Skip to content

L15: CSTR Digital Twin (Capstone Project)

Prerequisites: L00-L14 | Effort: 2-3 hours | Seborg: Chapters 1-20

Learning Objectives

  1. โœ… Build full Bronze/Silver/Gold data pipeline
  2. โœ… Implement CSTR reactor simulation with temperature control
  3. โœ… Add validation, quarantine, SCD2 tracking
  4. โœ… Calculate KPIs and control performance metrics
  5. โœ… Create production-ready digital twin

Project Overview

CSTR (Continuous Stirred Tank Reactor) Digital Twin

Bronze Layer: Raw sensor simulation - Reactor temperature, concentration, flow rates - Controller outputs - Disturbances (feed temp, flow rate) - Realistic sensor noise

Silver Layer: Validated, cleaned data - Data quality tests (range, rate-of-change) - Quarantine bad readings - SCD2 tracking for controller tuning changes

Gold Layer: Business metrics - Reaction conversion % - Energy efficiency (kJ/kg product) - Controller performance (IAE, settling time) - Alarm events

CSTR Dynamics

Mass Balance: 

dC/dt = F/Vยท(C_in - C) - kยทC

Energy Balance: 

dT/dt = F/Vยท(T_in - T) + (-ฮ”H)ยทkยทC / (ฯยทCp) + Q_jacket / (ฯยทCpยทV)

Reaction Rate: 

k = k0ยทexp(-E/(RยทT))  (Arrhenius)

Controller: - PID controls jacket heat duty (Q_jacket) to maintain reactor temp - PI ratio control maintains feed ratio

Odibi Implementation

Bronze: Raw Sensor Data

# cstr_digital_twin.yaml
nodes:
  - name: bronze_cstr_sensors
    read:
      format: simulation
      options:
        simulation:
          scope:
            start_time: "2024-01-01T00:00:00Z"
            timestep: "30sec"
            row_count: 2880  # 24 hours
            seed: 42

          entities:
            count: 1
            id_prefix: "CSTR-"

          columns:
            # (Full CSTR model with PID control)
            # See working example for complete implementation

          chaos:
            outlier_rate: 0.01
            duplicate_rate: 0.005

    write:
      connection: local
      format: parquet
      path: data/output/bronze_cstr_sensors.parquet
      mode: overwrite

Silver: Validated Data

  - name: silver_cstr_validated
    read:
      connection: local
      format: parquet
      path: data/output/bronze_cstr_sensors.parquet

    validate:
      - name: temp_range_check
        test: range
        column: reactor_temp_degc
        min: 50
        max: 100
        on_fail: fail
        quarantine: true

      - name: flow_rate_check
        test: rate_of_change
        column: feed_flow_gpm
        max_change: 10
        on_fail: warn

    transform:
      - name: scd2_controller_tuning
        transformer: scd2
        natural_key: entity_id
        tracked_columns:
          - Kp
          - Ki
          - Kd

    write:
      connection: local
      format: parquet
      path: data/output/silver_cstr_validated.parquet
      mode: overwrite

Gold: KPIs

  - name: gold_cstr_kpis
    read:
      connection: local
      format: parquet
      path: data/output/silver_cstr_validated.parquet

    transform:
      - sql: >
          SELECT
            DATE_TRUNC('hour', timestamp) as hour,
            AVG(conversion_pct) as avg_conversion,
            AVG(abs(temp_sp_degc - reactor_temp_degc)) as avg_temp_error,
            SUM(energy_kwh) as total_energy_kwh,
            SUM(CASE WHEN fault_detected THEN 1 ELSE 0 END) as fault_count
          FROM __this__
          GROUP BY hour
          ORDER BY hour

    write:
      connection: local
      format: parquet
      path: data/output/gold_cstr_kpis.parquet
      mode: overwrite

Working example: /examples/cheme_course/L15_cstr_digital_twin/cstr_digital_twin.yaml

Project Extensions

  1. Add cascade control: Inner loop on jacket temp
  2. Feedforward: Compensate for feed temp changes
  3. Alarm management: High temp, low conversion
  4. Batch tracking: SCD2 on batch ID
  5. OpenLineage: Track data lineage

Summary

Key Achievements: - โœ… Built full Bronze/Silver/Gold pipeline - โœ… Simulated realistic CSTR with control - โœ… Applied validation, quarantine, SCD2 - โœ… Calculated business KPIs - โœ… Production-ready digital twin

You've completed the ChemE Data Engineering course!

You now can: - Simulate process control systems - Build data pipelines for process data - Apply validation and quality gates - Generate synthetic data for ML - Understand control engineering from a data perspective

Additional Resources