🌎 Language:
- 🇺🇸 English (current)
- 🇨🇱 Versión en Español
Industrial environments generate large volumes of operational data from control systems, historians, sensors, PLCs, DCS, SCADA platforms, and monitoring applications.
Despite the availability of this information, many monitoring strategies still rely heavily on static thresholds and alarm limits, making it difficult to identify early operational degradation in dynamic processes.
This project explores a contextual monitoring approach based on operational behavior rather than fixed alarm limits.
The core idea is to build an Operational Health Score (OHS) capable of quantifying how far the current process behavior deviates from an expected baseline, allowing earlier identification of abnormal conditions while reducing dependence on static thresholds.
The framework was validated using the Tennessee Eastman Process (TEP), a well-known industrial benchmark widely used in process monitoring and fault detection research.
- Context-aware operational monitoring
- Multivariable process analytics
- Early detection of operational deviations
- Reduction of false alarms
- Continuous operational scoring
- Operational visualization and decision support
- Foundation for predictive analytics applications
Industrial Data Sources
(CSV / Historian / OPC-UA / SCADA)
↓
Exploratory Data Analysis
↓
Baseline Generation
↓
Operational Health Scoring
↓
Operational Classification
↓
Temporal Monitoring
↓
Visualization & Decision Support
Instead of relying solely on fixed thresholds, the framework evaluates operational behavior relative to an expected baseline.
The Tennessee Eastman Process (TEP) was used as a validation environment.
TEP is widely used for:
- Process monitoring
- Fault detection
- Fault diagnosis
- Multivariable analytics
- Control strategy validation
Datasets used:
- TEP_FaultFree_Training.csv
- TEP_FaultFree_Testing.csv
- TEP_Faulty_Training.csv
- TEP_Faulty_Testing.csv
Operational conditions explored:
- Normal operation
- Fault 1
- Fault 4
- Fault 7
OHS = mean( abs((x - μ) / σ) )
Where:
- x = current variable value
- μ = baseline mean
- σ = baseline standard deviation
Benefits:
- Contextual interpretation of process behavior
- Reduced dependence on static thresholds
- Continuous severity representation
- Multivariable aggregation into a single operational indicator
The Power BI dashboard includes:
- Operational KPIs
- Temporal monitoring
- Operational Health Score tracking
- Comparative fault analysis
- Executive-level operational visualization
Industrial Data Source
(CSV / Historian / OPC-UA / SCADA)
↓
Python Analytics Layer
↓
Operational Health Scoring
↓
SQL Operational Layer
↓
Visualization & Monitoring
(Power BI / Grafana / BI Platforms)
- Python
- Pandas
- NumPy
- Scikit-learn
- Matplotlib
- Seaborn
- SQL
- MySQL
- Power BI
- Jupyter Notebook
- VS Code
01_operational_exploration.ipynb
02_baseline_operational_analysis.ipynb
03_operational_generalization.ipynb
04_multifault_operational_analysis.ipynb
05_operational_scoring.ipynb
06_operational_dashboard_dataset.ipynb
07_model_validation_auc_f1.ipynb
| Metric | Value |
|---|---|
| ROC-AUC | 0.787 |
| Precision | 0.993 |
| Recall | 0.505 |
| F1-Score | 0.669 |
- Dynamic industrial processes naturally exhibit variability under normal operation.
- Static-threshold monitoring can generate false positives.
- Fault conditions produce measurable changes in operational behavior.
- Context-aware monitoring improves process interpretation.
- Baseline-driven scoring provides meaningful operational context.
- High precision was achieved while maintaining a very low false alarm rate.
- Industrial process monitoring
- Chemical plants
- Mining operations
- Manufacturing systems
- Utilities and water treatment
- Energy systems
- Refrigeration systems
- Predictive maintenance initiatives
- Advanced alarm management
- Dynamic operational state classification
- Adaptive alarm logic
- Real-time streaming analytics
- Multivariable anomaly detection
- Predictive maintenance models
- Fault-type classification
- Historian integration
- OPC-UA connectivity
- MQTT integration
- Automated operational alerting
- Explainable operational analytics
- Deployment in DCS and SCADA environments
This project demonstrates that contextual operational monitoring strategies can complement traditional industrial monitoring approaches based on static thresholds.
Rather than replacing existing monitoring systems, the framework aims to provide an additional analytical layer capable of extracting richer operational insight from already available industrial data.
The long-term vision is to evolve the proof of concept into a vendor-agnostic operational analytics framework capable of integrating with multiple industrial platforms through open standards and modern data architectures.