Equipment failures have long haunted manufacturers — and they come at a staggering price.
A 2024 Siemens report estimated that unplanned downtime costs the world’s 500 biggest companies up to $1.4 trillion a year combined. Breakdowns in machinery can then affect consumers through product shortages, higher prices, and lower-quality goods.
Major companies like IBM and Amazon have long relied on a strategy known as predictive maintenance to try to keep their industrial operations running smoothly. Predictive maintenance tech combines hardware and software to monitor the health of industrial machinery and identify issues, such as overheating components or pressure leaks, before they escalate. Sensors collect performance data from equipment, and software then analyzes that data to pinpoint problems.
The need for the technology is expected to rise: Fortune Business Insights projects the global market for predictive maintenance to climb by 26.5% a year, reaching $70.73 billion by 2032.
This is where AI comes in. Advancements in AI, including generative AI and AI-powered robotics, could contribute to that growth. As AI capabilities evolve, manufacturers are integrating advanced algorithms into their maintenance systems, betting that smarter, faster predictions will improve business operations and savings.
That is, if growing pains don’t get in the way.
AI and robots are enhancing predictive maintenance
Manufacturers’ older systems that are designed to identify potential machinery breakdowns often overwhelm maintenance teams with false alarms, triggering unnecessary part replacements and wasted labor. Now, some startups such as Aquant are banking on AI to spot issues both faster and more accurately.
Aquant said its AI-powered algorithms can analyze past maintenance data to determine whether a maintenance action is necessary. For example, cross-referencing technicians’ notes and industry benchmarks can help identify redundancies.
Its platform then analyzes machines’ live sensor data — such as vibrations, sounds, and temperature shifts — alongside historical records from customer relationship management systems and other information repositories.
Proprietary algorithms sort through the information to “find the needle in the haystack” and diagnose issues that need addressing as well as the causes, said Assaf Melochna, Aquant’s president and cofounder. The platform, which is also designed to filter out false positives, then recommends specific actions, such as shutting down the machine, to prevent failure, Melochna explained.
By focusing on factory machinery components that affect uptime, Aquant says its AI platform is helping clients such as Coca-Cola, HP, and Hologic cut downtime and avoid unnecessary repairs, saving up to 23% a year in service costs.
AI can also make inspections safer. Gecko Robotics builds wall-climbing robots, drones, and robot dogs equipped with ultrasonic sensors, high-resolution cameras, and light detection and ranging technology.
Ross Mantle for BI
These robots are designed to inspect critical infrastructure like dams, power plants, and oil and gas facilities, collecting detailed data on pipes, tanks, vessels, and other industrial equipment. Gecko’s AI platform, Cantilever, analyzes the data to detect corrosion, erosion, and cracking before failures occur.
The combination of the data obtained by the robots and the AI platform to analyze it helps give Gecko’s clients, including Siemens Energy and the US Air Force, a real-time, predictive view of asset health, said Jake Loosararian, Gecko Robotics’ cofounder and CEO, adding that this makes it easier to plan maintenance before problems escalate.
Generative AI is coming to maintenance
Ross Mantle for BI
Some companies are layering conversational AI onto their maintenance tools.
Since 2006, Waites Sensor Technologies, a predictive maintenance firm, has installed more than 500,000 vibration and temperature sensors across Amazon, Tesla, DHL, and other facilities.
Rob Ratterman, the cofounder and CEO, told BI that Waites had more recently integrated large language models into its predictive systems to let technicians query their systems directly. Waites’ LLMs are trained on machine manuals, repair history, and sensor data collected by the company’s sensors.
Using devices like on-site computers, maintenance personnel can ask Waites’ systems which machines are most likely to fail, what past repairs were needed, and what maintenance tasks should be prioritized. Technicians, for instance, can ask the chatbot questions like, “What are the most problematic pieces of equipment?” to identify issues, or “How do I relubricate a bearing?” to find instructions in a machine’s manual.
By simplifying access to maintenance insights, LLMs can help teams make faster, data-driven decisions without needing to dig through complicated dashboards or reports, Ratterman explained.
With advancements in AI, predictive maintenance is now evolving into prescriptive maintenance, Kevin Tucker, an advisory practice lead at Info-Tech Research Group, told BI.
While predictive maintenance identifies a problem, prescriptive maintenance will tell you what to do to solve that particular issue. The more data predictive maintenance systems capture and analyze, the more accurate their predictions become, enhancing their abilities to make informed decisions.
“It’s kind of like following your GPS in the car,” Tucker said, underscoring the potential for prescriptive maintenance to act as a trusted guide.
Upfront costs and fears around AI can hinder adoption
Ross Mantle for BI
While it has upsides, AI-driven predictive maintenance isn’t easy to implement.
One major hurdle is cost. Installing smart sensors, integrating AI platforms, and upgrading legacy equipment requires major upfront investments, said Jorge Izquierdo, the VP of market development at The Association for Packaging and Processing Technologies, a trade organization.
Izquierdo called predictive systems the “lowest hanging fruit” for smart manufacturing in the packaging industry, but Tucker said incorporating new technologies into legacy systems might be hard to prioritize. After all, returns on investment aren’t immediate, and system complexities — including data silos and the different softwares manufacturers use across the business — could make integration even harder.
“You’re probably going to get your return on investment in a year and a half to two years out,” Tucker said.
Loosararian emphasized that it’s tough for companies, especially those running on tight budgets, to allocate resources toward implementing new technology on a large scale. “Innovation budgets run out pretty fast,” he said.
Workforce skills gaps pose another roadblock. Predictive maintenance demands new expertise in data analysis and AI tools management — skills that many traditional maintenance teams don’t have and may be resistant to learn.
That resistance is compounded by the concern that predictive maintenance systems, especially with advanced AI capabilities, may replace maintenance technicians’ and inspectors’ jobs. But some manufacturing leaders say the technology can support, not replace, maintenance personnel.
Craig Malloy, the CEO of Arix Technologies, a company behind robotic and AI systems that detect corrosion in industrial pipes, said its AI tools help free up time for maintenance personnel to focus on decision-making and long-term planning. “The goal is smarter, safer, more effective inspections — not fewer people,” Malloy told BI.
Loosararian added that arming the workforce with AI tools can be one step toward addressing the growing labor shortage in US manufacturing.
Predictive maintenance leaders who spoke with BI agree that AI advancements are still in their early stages of transforming maintenance systems. But manufacturers willing to invest and stick through the growing pains could be the difference between thriving and falling behind.
As Malloy said: “AI has completely changed the game.”
