Skip to content
Mustafa Tarhan

Predictive Maintenance for Machine Axes

Accelerometers on linear axes, feature extraction on a Raspberry Pi and an SVM that classifies the wear state of each axis.

Industrial AI · work project2024 – 2026In progress
Role
Working-student project at Fraunhofer IPK; also the basis of my bachelor thesis.
Classifier
SVM, grid search
Sensors
3-axis accelerometers
Stack
Python, scikit-learn, NumPy / SciPy, Raspberry Pi, LIS3DH accelerometers, MySQL, MQTT, ThingsBoard

What it does

A demonstrator that detects wear on the linear axes of a machine. Accelerometers on each axis feed a Raspberry Pi backend that records labelled and unlabelled measurement series, extracts statistical features and classifies the wear state with a support vector machine; results go to a dashboard and an IoT platform. A later rework moved storage from a database to local CSV files with the same schema, removed a second feature calculation that produced different values than the training data, and added a script that records controlled experiment series with an undamaged axis as a built-in control.

  1. Sensors
    • Accelerometers
    • Raspberry Pi
  2. Features
    • Statistical features
  3. Model
    • SVM
    • Wear state
  4. Output
    • Dashboard
    • MQTT

The hard part

Making the live classification compute exactly the same features as the training data, so models trained on older measurements stay valid on new ones.