Plain-language summary
- What it solves
- Spotting a fault in a chemical plant before it becomes a problem, and knowing how often the system is wrong.
- Who it can serve
- Anyone working in industrial monitoring, and anyone studying anomaly detection.
- What it produces
- Plant status, detected and missed anomalies, false alarms and detection delay. On simulated data, not real.
Study industrial anomaly detection in a controlled environment before any real-plant application.
3 · What it produces
State, anomaly score, contributors, detected/missed events, recall, false alarms, delay and benchmark readiness.
The cards expose multiple views of the same workflow so result, method, evidence and history remain distinct. The current benchmark uses synthetic data (simulated data).
4 · How it works
Multivariate monitoring + frozen synthetic data (simulated data) evaluation era
1generic CSTR multivariate sensors, noise, missing data, step/drift/intermittent faults
2drift-aware ensemble: high-specificity PCA T²/Q + steam, cooling and reaction residuals + slow steam-residual CUSUM; Isolation Forest/autoencoder as comparators; fault classifier
3parameters frozen on a separate development simulation; same frozen 28-day synthetic data (simulated data) benchmark; event recall, missed rate, false-alarm episodes/24h, P50/P90 delay, v1.4.16 ensemble comparison and anti-leakage
4Persistence, rendering and governed publishing.
5 · Data and signals
| Element | Detail |
|---|---|
| Data and signals | generic CSTR multivariate sensors, noise, missing data, step/drift/intermittent faults |
| Industrial transferability | condition monitoring, early warning and assisted diagnosis |
| Research | detection-delay/false-alarm trade-off and model gating |
| Interpretation condition | performance applies only to the stated benchmark/simulator |
drift-aware ensemble: high-specificity PCA T²/Q + steam, cooling and reaction residuals + slow steam-residual CUSUM; Isolation Forest/autoencoder as comparators; fault classifier
parameters frozen on a separate development simulation; same frozen 28-day synthetic data (simulated data) benchmark; event recall, missed rate, false-alarm episodes/24h, P50/P90 delay, v1.4.16 ensemble comparison and anti-leakage
Python, NumPy, pandas, scikit-learn, SQLite, Matplotlib
performance applies only to the simulator; test-bed-specific process residuals and an uncalibrated classifier require validation before real-plant transfer
Workflow technical dossier
Methods, checks, stack and transferability conditions.
Data and signals
generic CSTR multivariate sensors, noise, missing data, step/drift/intermittent faults
Inference / calculation
drift-aware ensemble: high-specificity PCA T²/Q + steam, cooling and reaction residuals + slow steam-residual CUSUM; Isolation Forest/autoencoder as comparators; fault classifier
Validation
parameters frozen on a separate development simulation; same frozen 28-day synthetic data (simulated data) benchmark; event recall, missed rate, false-alarm episodes/24h, P50/P90 delay, v1.4.16 ensemble comparison and anti-leakage
Stack
Python, NumPy, pandas, scikit-learn, SQLite, Matplotlib
Limitations
performance applies only to the simulator; test-bed-specific process residuals and an uncalibrated classifier require validation before real-plant transfer
See recent outputs
The Publications page collects local previews and, on request, the public feed.
Potential transferability
Have a problem similar to Process Sentinel?
A first discussion can start from the objective, available data and success criterion without sending confidential information. A result observed here is not assumed to transfer automatically to another context.