How Does AI Understand Facilities?

[Cengiz Özemli]

​

## How Does AI Understand Plant Operations?

Operational technology has reached a critical juncture. In recent years, process manufacturers have leveraged the decreasing cost and rapidly increasing capacity of sensors to monitor more plant areas than ever before.

The new generation of sensors generates very large datasets, both structured and unstructured, much faster than OT personnel can manually evaluate. OT teams have discovered the value of data; for a long time, plant employees were concerned about whether data collection was being done excessively and about its management. These concerns have changed as artificial intelligence offers ways to transform this big data into meaningful and usable information.

Today, numerous AI applications successfully operate on large volumes of data. However, for OT teams, in an environment where uptime, risk reduction, deterministic processes, PID control loops, and first-principle methodologies are paramount, the intersection of big data and AI presents new challenges.

### Limitations of Artificial Intelligence

AI's recommendations, if not bound by constraints such as thermodynamic laws, equipment design limits, mass and energy balances, supervisory parameters, and safety mechanisms, can violate process reality. Incorrect or misleading advice can erode operator trust, delay decision-making, and pose risks in a low-latency critical environment.

OT teams will increasingly need AI to amplify the insights it provides and multiply innovation opportunities.

### Contextually Relevant Artificial Intelligence

When an operator asks an AI consultant for general information about the previous shift during a shift change, a well-tuned AI can provide meaningful answers regarding production rates, alarm sets, or quality trends. However, out-of-scope data can cause confusion, increasing information clutter and reducing trust in AI.

### Industrial AI Features

• Based on first principles and immutable constraints, taking into account physics, chemistry, equipment design, and control limitations.
• Recommendations offer feasible, meaningful, and budget-appropriate solutions.
• User-based; provides different recommendations and information for operators, maintenance engineers, and process engineers.
• Considers process hierarchy and is open to interpretation with appropriate language and threshold values based on roles.

### Foundation of Data-Driven Automation

Industrial AI can utilize raw and unstructured data, but its impact multiplies when supported by contextual data. Therefore, many organizations are building their automation infrastructure with a data fabric and an enterprise operations platform that ensures seamless data mobility.

The most advanced automation providers are integrating AI into these platforms, offering seamless and scalable solutions based on first principles and industry knowledge.

### Preparing for the Future

As AI solutions evolve, now is the perfect time to prepare to gain a competitive advantage. A new industrial era is on the horizon, and those who invest in this field will be pioneers of future success.