Imagine how much simpler car ownership could be if you knew exactly when to change the oil or replace the timing belt based on your car’s actual condition. Instead of following the manufacturer’s guidelines, you could go to the mechanic only when needed.
Condition-based maintenance isn’t here yet for the personal automobile, but it’s coming to the industrial boilers, chillers, and other equipment that keeps large enterprises running. IBM is integrating AI agents into its Maximo Application Suite to make tracking the condition of physical assets easier, more efficient, and potentially more economical.
Maximo was developed nearly 40 years ago to help enterprises manage their physical assets, from their initial purchase to ongoing maintenance, disposal, and eventual replacement. In recent years, sensor and visual inspection data have been pulled in, giving data scientists the ability to monitor the condition of individual machines and detect failures before they occur.
Maximo is now rolling out several new features, born out of IBM Research, that will allow more people to access this data and manage their assets more effectively.
“Now operators, reliability engineers, and technicians can interact with the AI directly and do their jobs much more efficiently,” said Anuradha Bhamidipaty, a distinguished engineer at IBM Research co-leading the project. “It’s a huge transformation in how clients not only monitor and maintain their systems but plan their investments to replace them.”
Maximo Assistant, a new LLM-powered interface, was recently integrated into the platform. Users can now query Maximo’s databases through a chat window, and an IBM Granite-powered agent will fetch the information. Users no longer have to call APIs or know how to code in SQL to interact with Maximo’s relational databases.
In July, the SaaS platform will add an optimizer to help users calculate, based on budgeting constraints and the condition and performance of their assets, when they should ideally be replaced. By the end of the year, a second agent, Maximo’s Condition Insights agent, will be added to gather statistics and operational data for each asset a company tracks. The agent will estimate the asset’s current condition and projected replacement date as part of a broader move from regular check-ups to more sporadic visits — only when Maximo detects signs of trouble.
By shifting the focus from corrective maintenance to servicing only when needed, researchers expect that Maximo’s AI capabilities can reduce unnecessary labor, help enterprises meet their sustainability goals by keeping machines at peak performance, and ultimately, extend the useful life of expensive, multi-year investments.
As an enterprise scattered across nearly 600 sites, IBM itself knows something about managing physical assets. The company’s global real estate (GRE) division spent $1.5 billion last year on leases and building maintenance, which extends to more than 115,000 pieces of industrial equipment — things like chillers, generators, and boilers.
From IBM’s Poughkeepsie office, Sal Rosato and Kathleen Kennett lead the respective technology transformation and engineering arms that keep IBM’s physical operations running smoothly. For years, they and their teams have used Maximo to help track, service, and plan for replacing industrial equipment across IBM.
They have been among the first to test new Maximo capabilities. Under a long running “smarter buildings” project, they installed automated fault detection in machines across 30 IBM facilities. Teams of technicians and engineers worked together to identify points of failure for each asset and manually develop algorithms to detect them. In the end, GRE documented about $1 million a year in energy savings.
A new pilot study is now underway. This time, its Maximo’s Condition Insights agent that’s being put to the test, and the process is expected to go faster and perhaps deliver even more value. The agent’s first assignment is to monitor a $2 million chiller that provides essential cooling for offices, data centers, test labs, and manufacturing areas at IBM’s Poughkeepsie site.
Like fans and other machines with rotating parts, chillers have ball bearings to reduce friction. As they start to wear out, they can emit a faint vibration. “These vibrations aren’t detectable by humans, at first,” said Rosato, the technology transformation manager for GRE. “But if an AI is monitoring them, it can flag the anomaly and predict when a failure is likely to occur. We can then schedule a shutdown to reduce the chance that the issue becomes an expensive, time-consuming failure.”
Maximo currently tracks each machine’s make and model, its expected life span, maintenance history, and all the things that could potentially go wrong with it. It also receives a firehose of sensor data, which in the case of a chiller, includes vibrations. The job of Maximo’s Condition Insights agent is to consolidate these data silos into one for a holistic assessment of an asset’s health.
The proverb often attributed to Benjamin Franklin, “prevention is worth an ounce of cure,” could be thought of as the guiding inspiration for Maximo’s Condition Insights agent.
“The typical chiller lasts about 40 years,” said Roman Vaculin, a principal research scientist at IBM Research co-leading the project. “If the AI agent senses that efficiency is declining or the machine is consuming too much power, that could signal a failure mode. Technicians and engineers can act on these insights before it turns into something serious.”
AI agents for greater automation
The agent underpinning Maximo Assistant will be upgraded to a Granite 4.0 model in the near future. Equipped with a native understanding of Maximo’s query syntax, the new assistant will provide a smoother user experience. As more specialized agents are added to the SaaS platform, Maximo Assistant will continue to evolve, to orchestrate multiple agents and workflows to take appropriate action.
IBM’s time-series foundation models will also be integrated into Maximo to help the Condition Insights agent pick out trends and other meaningful patterns in sensor data.
“These machines have been producing and collecting data since they went live, but you had to know how to analyze time series data and combine those insights with other structured and unstructured data,” said Bhamidipaty. “The Condition Insights agent will soon be able to do all that for you.”
Next year, researchers plan on introducing a third agent focused on asset investment planning. The agent will go beyond Maximo’s default optimizer to allow users to set the conditions, whether that’s operating costs, budgeting constraints, or sustainability targets, for replacing a piece of equipment.
To make all this happen, IBM researchers created AssetOpsBench, the first agentic environment and benchmark of its kind to guide the development, orchestration, and evaluation of specialized agents for asset management tasks. IBM Research recently open sourced the environment so that Maximo partners and collaborators can now develop and put their agents to the test, too.
“Our intention is for the community to improve on existing asset lifecycle maintenance agents and build new agents that can be quickly validated across more than 140 industrial test scenarios,” said Jayant Kalagnanam, director of AI applications at IBM Research.