Extending catalyst lifetime with predictive deactivation modelling
Forecasts of the days until a catalyst repack, within about 15 days, so repacks are planned instead of forced.
Challenge
A producer of aniline catalysts suffered unplanned shutdowns from catalyst deactivation, each requiring a costly repack. The activation state cannot be measured directly, repack events are rare, and normal and deactivating operation look different in the data.
Solution
Arundo built a multi-stage workflow: survival analysis confirmed the strongest proxy for failure risk, regression models emulate lab results with 6% error so the catalyst state is estimated continuously, exponential fits translate predicted lab values into days until repack, and the gap between predicted and actual hydrogen partial pressure flags the onset of deactivation.
Impact
Time-to-repack forecasts within about ±15 days on historical cycles, enabling planned repacks and continuous production.