Lab 06
Predictive Maintenance
Spot trouble before a machine stops
Run a simulated pump, introduce a fault and watch a trained AI look for warning signs.
Overview
A machine can start going wrong before anyone notices.
How this could help a business
The same application pattern can help maintenance teams investigate unusual equipment behaviour. Real deployment needs actual machine data, operating limits and a plan for handling alerts.
What I built
- A repeatable pump simulator in TypeScript and Python, checked against the same recorded inputs.
- Trained anomaly detectors and a browser worker that runs the published autoencoder.
Live demo
Start the pump with one click. Let it collect readings, then introduce a fault and watch it develop.
What I built
Implementation, decisions and evidence.
What Sebastian built
The application around the technology.
- A repeatable pump simulator in TypeScript and Python, checked against the same recorded inputs.
- Trained anomaly detectors and a browser worker that runs the published autoencoder.
- A monitoring interface connecting sensor readings, fault controls, the trained threshold and a persistent alarm.
One engineering decision
Require an unusual pattern to persist before raising an alarm, rather than treating every noisy reading as a breakdown.
What to check yourself
Watch the simulated sensor readings and the real detector score. The trained model files and model card are published on Hugging Face.
Tools and model credits. ONNX Runtime runs the trained model. The pump and faults are simulated; the model's scores are computed during the run.
These are portfolio demonstrations. Explore professional experience separately, or discuss how the approach could fit your team.
Learn
Understand what the application is doing.
Watch the readings settle, then choose one fault. Compare the sensor pattern and the alarm message; a warning can take time to develop.
Pause and predict
Should one unusual sensor reading always trigger a breakdown alarm?
A learning question, not a score. Try this in the working demo.
Try changing something
Now make it your experiment.
Let the pump run normally, introduce bearing wear, then remove the fault. Watch the recovery rather than expecting an instant reset.
The pump is simulated, so this is not a validated factory safety system. The Isolation Forest comparison and per-sensor explanations are trained and benchmarked but not wired into this view yet.
A few words, explained
- Anomaly
- A pattern that looks different from the healthy examples the detector learned.
- Threshold
- The measured score must cross this boundary repeatedly before the alarm turns on.
Data handling & current limitations
The pump is simulated. The trained detector, alarm threshold and measurements are real.
Demo only. Each lab below explains which inputs stay on your device and which data is sent for server-backed features. Model files are downloaded and may remain in your browser cache. Hosting and security services also receive the network requests needed to serve the site.
The simulator and the anomaly detector run in this browser. Simulated telemetry is not uploaded. Downloaded model files may remain in your browser cache.
The pump is simulated, so this is not a validated factory safety system. The Isolation Forest comparison and per-sensor explanations are trained and benchmarked but not wired into this view yet.