top of page

Case study

Root cause analysis

Discovering the causes of process faults and inefficiencies.



  • Identifying the true origins of process failures using traditional correlation-based approaches is a significant challenge. As upstream faults propagate downstream, these methods often mistake symptoms for root causes.

  • Conventional models do not generalize well to never-before-seen data points and new problems are often misdiagnosed.

  • Models lack explainability and trust from domain experts.


  • Human-guided causal graph discovery: A preliminary causal graph is defined based on the input of a domain expert. This framework is validated and continuously refined through algorithms like the PC constraint-based method.

  • Causal model building: The causal model is trained with respect to the detected anomalies (i.e. faults).

  • Intelligent decision-making: Each identified issue undergoes a root cause analysis, attributing the causal impact of individual process steps to the end product. Process steps with highest potential for improvement are highlighted.

  • On-the-spot optimization: By understanding the origins a process faults in real-time, immediate actions can be taken to mitigate these issues.

  • Reduction in downtime due to manufacturing line faults.

  • Significant improvement of the overall equipment effectiveness (OEE).

bottom of page