Rockwell Says Factory AI Cut Downtime 33% and Will Reach More Plants

Rockwell Automation says its AI maintenance assistant has cut machine downtime by 33% at its Singapore factory, where technicians have used it since October 2025. The company now plans to bring it to plants in Ohio, Poland and Mexico. That expansion puts a practical question behind the reported gain: can a tool built around one factory’s experience help crews elsewhere?

A repair guide built from people and records

Before using the assistant, engineering assistant Mangleswaran Mahalingam searched lengthy manuals for error codes, then sought out a colleague if the suggested fix failed. Now he can type a question on a tablet. That change is modest on its face, but it gives technicians a way to consult past repairs without first finding someone who remembers them.

Rockwell’s in-house GenAI-Powered Maintenance Copilot draws on manufacturing-software data, machine manuals and a database created by veteran engineers. It organizes their knowledge by symptom, cause and response. When a machine shows an error, technicians can see likely causes, read colleagues’ comments and check whether the problem appeared on another production line. They can also ask questions about a manual and get a step-by-step answer.

The assistant runs on GPT-5.4-mini and was built with Microsoft Azure and Azure OpenAI in Microsoft Foundry. Its useful feature for a technician is not the model name; it is the ability to bring scattered instructions and prior experience into a single troubleshooting exchange.

The gains Rockwell is counting

Singapore plant director Li Wang attributes a 33% drop in machine downtime to the more targeted, consistent approach. Rockwell’s internal tracking also puts servicing and spare-parts costs down by about 25%. These are company-reported results from the Singapore site, not measurements of how the assistant will perform at every plant.

Rockwell also estimates that new workers learn to troubleshoot hundreds of machines in three months rather than nine, according to chief supply chain officer Bob Buttermore. Faster training addresses a different problem from machine downtime: how long a new technician needs before taking on complex faults. The two outcomes matter together if a factory needs capable workers on every shift.

Why the workers’ knowledge matters

The veteran engineers behind the database each have 20 to 30 years of shop-floor experience. Wang says Rockwell wants to keep that knowledge from leaving when workers retire. She also wants technicians to follow a standard approach to faults regardless of who is on duty. The assistant is therefore meant to preserve hard-won experience and make it usable by people who did not acquire it firsthand.

That does not make the copilot responsible for every improvement at Singapore. Wang identifies it as one of several AI-driven capabilities at the site, alongside a system for detecting assembly-line defects and another for predicting maintenance needs. Keeping those tools distinct matters when judging which results the maintenance assistant itself can reproduce.

A local system heads to other plants

Buttermore says the maintenance copilot will be rolled out in Twinsburg, Ohio, and at plants in Poland and Mexico. Singapore serves as a model site for Rockwell’s other operations and for customers; the company has more than 25 plants worldwide. Rockwell’s stated aim is to test approaches in its own facilities before offering lessons to other manufacturers.

The expansion will test more than whether the software can run in another location. The Singapore assistant depends on documented experience, machine information and manuals that technicians can actually use when a fault occurs. Rockwell has shown how it assembled those ingredients at one factory and reported gains there. Whether the same approach delivers similar results at the next three plants remains the open question.