Case studies/Causal AI Case-Study
Root-cause analysis for production downtime.
Industry · 2024
Challenge
Conventional correlation-based methods often confuse symptoms with the true causes of process faults. New fault patterns were frequently misdiagnosed, and the models lacked the trust of domain experts.
Solution
A causal graph was defined together with domain experts and continuously refined with algorithms such as the PC method. The causal model is trained on detected anomalies and performs a root-cause analysis for each fault that quantifies the contribution of individual process steps.
Impact
Fewer downtime events from production faults and a clear improvement in overall equipment effectiveness (OEE).