Every device
has a point of
failure.

A new machine fresh from the manufacturer is healthy and problem-free. But due to wear and tear, its health slowly deteriorates — and eventually it fails. When that happens, you need to perform maintenance to get it back to a working condition.

The question isn't if it will fail. The question is what you do about it before it does. There are three approaches — and the difference between them is measured in production hours, maintenance costs, and competitive advantage.

The chart below shows real sensor data from an industrial motor. The signal isn't smooth — it drifts, spikes, and recovers. That's what real degradation looks like. Switch between the three maintenance strategies and watch how each one interacts with that reality.

Industrial Motor — Reactive
Equipment
Industrial Motor
Y axis
Vibration mm/s
Failure threshold
14.5 mm/s
Life saved
0 days
Strategy comparison
Maintenance trigger Device failure
Unplanned downtime High risk
Useful life wasted None
Data required None
The natural progression
01 — Reactive
React
Something breaks. You fix it. No data needed — but every failure is a surprise.
Trigger: failure
02 — Preventive
Plan
You set a schedule. Downtime is controlled — but you pay for maintenance whether the machine needs it or not.
Trigger: calendar
03 — Predictive
Predict
The data tells you when. You intervene at exactly the right moment — maximum life, minimum downtime.
Trigger: data pattern
04 — Prescriptive
Act
The model doesn't just predict — it tells you what to do, who should do it, with what part, and when.
Trigger: connected systems
Predictive tells you
the future.
Prescriptive tells you
what to do about it.
Prescriptive maintenance

Once you start making decisions based on data, something changes. You realize the data doesn't stop at the machine. It connects to your inventory. To your workforce schedule. To your production plan. To your supply chain.

Prescriptive maintenance is what happens when those connections exist. The model doesn't just say "bearing will fail in 8 days" — it says "bearing 3B will fail in 8 days. Technician available Thursday. Part in stock at A7. Schedule the intervention at 2pm to minimize OEE impact."

That's not science fiction. That's the natural destination of every plant that starts asking the right questions about its data. Every dataset you build today is infrastructure for the decisions you'll make tomorrow.

What prescriptive requires
CMMS connected in real time — not just logged
Live inventory and spare parts visibility
Workforce availability as a live variable
Production schedule optimization integrated
Most plants are still building this infrastructure. Predictive is the foundation — prescriptive is what comes next.
How does the model actually read the sensors?

From sensor signal
to anomaly detected

You don't need to be an engineer to understand this. The concept is basic — and once you see it, you can't unsee it.

Industrial Motor
ENVIRONMENTAL SENSORS Cameras normal / night vision Smoke & Gas leak detection Humidity ambient moisture Temperature ambient air Radiation nuclear / pharma MACHINE SENSORS Sensor A · Vibration Sensor B · Thermal Position shaft alignment Pressure internal / fluid Level oil / coolant Strain mechanical stress Sound / Vibration acoustic emission Environmental Machine (attached) Active in this demo
Readings collected
0
Model status
Learning...
Last reading
Sensor A vs Sensor B
Think of it like...
Every time the motor runs normally, you write down what both sensors read. After enough readings, you know what "normal" looks like on a map. When a new reading falls far outside that map — the model flags it. No formulas. Just memory.
Demo 02 — Sensor Anomaly Detection

How the model
learns what's normal

Two vibration sensors are installed on the motor — Sensor A on top, Sensor B on the bottom. Each reading gives you two numbers. Plot them together and a pattern emerges: normal operation always falls in the same zone. When a reading falls outside that zone — the model already knows something is wrong.

Phase 1 — Collecting normal readings
The motor is running normally. Each reading is a data point. The model is learning what normal looks like — building its map.
The connection to KNN
This is exactly how K-Nearest Neighbors works — it measures the distance from a new point to what it already knows. If the nearest neighbors are all "normal", the reading is normal. If not — it's an anomaly.
→ See KNN demo in Models
Not a calendar.
Not a rule.
A pattern the data
already knows.
How predictive ML works

How do we predict when a device fails? Using data collected from similar devices in the past. With industrial motors, there are already thousands of similar machines installed across plants worldwide — each one generating vibration, temperature, and current data throughout its lifecycle.

By analyzing patterns from motors that already failed, a model learns what the degradation curve looks like before breakdown. When it sees those same patterns in a new motor, it can estimate how much life is left.

The fancy term we currently use — or maybe overuse — is AI. But the basic principle is simpler: analyze historic data, recognize patterns, make better decisions about the future. That's it.

We're still in the early phases of this transition in industry. Preventive maintenance has decades of trust behind it. Predictive maintenance is still earning that trust — one dataset at a time.

Real project — Motor Failures Prediction
KNN applied to industrial motor failure
This is exactly what the Motor Failures Prediction project does: KNN finds the K most similar motors that already failed — by vibration profile, temperature, and load — and predicts when yours will reach the same point. Real dataset. Open code.
Full Repository
The complete picture

From sensor to work order —
the full ecosystem

This is what predictive — and eventually prescriptive — maintenance looks like in practice. Click any block to learn what it does.

Sensor A Sensor B MACHINE Vibration · Thermal · Pressure Sensor Data Real-time readings Knowledge Base Manuals · History ML Pattern Recognition KNN · Random Forest Logistic · SVM ERP Operations Production schedule CMMS Work orders Alerts Notifications Prescriptions SAP PM · Maximo Inventory Asset Data Maintenance History Technician App Work order · Instructions · AR ① Detect ② Collect ③ Predict ④ Decide ⑤ Context
Selected block
Click any block
to learn what it does
Each block in this diagram plays a specific role in turning raw sensor data into a maintenance action. Click any block to understand its role in the ecosystem.
What's next

See the models behind the prediction

KNN, Logistic Regression, Decision Trees, SVM — 11 supervised learning models explained through interactive visualizations. Understand the math, the geometry, and when to use each one.

Quality · Operations · SCM — coming soon