Are LLMs Ready for Industrial Automation?

September 1, 2026

Peter Welander

See what generative AI, such as GPT-5 and other tools, can accomplish in different industries

Image Source: Pixel World/Stock.adobe.com; generated with AI

By Peter Welander for Mouser

Published September 2, 2026

Artificial intelligence (AI), and more specifically generative AI (GenAI), is becoming common in everyday life. Those same technologies are starting to find a place in industrial automation, as well. Large language models (LLMs) are critical to this development because they allow people to interact with complex information using ordinary language. To understand what that could mean on the factory floor, it helps to start with a simple explanation of how they work.

LLMs are deep learning models trained on enormous collections of text. During training, they learn statistical relationships among words and other pieces of language, which allows them to recognize patterns and generate responses based on context. Ask an LLM to write a love poem, for example, and it will not retrieve a poem it has seen before. Instead, it generates new text based on patterns learned during training. The same basic capability allows an LLM to summarize a technical document, explain a piece of software code, answer questions about information it has been given, or help turn written requirements into programming instructions.

Those capabilities open up interesting possibilities for industrial automation. Manufacturing environments generate enormous amounts of information, from operating records and maintenance histories to engineering documentation and control code. Engineers and operators must find, interpret, and act on that information. An LLM provides a new way to interact with it, potentially reducing some of the work required to move from a question or engineering requirement into useful technical information.

What might that look like in practice? This article addresses that question by examining two hypothetical examples: an oil refinery and a packaging line.

Using LLMs to Unlock Industrial Knowledge

One way to understand how LLMs could fit into industrial automation is to assess a situation in which the information needed to make a decision already exists, but finding and connecting it takes time. Such a situation may exist at an oil refinery (Figure 1).

Figure 1: Can an LLM help a refinery adjust when a new and different load of crude oil hits its crude distillation unit? (Source: Aleksej/stock.adobe.com)

Consider a mid-size oil refinery that processes approximately 250,000 barrels per day. The refinery is independent, so it is not tied to any specific crude oil source and, therefore, shops for the best prices. The result is a high degree of variability in the composition of “opportunity crudes.” Consequently, after optimizing the crude distillation unit for one kind of oil, the refinery must switch to run a few million barrels of a bargain offering. Now the balance of what comes out of each fraction changes, forcing operators to re-optimize the subsequent fluid catalytic cracking (FCC) unit, alkylation unit, reformer, etc. These changes take time, and processes lose efficiency until they are restabilized.

So how do the operators minimize such disruptions? The refinery’s lab routinely analyzes incoming crude for characteristics such as corrosiveness and contamination. Over time, those results create a history of the different feedstocks the refinery has processed. When a new shipment arrives, operators can compare its characteristics with those of similar previously used crude processes, then turn to operational records to see what happened during those earlier runs and how the refinery responded.

The information may already exist. The challenge is finding the right information quickly and connecting data that could be spread across lab results, operating histories, procedures, and other plant documentation.

An LLM could help in this case. Instead of controlling the process itself, an LLM could provide a natural-language interface to approved plant information. Using an approach such as retrieval-augmented generation (RAG), the system could pull up information from designated internal sources and use it to answer an operator’s questions.^[1]

For example, an operator might ask which previous crude shipments had characteristics similar to the one arriving. The LLM could retrieve relevant lab results and operating records to help determine what problems occurred during those runs and what adjustments were made. Instead of manually searching multiple systems and years of records, the operator gets a starting point for evaluating the new feedstock.

The available information is not limited to the refinery itself. The global energy industry has a wide range of technical information on crude sources and characteristics. If appropriate sources were authorized and connected to the system, an LLM could retrieve external information, and the combination could provide engineers with additional context when they encounter unfamiliar feedstock.

It’s important to make the distinction that the LLM is not replacing the distributed control system (DCS), nor should its response automatically determine how the refinery operates. The DCS continues to perform process control while operators and engineers evaluate the information and determine what changes are appropriate. In this example, the value of the LLM lies in its ability to help operators tap into years of knowledge when they need it.

Applying an LLM to Packaging Line Design

Another way to look at LLMs in automation is to take into account what happens when engineers are building something new rather than searching through what happened in the past. For example, consider a personal care products manufacturer’s packaging line that must accommodate three bottle sizes, with some products using conventional screw caps and others using pumps. The line also needs to be flexible enough to be reconfigured for other products in the future. The company has experience designing automation systems and is considering whether to handle more of the engineering and integration work in-house instead of using an outside original equipment manufacturer (OEM) or system integrator. Such a packaging line (Figure 2) requires a wide variety of components related to liquid handling, conveyors, filling and capping stations, and labeling stations, all of which must operate as a coordinated system.

Figure 2: How does an LLM learn about something as complex and unique as a packaging line? (Source: nordroden/stock.adobe.com)

Asking an LLM to design the entire packaging line would be unrealistic, considering what it is and is not capable of doing. An LLM can be fed vast amounts of product information from a range of component manufacturers. It can also look at the company’s internal data, examining and learning from all the existing lines. Still, while the LLM may understand the mechanics and sizing of bottle conveyors or cap mounters, it may not be able to learn enough about how these elements are integrated into a single operating unit. Additionally, there may not be enough information about packaging lines to serve as a guide.

The LLM’s more immediate value could be helping engineers with specific tasks within that larger design process, such as control programming. Engineers usually begin with requirements that describe how the line should operate (e.g., detect a bottle, position it, dispense the correct quantity of a product, apply the appropriate cap or pump, inspect the finished package, and move it to the next station). These requirements eventually have to be implemented as control logic.

LLMs have demonstrated skills for writing code to generate IEC 61131-3 Structured Text, one of the programming languages used for programmable logic controllers (PLCs).^[2] An engineer could describe part of an operating sequence in natural language and use an LLM to generate the first version of the corresponding control code. The model could also help explain existing code, create documentation, or assist with any adjustments when the line needs to be reconfigured for a new product.

However, research into LLM-generated code also shows why human oversight remains necessary. Syntactically correct code may fail to account for real-world timing, equipment states, interlocks, faults, and safety requirements. For that reason, LLM-generated control code must pass through the same engineering safeguards that are applied to other control software.^[3] Approaches that combine LLM generation with compilation, verification, and iterative correction are already being explored to improve the reliability of AI-assisted PLC programming.^[4]

Therefore, if the LLM can turn an engineer’s requirements into a useful first draft of control code, explain unfamiliar logic, or reduce the work involved in documentation and reconfiguration, it becomes another engineering tool, not an autonomous system designer.

The Future of LLMs in Industrial Automation

The potential for LLMs in industrial automation is growing, but so is the need to understand their limitations. These models can produce convincing responses even when information is incomplete or incorrect, a problem known as hallucination. Therefore, LLM-generated control code requires engineering review, testing, and validation, especially when errors could affect physical equipment.^[5] The closer an AI-generated output gets to controlling machinery or processes, the more important these safeguards are.

So, are LLMs ready for industrial automation? In many supporting roles, they already show great potential. They can help engineers find information, interpret technical records, generate first drafts of code, and reduce time spent on repetitive tasks. However, they are not ready to replace the deterministic control systems, safety mechanisms, and engineering judgments on which industrial operators depend. For now, the most useful role for an LLM may be alongside the engineer rather than in place of one.