Batch Duration Anomaly Detector
Flag unexplained long batches, heats, and runs
Reference implementation. Industrial use requires validation on plant-specific data. Metrics are from open evaluation splits, not customer plants.
Intended use
Flagging batches, heats, and runs that take longer than comparable ones without a clear cause, for focused investigation.
Inputs & outputs
Inputs
Batch durations, recipe and grade context, process conditions
Output
Anomalous-duration flags with likely drivers
Training data & evaluation
Training-data classification
Synthetic batch-duration records with recipe/grade covariates.
Evaluation approach
Anomaly-flag precision on a held-out synthetic split.
Precision 0.82· open split
Industrial relevance
Applies across Chemicals, Battery Materials. The value is highest where the affected asset is a constraint and additional output can be monetized.
Limitations
- Recipe and grade context is required to avoid false positives.
- Flags anomalies; root cause still needs plant investigation.