AI anomaly detection: hundreds of machine sensors monitored automatically
A mid-sized manufacturer replaced daily manual monitoring of machine data with AI anomaly detection — 24/7, without anyone having to look.
By Gabriel Michelberger | June 2025
€30–60k
estimated annual savings
~95%
detection accuracy
Quality monitoring
Manual
Before
AI-driven
After
AI watches, specialists work
A mid-sized manufacturer monitors hundreds of sensors on machines and equipment. Until now, staff had to review machine data and readings manually every day.
SimplifieD Solutions built an AI system that analyses machine data automatically and only raises an alert when something is actually wrong. The daily routine check disappears entirely.
Four problems with sensor monitoring
Daily routine checks
Each morning, hundreds of sensor readings reviewed by hand — a huge time sink and error-prone.
Hidden anomalies
Slow drifts over weeks are nearly impossible for the human eye to spot across hundreds of data points.
False alarms from maintenance
Scheduled shutdowns and maintenance windows produced outliers that looked like faults.
Specialists tied up
Qualified staff wasted time on routine monitoring instead of value-adding work.
How the anomaly detection works
Reads existing data directly
No migration required. The system taps into the available readings, cleans them, and prepares them for analysis.
Detects anomalies with a double check
Two neural networks analyse independently. An alert is only raised if both models flag something — minimal false-alarm rate.
Filters out maintenance cycles
The system recognises scheduled shutdowns automatically and excludes them — including a time buffer. No false alarms from maintenance.
Tunable per machine
Three configurable parameters allow adjustment to different machines — every asset has different normal-operating ranges.
Clear visualisation of anomalies
Normal signal as a line, anomalies highlighted, scheduled shutdowns as green zones. One glance is enough.
Before vs. after
| Metric | Before | After |
|---|---|---|
| Daily review effort | Manually reviewing all sensor data | Only when there’s an alert |
| Detection method | Human eye scanning curves | AI ensemble with double check |
| False alarm rate | High (scheduled shutdowns) | Minimised via ensemble + maintenance exclusion |
| Slow drifts | Often detected late | Detected automatically via autoencoder |
| Monitorable data points | Limited by staff time | Hundreds of sensors in parallel, 24/7 |
| Skilled staff utilisation | Routine monitoring tied up specialists | Specialists freed up for project work |
2–3 months
From start to production pilot
€30–60k
estimated annual savings
“My engineers used to start their day clicking through data reviews. Now the system only speaks up when something is actually wrong. That’s exactly the kind of AI I want — invisible until it’s needed.”
— Managing Director, Specialised Machinery
Similar outcomes for your business?
In a free 30-minute call we’ll find out where your biggest levers are.
“I’ll take the time personally.”
Oliver Bührer
Managing Director
