Steel Blast Furnace Gas Recovery and CCUS Integration Workflow

This workflow solves the high-stakes optimization problem of routing BFG to maximize captured CO2 against energy penalties and market prices. By making real-time decisions on whether to send gas to the power plant, heating ovens, or the new capture unit, it directly increases the net carbon abatement per ton of steel, turning a compliance cost into a strategic asset. The architecture uses reinforcement learning agents integrated with plant historians and carbon market APIs to evaluate trade-offs continuously.

Automating the BFG routing decision eliminates the suboptimal, rule-of-thumb allocations that waste high-calorific gas. The workflow dynamically balances internal consumption (for plant heat and power) against external sales or capture, prioritizing the highest-value use at any given moment. This reduces the need to purchase supplemental natural gas or grid electricity, directly lowering the operating cost per ton of steel. Integration with the DCS and power plant control system is critical for safe, closed-loop execution.

Manually integrating a carbon capture unit into the volatile BFG system creates operational risk and delays value capture. This custom workflow acts as an intelligent orchestration layer, safely ramping up capture operations by modeling gas composition and volume impacts before making changes. It provides operators with a predictable, controlled interface to the new asset, shortening the learning curve and time-to-positive-cash-flow for the capital-intensive CCUS project.

The workflow automates the complex emissions accounting required when BFG is split between combustion (Scope 1) and capture. Agents tag each cubic meter of gas with its ultimate fate, calculate resultant emissions or abatement, and populate live compliance dashboards and audit trails. This continuous MRV (Measurement, Reporting, Verification) readiness slashes the quarterly manual effort for environmental reporting and provides defensible data for carbon credit verification.

Beyond routing logic, the workflow includes diagnostic agents that monitor the health of the entire BFG network—from gas holders and compressors to the capture unit's absorber. By analyzing pressure, flow, and composition telemetry against digital twin simulations, it predicts fouling, leaks, or equipment degradation. It automatically generates prioritized work orders in the CMMS (like SAP PM or Maximo), preventing unplanned downtime that would force wasteful gas flaring.

This implementation is not a point solution but an architectural foundation. The agentic layer, built on frameworks like LangGraph or custom orchestrators, establishes a pattern for integrating real-time optimization, safety interlocks, and human oversight. It becomes the central nervous system for future decarbonization projects—such as hydrogen injection or more advanced capture technologies—turning the steel mill's operational data into a persistent competitive advantage in a carbon-constrained market.

This agent ingests real-time data from gas chromatographs, flow meters, and calorimeters across the BFG network. It validates data quality, calculates the available energy (GJ/h) and carbon content (tCO2/h) of the gas stream, and serves as the single source of truth for downstream routing decisions. It flags instrument drift or calibration issues to maintenance systems.

The core decision engine. This agent runs a constrained optimization model every 30-60 seconds, weighing: 1) Power Market Price (for on-site power plant dispatch), 2) Process Heat Demand (from coke ovens, slab reheat furnaces), 3) Carbon Capture Unit (CCU) Capacity, and 4) Flaring Penalties. It outputs setpoints for valve banks and dampers to direct BFG flows, maximizing either total economic value or carbon abatement based on the plant's operating mode.

This agent manages the interface with the carbon capture unit. It monitors the CCU's solvent regeneration cycle, compressor status, and CO2 product buffer tank levels. It modulates the BFG feed rate to the CCU to keep it operating at peak efficiency, and can signal the Routing Optimizer to temporarily divert gas if the CCU is down for maintenance, preventing flaring. It also tracks captured CO2 purity for offtake compliance.

A guardrail agent that operates in parallel. It continuously checks all routing decisions against hard safety limits (e.g., minimum gas holder levels, boiler safe operating windows) and environmental permit conditions (e.g., hourly flaring limits, SOx emissions from combustion). It can override the Optimizer and initiate safe fallback states. It also auto-generates compliance logs for BFG utilization and flaring events.

Handles the non-steady state. When the Monitoring Agent detects a fault or the plant DCS signals an equipment trip (e.g., power boiler outage), this agent executes pre-defined contingency workflows. It recalculates feasible gas routes, generates prioritized work orders in the CMMS (e.g., SAP, IBM Maximo), and dispatches alerts to shift supervisors and maintenance crews with context-specific SOPs.

A physics-informed simulation layer used for planning and offline analysis. Engineers can use this agent to run 'what-if' scenarios—like adding a new hydrogen injection point or a second capture train—on a calibrated digital twin of the BFG network. It assesses impacts on gas quality, heating value, and carbon balance, providing data to justify capital projects or major operational shifts.

The core decision agent ingests live BFG pressure, calorific value, and plant-wide demand signals from the Distributed Control System (DCS). It runs a continuous linear programming model to allocate gas between the power plant (CCPP), hot stoves, and the new carbon capture unit, prioritizing the highest marginal value (€/GJ) or carbon abated (tCO₂/GJ) based on dynamic price and penalty inputs.

A dedicated control agent interfaces with the amine-based capture plant's PLC to modulate solvent circulation and stripper steam based on incoming BFG composition and flow. It ensures the capture unit operates at its most efficient point, minimizing the energy penalty while meeting target capture rates. The agent also handles safe startup/shutdown sequences during mill disruptions.

An external market agent connects via API to day-ahead and intraday electricity markets (e.g., EPEX Spot) and natural gas TTF indexes. It feeds real-time price forecasts into the routing engine, enabling the system to opportunistically shift BFG to power generation during high-price periods or reduce CCPP output when carbon capture is more valuable, creating a dynamic revenue optimization layer.

A safety-critical orchestration layer built with LangGraph enforces hard operational constraints. It prevents gas routing that would violate boiler safety limits, heating furnace minimum temperatures, or permitted stack emissions. All major setpoint changes require validation against a physics-informed digital twin before execution, with human-in-the-loop approval gates for novel operating regimes.

Reporting agents automatically pull validated data from Continuous Emissions Monitoring Systems (CEMS), gas meters, and the capture plant to calculate real-time Scope 1 emissions. They map abated CO₂ to the relevant carbon credit methodology (e.g., CDM, VERRA) and generate verification-ready data packages, feeding directly into ERP systems (SAP) for internal carbon pricing and financial reporting.

Agents monitor telemetry from BFG compressors, valves, and capture unit heat exchangers. Using vibration and performance models, they predict failures and automatically generate detailed work orders in the Computerized Maintenance Management System (CMMS, e.g., IBM Maximo). The system schedules maintenance during planned mill stoppages, preventing unplanned downtime that would force BFG flaring.