Monitoring shows what happened. AI thinks ahead.
Purpose-built AI modules trained on data from 2,000+ machines. They score machine health daily, warn before failures, balloon drawings in minutes and answer plain-language questions about your floor, all running inside your own deployment.
A health score every morning. A forecast every week.
The two numbers maintenance and planning teams open first: how healthy each machine is today, and what next week's OEE is likely to be.
Daily machine health, 0–100
ML models score every machine each day from vibration, load and thermal trends, per machine type.
Next-week OEE forecast
Forecast from health scores, scheduled PMs and recent history, so planners see capacity before the week starts rather than after it ends.
Each module replaces a job that used to take hours.
Pick a module to see what it reads, what it hands back and who on your team uses it.
Daily machine health, 0–100
Every monitored machine gets a single number each morning, computed from its own vibration, load and thermal trends by models specific to its machine type. Worst machines first, so the maintenance meeting starts with the right one.
Condition trends from vibration, spindle or motor load, current and temperature, plus the machine's own history.
VMC-09 · 74 ▼ (was 82 on Monday)
Outer-race band rising at spindle front bearing. Trending toward a planned intervention window.
Time-to-fault, with the parts to pre-order
When a health score starts falling, the model estimates how long the component has left, names the likely component where the signature allows, and lists the part to order, so the change happens in a planned stop instead of an unplanned one.
Health-score trajectory, defect-frequency bands, running speed, PM history and past failures on that machine.
Est. time-to-fault 48–96 h · bearing 6206 · schedule change into Saturday's planned stop · ticket raised for Maintenance.
From drawing to inspection plan in minutes
Upload a drawing as a PDF, DXF or phone photo. The AI numbers every critical-to-quality dimension, extracts nominal values and tolerances into an inspection table and produces a ready QC checklist your inspector can start from.
A part drawing: PDF, DXF or a clear photo.
Ballooned drawing + inspection table: #14 Ø32 H7 (+0.025/0) · #15 Ra 1.6 · checklist exported for the first-article inspection.
Questions in plain language, answers with evidence
Ask what happened, in the words you would use with a colleague. Answers are grounded in that machine's own states, cycles, alarms, sensor trends and PM records, and they show the evidence behind them rather than a confident guess.
Your question, plus the machine's recorded states, cycles, alarms, sensor trends and maintenance history.
“Why did CNC-06 lose 18% OEE last Tuesday?” → Availability held at 96%; cycle time ran 24% over baseline after the 11:14 insert change with feed override at 70%.
Structured 5-Why, pre-filled with the evidence
Pick any downtime event or rejection. The module walks a structured 5-Why, pre-populates each level with the sensor and event evidence it can find, marks how confident it is at each step and exports a QMS-ready report your team reviews and signs.
A downtime or rejection event, with everything recorded around it.
5-Why report: stop → spindle overload alarm → worn insert → tool change skipped at 380 parts vs 300 limit → counter reset missed on shift change. Confidence shown per level.
The exact page of the manual, mid-breakdown
Upload machine manuals, alarm lists, SOPs and standards once. A technician asks a question in plain words and gets the exact cited section back, instead of paging through a 300-page PDF with a machine stopped.
Manuals, alarm references, SOPs and internal standards you upload.
“Alarm 1004 on the HMC pallet changer?” → cited section from the maintenance manual with the reset procedure and page reference.
Next week, predicted per machine
Next-week OEE and energy use are forecast for every machine from its health score, scheduled PMs and recent history, giving planners a forward view of capacity instead of last week's report, and feeding capacity mapping for route cards.
OEE and energy history, health scores, planned maintenance and the production plan.
CNC-6 · forecast OEE 72–78% next week · PM due Thursday · 3 route cards better placed on CNC-8.
From machine signal to recommended action, inside your plant.
The same data that drives your OEE dashboards feeds the models. Nothing extra to install on the machines for the explain, root-cause and forecast modules; condition sensors unlock health scoring.
Sense
States, cycles, alarms, spindle load, vibration, current and temperature, time-stamped at the machine.
Baseline
Each machine learns its own normal: per job, per speed range, per shift pattern.
Detect
Drift from that baseline is flagged: a rising defect band, a cycle creeping, an alarm pattern repeating.
Reason
Models score, forecast and explain, and attach the evidence and a confidence to every output.
Act
One tap turns a finding into a ticket, PM, route card or WhatsApp summary that a person approves.
Your machines
- CNC, VMC, HMC, lathes, grinders, EDM
- Presses, moulding, furnaces
- Compressors, gearboxes, test beds
- Conventional machines via digital I/O
Data, models and AI
- Machine data store, on-premises
- Health, failure and forecast models
- Language models for questions, drawings and manuals
- Your uploaded manuals and SOPs
Where answers land
- Dashboards and TV andon walls
- WhatsApp, email and alerts
- Maintenance tickets and PMs
- Route cards and reports
One set of models. Five people who stop guessing.
Where did the week go?
Ask in plain words; get losses ranked with evidence and a one-line action for each.
Which three machines cost us most OEE this week, and why?
What fails next?
Fleet ranked by health score, time-to-fault estimates and the parts to order.
Which machines need attention before Saturday's shutdown?
Why was it rejected?
5-Why pre-filled with machine evidence; ballooned drawings in minutes.
What was the machine doing when batch 2214 went out of tolerance?
What will next week hold?
OEE and energy forecast per machine, and where each route card should run.
Where should RC-25-54 run to finish by Friday?
How do I fix this?
Cited answers from your own manuals and SOPs, at the machine.
What is the reset procedure for alarm 1004?
AI your maintenance team will actually trust.
An alert that is wrong three times gets muted, and then the fourth one, the real one, is ignored. Every design choice below exists to stop that happening.
Grounded in your machines
Every answer and prediction is computed from that machine's own recorded data, not a generic model of a machine like it.
Shows its evidence
Which band rose, which alarm fired, which cycle drifted. You can check the reasoning before you act on it.
States its confidence
Estimates come as ranges with confidence. A prediction with fake precision is worse than an honest range.
Your data stays with you
Models run inside your MachineWise deployment. Machine data never leaves it.
People approve, AI proposes
Nothing is changed on a machine. Recommendations become tickets and route cards that a person accepts.
Learns per machine
Two identical machines on different foundations run differently. Baselines are learned for each one.
What the AI does, and what it will not pretend to do.
It will
- Warn about wear-out failures (bearings, drives, gearboxes, lubrication) days ahead
- Explain OEE losses with the cycle, alarm and sensor evidence behind them
- Turn a drawing into an inspection plan and a checklist in minutes
- Find the right section of a manual while the machine is stopped
- Say “not enough data” when there is not enough data
It will not
- Predict events such as crashes, program errors or an unclamped fixture
- Write to your machines or change a parameter
- Invent a reading that was never recorded
- Send your machine data outside your deployment
- Replace the judgement of the person who signs the ticket
Why the event-versus-wear boundary matters, component by component: predictive maintenance for CNC machines →
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 MachineWise AI?
Does MachineWise AI need the cloud or an internet connection?
What data does the AI use?
Can the AI change settings on my machines?
How early does the predictive maintenance module warn before a failure?
Does it work on old machines?
Is MachineWise AI priced separately?
Where this connects to the rest of the platform.
See MachineWise AI working on your own machine data.
A 30-minute walkthrough on a live deployment, then a free 2-machine pilot on your floor. Your data stays on-premises throughout.