Cognitive Maintenance: Beyond Predictive, All the Way to Root Cause
On 20 November 2026, IIoT World published an interview with the CEO of GroundUp AI on cognitive maintenance, which fuses sound, vibration and thermal data to diagnose the cause of a failure remotely. The claims come from the vendor and have not been independently verified.
On 20 November 2026, IIoT World published an article based on a video interview with Leon Lim, CEO and founder of GroundUp AI. The subject is cognitive maintenance, presented as the next step beyond predictive maintenance. Bear in mind that this is a vendor's account: its claims should be treated as unverified.
Predictive vs. cognitive: what is the difference?
Predictive maintenance estimates when an asset might fail and usually returns a time window and a severity level. Someone then has to open up the machine to find out why: misalignment, cavitation, a worn bearing. According to Lim, cognitive maintenance aims to deliver the root cause and the failure mode remotely.
"We don't just tell you when your machine will spoil. We say why your machine spoils, and how it's going to spoil."
Sensor fusion
The approach described synchronizes sound, vibration and temperature readings on the same timestamp, so that the three signals validate one another. The article claims these signals change before conventional alarm thresholds trip, and contrasts the approach with typical single-modality predictive systems that rely on vibration alone.
The naval defense case
Before the change, the crew inspected rotating groups of about 10 machines at each port call. Docking windows lasted only a few hours, so many machines went unchecked. After adoption, maintenance priorities come from remote alarms ranked by severity, each with a specific diagnosis, and procurement planning starts from the diagnosed failure mode.
The stated result is 22 consecutive months without unplanned downtime, with the same crew covering the entire fleet. However, the article provides no baseline data, and the result is reported by the vendor itself.
The human factor
The average age of experienced maintenance technicians has risen from 48 to 52 over five years. The GroundUp Asset Library catalogs failure patterns at the sub-component level, training on each failure type separately, with the aim of preserving veterans' diagnostic knowledge after they retire. Some older technicians support it, seeing it as the automated version of the standard operating procedures they have always wanted documented.
Looking ahead, the article discusses "physical AI": machines that detect their own faults and coordinate with neighboring systems, something that would require the same multimodal diagnosis.
Why it matters for those building monitoring projects
Practical takeaways for anyone designing monitoring and AI systems:
- An alarm should carry a diagnosis and a priority, not just a predicted failure date.
- You need time-synchronized multi-sensor acquisition and a library of labeled failures.
- Clarify how diagnoses will be integrated with SCADA and CMMS, for example to generate work orders and plan spare parts.
- Plan how to capture experienced technicians' knowledge in labels and procedures.
Caveats
The article reports no accuracy metrics, and the only case cited, the naval one, is supplied by the vendor. Before drawing conclusions, we would need data on false positives, missed diagnoses and comparison against a baseline. The topic is related to the one covered in our earlier article on AI fault diagnostics and Honeywell's vision, but here the emphasis is on multimodal fusion, root cause and knowledge preservation.