CS-07 — Automation & AI · Data platforms
Predictive maintenance with industrial anomaly detection
Anomaly detection on telemetry from industrial assembly and welding machinery to anticipate mechanical failures, together with an on-premise IoT infrastructure for real-time ingestion and analytics.
- My role
- Data Engineer and researcher (IKERLAN).
- Period
- Not documented
- Disclosure
- Anonymised details
- Technologies
- PythonJupyterApache KafkaApache StormApache Spark
- Last reviewed
- 13 November 2026
Context
Industrial manufacturing machines with gears and components subject to pressure, force and wear. An unexpected failure could cause serious downtime, equipment damage and costly maintenance.
Challenges
- 01Detecting failures before components break.
- 02Interpreting sensor time series (force, pressure, operating cycles).
- 03Turning model outputs into useful information for maintenance.
Solution & architecture
Preprocessing of industrial telemetry, exploratory analysis, feature engineering from machine data, and training and evaluation of anomaly detection models (including deep learning where applicable). Contribution to an on-premise IoT streaming infrastructure with Kafka, Storm and Spark for real-time ingestion and analytics.
Key decisions
Not documented yet
Trade-offs
Not documented yet
Results
Results will be published once verified
Lessons learned
Not documented yet
Goal: identify potential failures early enough for preventive maintenance and reduce downtime and mechanical damage. No savings or accuracy metrics are published.
Ask about this case study
Questions here are answered with the case study “Predictive maintenance with industrial anomaly detection” as context.