and chemicals industry cited high upfront investment costs as the leading barrier to wider adoption of autonomous control. Senior leaders in North America are especially sensitive to cost concerns, with nearly 60% of U.S. executives citing upfront investment costs as a top barrier to autonomous deployment.

Connecting AI-driven autonomous control technologies with legacy infrastructure is another barrier that industry leaders frequently cited. While integrating AI capabilities into a single unit or facility might seem manageable, replicating that approach across sites with disparate control systems, configurations and operating practices is an entirely different challenge. As a result, chemical producers such as Evonik are focusing on pilot projects in plant types that can be standardized and repeated, rather than attempting enterprise-wide rollouts all at once.

Not all the barriers to autonomy are technical. In the Schneider Electric survey, 27% of executives said resistance to organizational change is a key obstacle. Moving from pilot to production requires coordination across plant teams, corporate leadership and, in some cases, unions — particularly when new technologies begin to change day-to-day workflows. Without organizational alignment, even successful pilots can stall before broader deployment.

It’s a challenge that’s top of mind for Evonik’s Henrik Hahn.

“Integration is as much organizational as it is technical,” said Hahn in an email interview. As Evonik and Siemens develop a new technology stack that supports autonomous operations, they’re also reimagining “the organizational framework — future roles, skills and responsibilities — so operators and engineers remain in control and focus their time on expertise, problem‑solving and quality as autonomy expands,” he added.

Collectively, these dynamics highlight a broader reality: While many of the building blocks of autonomous operations are already in place, scaling those capabilities across complex, multi-site environments won’t happen overnight. For many chemical producers, autonomy will be an evolution that unfolds as technology, operations and company cultures align.

“Evonik and Siemens believe that autonomy is a progressive, structured journey rather than a ‘big bang’ transformation,” Hahn said. “The autonomy vision becomes achievable only through strong partnerships that combine deep operational know‑how with automation, digitalization and AI expertise. This is exactly why Evonik and Siemens are pursuing this work together: Autonomy requires the integration of technology, process knowledge and organizational readiness.”