Hear the failure coming — 48 to 96 hours early.
Machine-learning models score every machine's health from 0 to 100 each day from vibration, current and thermal trends. When a component starts to degrade, you get the likely component, an estimated time-to-fault and the part to pre-order, so the fix lands in a planned stop.
Every unplanned breakdown was once a prediction nobody made.
“It just failed”
Machines rarely just fail. They announce it in their vibration signature for days or weeks. Nobody was listening; now something always is.
Servicing the wrong machines
Fixed-interval PMs service healthy machines and miss dying ones. Condition data puts the effort where the risk actually is.
The part that isn't there
The repair takes four hours; waiting for the bearing takes four days. A time-to-fault estimate buys you the procurement window.
Failure is a slope, not a moment. The earlier you see it, the cheaper it is.
Between the point where a defect first becomes detectable (P) and the point where the machine stops doing its job (F), each signal shows up at a different time. Vibration is almost always first. Heat and noise arrive late, when the only choice left is an unplanned stop.
Drag through thirty days of a bearing failing.
A spindle bearing develops an outer-race defect. Watch the health score, the defect band and what MachineWise tells your team on each day, long before anything is audible.
ILLUSTRATIVE MODEL OF A TYPICAL BEARING DEGRADATION · REAL CURVES VARY BY MACHINE, LOAD AND FAILURE MODE
Six steps from a sensor on a bearing housing to a part on the shelf.
Sense
Tri-axial vibration at the bearing positions that matter, plus current, temperature, speed and load.
Baseline
Each machine learns its own healthy signature, per speed range, over its first weeks of normal running.
Detect
Spectral bands tracked at the real running speed; sustained slopes flagged, single spikes ignored.
Score
Evidence fused into one health score, 0–100, per machine per day. Fleet ranked worst first.
Predict
Time-to-fault estimated as a range with confidence, and the likely component named where the signature allows.
Act
Ticket, parts list and a suggested intervention window around production. Outcome logged to sharpen the model.
Failure modes, their signatures, and how early they show.
These are the wear-out failures that account for most of the cost of unplanned downtime on rotating equipment. Work out the defect frequencies for your own bearings with the bearing frequency calculator.
| Failure mode | Signature MachineWise tracks | Typical lead time | What your team is told |
|---|---|---|---|
| Rolling-element bearings | Envelope energy at the bearing defect frequencies (BPFO, BPFI, BSF, FTF) | Weeks to days | Component, band, trend and the bearing to order |
| Imbalance | Rising amplitude at 1× running speed, mainly radial | Weeks | Imbalance flagged with the speed it appears at |
| Misalignment | Energy at 2× (and 3×) running speed, strong axial component | Weeks | Coupling and alignment check recommended |
| Mechanical looseness | Many harmonics of running speed, raised noise floor | Days to weeks | Mounting or foundation inspection |
| Gear wear | Gear-mesh frequency with growing sidebands | Weeks | Stage and gear pair where the signature allows |
| Lubrication breakdown | High-frequency floor rising, with bearing temperature | Days | Lubrication task raised before damage starts |
| Motor and drive faults | Current signature changes, load drift at constant work | Days to weeks | Drive or winding inspection |
| Thermal degradation | Longer warm-up, higher steady-state temperature | Days to weeks | Cooling, chiller or filter service |
LEAD TIMES ARE TYPICAL RANGES, NOT GUARANTEES. SUDDEN EVENTS (CRASHES, PROGRAM ERRORS) ARE NOT PREDICTABLE AND ARE HANDLED BY CRASH DETECTION AND DOWNTIME CAPTURE.
Predictive maintenance for the machines that stop your plant.
CNC spindles
- WATCHSpindle bearings
- WATCHBallscrews and axis drives
- WATCHTool changer mechanics
- WATCHChiller and thermal behaviour
Gearboxes
- WATCHGear-mesh and sidebands
- WATCHStage-wise bearings
- WATCHLubrication condition
- WATCHLoad-normalised trends
Air compressors
- WATCHAirend bearings
- WATCHMotor and coupling
- WATCHDischarge temperature
- WATCHSpecific power drift
Motors, pumps and fans
- WATCHImbalance and misalignment
- WATCHBearing wear
- WATCHCurrent signature
- WATCHCavitation and flow faults
Presses and hydraulics
- WATCHMain drive and flywheel
- WATCHHydraulic pump wear
- WATCHOil temperature trend
- WATCHClutch-brake behaviour
Furnaces and blowers
- WATCHCombustion and circulation fans
- WATCHElement ageing
- WATCHRecovery-time creep
- WATCHZone-to-zone drift
What are breakdowns on your critical machines costing?
Enter your own figures. The range applies the 60–80% reduction plants typically see on monitored machines.
Count hours from the stop to the machine running good parts again, including the wait for parts. Cost per hour is the contribution the machine would have earned, plus overtime or outsourcing to recover it.
Predictive maintenance handles wear. It does not predict accidents.
Predictable, with warning
- Spindle, motor, gearbox and fan bearings
- Imbalance, misalignment and looseness
- Gear wear and lubrication breakdown
- Drive, winding and thermal degradation
- Tool changer and axis mechanics that wear gradually
Not predictable, by nature
- A crash from a program or setup error
- A tool that breaks on a bad insert
- An unclamped fixture or out-of-spec bar
- Operator error and power events
- Anything with no measurable build-up
Events like these are caught the instant they happen and recorded with evidence, which is a different job. Component by component on a machining centre: what is predictable on a CNC, and what is not →
Five ways to maintain a machine. One platform runs all of them.
Most plants run all five at once on different machines, and that is correct. A spare-rich pump can run to failure; a constraint CNC cannot. MachineWise lets each machine sit on the stage its cost of failure justifies.
Straight answers.
What is AI predictive maintenance?
How early can MachineWise warn before a machine fails?
Which machines and components can it monitor?
Which sensors are needed?
How do you avoid false alarms?
What is the difference between preventive, condition-based and predictive maintenance?
How many breakdowns does predictive maintenance prevent?
Does the data go to the cloud?
Where this connects to the rest of the platform.
Instrument one critical machine. See its health score within weeks.
Start with the machine whose breakdown hurts most. Two machines, thirty days, zero cost to start, with your data on-premises throughout.