Glossary · IIoT, data and AI
Feature engineering
Also known as: Feature extraction
German: Merkmalskonstruktion
In machine learning, feature engineering is the selection, transformation and creation of input variables (features) from raw data so that a model can learn the relevant patterns more effectively. In industrial data it often draws on domain knowledge, such as frequency bands of vibration signals or cycle-based statistics.
- IIoT
- AI
In one sentence
Feature engineering selects and derives model input variables from raw data, often using domain knowledge about machines and processes.
Example
Instead of feeding raw motor current, engineers compute the RMS value, peak count and energy per cycle, which lets a simple model detect tool wear.
How it applies
- Engineering: Good features encode physical understanding: normalizing by speed or load, aligning data to machine cycles, or separating operating modes. This often matters more than the choice of algorithm.
- Deployment: Features must be computed the same way in training and in operation. Differences are a common cause of Training-serving skew.
- Documentation: Document each feature: source signals, calculation, units and rationale. This supports Explainability (AI) and makes later maintenance of the model possible.
Feature engineering vs. deep learning
Deep learning models such as Neural networks can learn features from raw data themselves, but they need more data and are harder to interpret. In industrial settings with limited labeled data, engineered features combined with simpler models are often more robust.