🚀 [B]The Use of Artificial Intelligence in Large-Scale Control System Migrations: Efficiency and Opportunities![/B]

[Ahmet Ö.]

💡 Automating Repetitive Tasks​

Large-scale control system migrations create a large number of repetitive engineering tasks, such as thousands of control loops, legacy files, descriptions, graphics, and configuration records. Many of these tasks involve processing large amounts of data with repeatable steps, making them practical use cases for AI.

📝 Converting Legacy Data to Control Descriptions​

In a recent migration project, over 1,000 control descriptions were required, each detailing how a control loop should operate. The necessary data was already available in the legacy system but needed to be extracted, organized, and transferred into a consistent set of Word document templates. This work was previously done with Excel macros, which are effective but slow to build. In this project, an AI model was given the templates and a table of extracted legacy system parameters, then asked to write a script that would automatically populate the templates.

The use of AI reduced the time to develop the script from an estimated one or two days to minutes. The engineering team reviewed the script, ran it against the project data, and performed a back-check to verify that the information was correctly transferred. The same approach also facilitated large-scale revisions. Adding a field to every description or changing words required a new script instead of hundreds of manual edits.

This use case was successful because the task had clear inputs, a consistent template, and an output that the engineering team could verify. The AI model was not determining how the loops should operate; the engineers provided the source information, built the template, specified the desired changes, and checked the final output.

🔍 Making Legacy Information More Accessible​

Control system migrations often involve equipment that has been in service for decades, and documentation may be missing, out of date, or only available as a scanned copy of a printed manual. In practice, this can mean a folder of technical information becomes a folder of images. An engineer can review page images one by one, but cannot use a standard keyword search to quickly find a specific term.

An AI model that reads scanned pages can convert them into searchable information, allowing engineers to find specific terms or hardware references more quickly. This does not solve every document problem, as an old manual may still contradict the installed system, but engineers spend less time looking for information and more time evaluating what they find.

💻 Working with Text-Based Control Files​

Many control system platforms allow logic or configuration data to be exported into formats that can be parsed and modified with scripts, such as XML or HTML. When a required change is consistent and well-defined, AI can help the engineer develop a script that applies it across hundreds of files, using a representative section of the exported logic to show the model the existing structure and the intended change.

Using an AI-generated script across many files can also improve standardization in a multi-engineer migration. When work is divided among a team, variations can emerge in logic, tags, and documentation. A common script can reduce this variation and make the finished work easier to review and maintain.

The use of AI becomes less reliable when the task requires it to interpret large amounts of legacy information rather than apply a defined change. AI output is not always correct. When a team feeds an entire legacy database to a model and asks it to identify required parameters, the model may return extraneous information, misunderstand how values are used, or omit important context. AI is not yet reliable enough to take control logic from one platform and automatically convert it into working logic for another platform without significant engineering review.

AI has similar limitations when creating operator graphics. While an AI-generated screen may contain all the desired information, it can still be cluttered, difficult to navigate, or inconsistent with operator expectations. Designing an effective operator interface still requires engineering judgment, process knowledge, and an understanding of how operators use the system.

🛠️ How Do You Incorporate AI into a Migration Workflow?​

Teams incorporating AI into a migration project should start with a small, measurable task and a limited example to avoid spending ten hours automating a five-hour job. Teams should clean and standardize source data before using AI, as two pieces of code may perform the same function but use very different structures, which can cause the model to treat them as separate use cases.

Customer-specific information should be removed from project material before using an online AI model. When possible, engineers can also ask the model to generate a script rather than make changes directly in the project data, then run that script offline against the actual data.

AI works best in control system migrations when the task is narrow, repetitive, and reviewable. The most useful applications allow AI to handle the work at scale, while control engineers remain responsible for the data, verification, and final technical decisions.