All case studies

Case studyUnder NDA Predictive maintenance for industrial machines

A manufacturer's line stopped without warning when bearings wore out, and the machines were checked by hand once a month. We built a system that spots problems in vibration data before they stop the line.

Client
Manufacturing company, Central Europe
Industry
IoT, Manufacturing
Platform
Web, edge
Scope
Data, backend and UI
Line overview with vibration levels per machine and conditions to act on
~30%less equipment downtime
5months in development, ongoing
3people: Python/ML, backend, frontend
The challenge

Monthly checks, sudden stops

Worn bearings stopped the production line without warning. Machines were monitored by hand, once a month.

The solution

What we built

01

Data pipeline

Vibration and acoustic data from the machines' PLCs reaches the system through an edge gateway, over OPC UA and Modbus TCP, and is stored in ClickHouse.

02

Anomaly detection

A PyTorch module that finds anomalies in the data.

03

Engineer dashboard

A React dashboard where engineers see the state of the machines.

04

Light on the uplink

High-frequency vibration data is compressed on the edge gateway so it does not clog the connection.

Machine page with the anomaly score and the vibration spectrum
Anomaly detection
Maintenance plan with jobs raised by the model
Maintenance plan
The stack

Technology stack

Backend and ML
  • Python
  • FastAPI
  • PyTorch
Frontend
  • React
Data
  • ClickHouse
Industrial
  • OPC UA
  • Modbus TCP
Infrastructure
  • Docker
Results

What the client got

  • Maintenance is planned instead of done after a breakdown
  • Equipment downtime is down by about 30%

Building an IoT platform?

Tell us about it. We'll suggest an approach and estimate the cost.