MCIFT combines multiple operational signals into transparent early-warning rules. Two public bearing benchmark runs now document distinct frozen protocols, decision definitions, outcomes and limitations.
Public dataset · Reproducible artifacts · Clear limitations · No black-box AI required
Install, inspect and test the deterministic multichannel condition-monitoring implementation. The public release includes versioned profiles, safe model serialization, reproducibility documentation and an initial NASA IMS bearing case study.
python -m pip install "mcift==0.1.0a1"Inspect the derived score timeline, persistence rule, baseline comparison, methodology and limitations for one IMS Set 2 run.
Explore how the same MCIFT thinking can reduce downtime, surface risk, and make complex systems easier to manage.
Show how heavy AI workloads can spread heat, network pressure, queue growth, and capacity stress across a connected data-centre cluster.
Spot slowdowns and missing work across APIs, databases, queues, and ETL before users feel the full impact.
Combine vibration, heat, and pressure changes into an early warning for engines, turbines, and pumps.
Rebalance inventory, energy, computing capacity, or budget without losing track of the total.
Find unusual repeating patterns in rotating machinery, coverage scans, or other cyclic measurements.
Show how far an observation has moved from expectation while keeping uncertainty visible.
Require several independent checks to agree before a workflow, sensor decision, or transaction moves forward.
Every result comes from repeatable mathematical rules. AI is optional—not required to create the alerts shown here.
Separate a meaningful repeating pattern from normal background noise.
Measure change against the way that system normally behaves.
Track what enters, leaves, and waits so missing work becomes visible.
Use several independent checks before an important decision moves forward.
AI can adapt normal ranges to each customer or learn from history. The decision path stays visible and can run without AI.
Each demo shows how ordinary signals can become an understandable warning or action.
Replay GPU-cluster load. The field tracks whether heat, network pressure and queue growth remain local or spread through connected nodes.
Connected pressure forming
Whether node relationships help distinguish a faulty sensor from an event spreading through infrastructure.
10 of 27 nodes cross the illustrative review level.
An alert appears after a local node crosses its configured limit.
Neighbour changes form one pattern. Lead time and localization require validation on recorded incidents.
Each use case must prove measurable value before it becomes an operational product.
Choose the machine or digital process, its normal behavior, and the decision that matters.
Turn several weak clues into evidence that a business user can understand.
Test against real history and compare warning time, accuracy, and cost with today’s process.
The public package documents and implements the current stable software profiles. Experimental research ideas and future validation work remain distinct from that versioned behaviour. These demos illustrate computational mappings. They are not certified diagnostics and do not establish general scientific validity or operational performance. Real data must test each use case independently.
MCIFT · PUBLIC SOFTWARETwo frozen benchmark runs evaluate MCIFT on a public bearing dataset. The next step is to test whether the approach adds useful warning time on another machine, workflow or digital system.
A bounded validation study compares MCIFT with your existing thresholds, monitoring rules or an established baseline—without assuming a positive result.