IMARC 2025 Keynote: How to Master the Full Spectrum of Failures in Mining
November 4, 2025
At IMARC 2025, Razor Labs CTO and co-founder Michael Zolotov presented a keynote “Mastering the Full Spectrum of Failures: From Fixed Assets to Mobile Fleets and AI Vision.” Drawing on insights from mining sites across six continents, Michael laid out a clear, engineering-first framework for closing the blind spots that still cause critical equipment failures – even in highly instrumented operations.
Why failures still happen – despite all the existing systems
Michael opened with a simple but critical question: Why do catastrophic failures still occur in mining operations despite having SCADA, historian logs, vibration reports, oil analysis, and OEM alerts?
He mapped out the typical failure coverage at mine sites and showed that up to 75% of failures remain undetected due to:
Delayed or manual inspection cycles (e.g. monthly vibration or oil analysis),
Operational noise masking real signals,
Alarms that rely only on existing sensors and must be manually investigated,
Gaps in sensor coverage – especially visual, mechanical, and contextual anomalies.
The result? High-impact failures that aren’t prevented, just reacted to.
From thousands of alarms to a handful of root causes
One of the major themes was reducing “alarm overload.” Michael showed how Razor Labs’ system filters thousands of daily alerts into a small number of actionable failure patterns, by automatically correlating data across systems (e.g. pressure, oil, harness feedback) to surface true root causes.
In one case, what seemed like multiple sensor faults in a Caterpillar haul truck was traced back to a single issue with the engine harness – avoiding costly misdiagnosis and repeated downtime.
Cutting through operational noise
Mining environments are inherently noisy – from fluctuating payloads to varying engine loads. That noise often masks real failure trends, particularly in systems like transmissions or ball mills.
Michael demonstrated how AI-based noise compensation allows the system to extract failure patterns weeks before any OEM alert is triggered, giving teams time to act before damage occurs.
Visual AI: Detecting what sensors can’t
From oversized ore at crusher discharge to misaligned conveyor belts and flotation cell bubble size, some of the most important failure modes are simply invisible to traditional sensors.
With real-time cameras and AI models, Razor Labs provides early detection of:
Material flow issues,
Belt damage or misalignment,
Recovery losses in flotation,
Mechanical wear in splices, liners, and clips.
Holistic coverage – not just data, but context
What sets Razor Labs apart is the unified approach: vibration, historian, visual data, fluid reports, operator behavior, and maintenance logs are all treated as pieces of the same diagnostic puzzle.
Michael concluded with a clear message:
Only a system that connects all failure modes – across fixed assets, mobile fleets, and visual inspection – can reliably reduce downtime and prevent recurrence.
