Low NO_x Configurations in an Industrial Boiler via Genetic Algorithm & CFD

Overview

Industrial coal-fired power plants must balance two conflicting objectives: reducing nitrogen-oxide (NO_x) emissions to comply with increasingly strict environmental legislation, and preserving the mechanical integrity of the boiler’s water-wall tubes, which are exposed to aggressive corrosion under reducing atmospheres created by low-NO_x operating strategies.

This project demonstrates how Genetic Algorithms (GA) combined with Computational Fluid Dynamics (CFD) can automatically explore the enormous configuration space of a 600 MW corner-fired boiler and identify settings that achieve the best trade-off between pollutant reduction and operational safety.

Boiler Description

EDF operates three 600 MW tangentially-fired pulverized-coal units in France (Cordemais 4 & 5, Le Havre 4). Key characteristics:

• Total height: ~80 m; cross-section: 16.6 × 16.6 m
• Four firing corners (A1–A4), three burner elevation groups (Group 1, 2, 3)
• Six mills (A–F) feeding coal to the burners; typically four mills active at full load
• Adjustable parameters: air-damper positions (on/off), coal mass-flow per mill, vertical tilt of burner nozzles (−30° to +30°)

CFD Model (Code_Saturne)

All boiler configurations are evaluated with Code_Saturne, EDF’s in-house finite-volume solver, on a structured mesh of 470,000 cells.

Genetic Algorithm

The GA is implemented with the ParadisEO library (INRIA). Each individual (chromosome) encodes a complete boiler operating point:

• Binary genes — on/off state of each air/secondary-air nozzle
• Integer genes — vertical tilt angles for corners A1/A3 and A2/A4
• Continuous genes — coal mass-flow rate per mill (0 – 100 % capacity)

Cost Function

Rather than optimising a single metric, a composite economic cost (€/year) penalises:

1. O₂ deviation — incomplete combustion penalty if O₂ departs >0.5 vol.% from target
2. CO in flue gas — heavy penalty above 100 ppm@6%O₂
3. NO_x emissions — SCR ammonia & catalyst savings vs. penalty above 200 mg/Nm³@6%O₂
4. Water-wall corrosion — metal-loss model (EPRI/PowerGen UK equations) as a function of local CO concentration and metal temperature; coating cost when wastage > 50 µm/year
5. Heat-flux heterogeneity — forced-outage penalty when flux variance exceeds a threshold
6. Unburned carbon — ash-recycling penalty when carbon-in-ash > 7 %

Numerical Setup

Key Results

Low-NO_x Configurations

Most high-performing configurations share a common pattern:

• Four mills (A, B, C, D) feeding the lower and middle groups only — concentrates fuel-rich zones in the lower furnace, promoting fuel-N reduction before additional OFA air is introduced.
• Horizontal burner tilt (0°) — upward tilt consistently increases NO_x; when the sum of both tilt angles exceeds +45°, NO_x hardly drops below 900 mg/Nm³@6%O₂.
• Selective opening of FOE, FOF, FOO-up nozzles — staged air injection reduces local O₂ near active burners; the GA found non-intuitive asymmetric patterns (e.g. opening nozzles only at two diagonally opposite corners).

NO_x vs. Corrosion Trade-off

• Configurations below 500 mg/Nm³@6%O₂ almost always require industrial coating to protect water-wall tubes.
• In the 500–600 mg/Nm³ range, many individuals achieve a low corrosion cost, making them the best compromise for long-term operation.

Burner Tilt & Heat Flux

• Downward or horizontal tilting concentrates heat on the ash hopper, increasing average flux and its spatial variance.
• Upward tilting reduces both quantities — consistent with on-site observations where operators tilt burners up to relieve temperature peaks on water-wall tubes.

Unburned Carbon

• A classic NO_x–burnout trade-off is observed: deeper air staging raises unburned carbon.
• With the coal blend used (high volatile, moderate N content), carbon-in-ash remains below 2 % even in deep staging, always permitting fly-ash recycling.

Validation Against On-Site Measurements

Results were validated during two test campaigns:

Both campaigns confirmed low corrosion risk for moderate staging, and high CO near walls for deep staging — fully consistent with GA/CFD predictions.

Conclusions & Perspectives

• Genetic algorithms + CFD can automatically discover non-intuitive boiler settings that a human operator would not easily find through trial and error.
• The composite cost function successfully steers the search away from configurations that are good for NO_x but catastrophic for corrosion.
• Future work: (1) couple a thermo-hydraulic tube model to predict metal temperature peaks more accurately; (2) extend to partial loads and variable coal blends; (3) replace the scalar cost function with a proper multi-objective Pareto formulation.

Citations

@Article{DalSecco2015,
  abbr         = {Article},
  author    = {Dal Secco, Sandro and Juan, Olivier and Louis-Louisy, Myriam and Lucas, Jean-Yves and Plion, Pierre and Porcheron, Lynda},
  journal   = {Fuel},
  title     = {Using a genetic algorithm and CFD to identify low NOx configurations in an industrial boiler},
  year      = {2015},
  pages     = {672--683},
  volume    = {158},
  doi       = {10.1016/j.fuel.2015.06.021},
  publisher = {Elsevier},
}
@InProceedings{DalSecco2010,
  abbr         = {Proceedings},
  author    = {Dal Secco, Sandro and Juan, Olivier and Lucas, Jean-Yves and Plion, Pierre},
  booktitle = {International conference on metaheuristics and nature inspired computing},
  title     = {Evolutionary algorithm for pollutant emission minimization in coal power plants},
  year      = {2010},
}