Tata Steel let a predictive controller run its annealing furnace, cutting fuel 8%

Continuous annealing sets the mechanical quality of cold-rolled steel, the highest-value stream at Tata Steel and the grade automakers buy. The furnace has slow temperature dynamics, yet the strip's mass flow keeps changing as one steel grade follows another. The standard fix — PID controllers that react after temperatures drift — produced off-specification strip exactly at the grade transitions.

Tata Steel and IIT Bombay built a model predictive control (MPC) supervisory layer that solves a finite-horizon optimal-control problem every minute, steering furnace temperatures in anticipation of transitions while respecting operational constraints. The system went fully operational in January 2023.

The results were measured, not estimated: 20% of annealed products met premium quality standards, and fuel consumption and CO2 emissions each fell 8% per ton of steel annually. The project made Tata Steel and IIT Bombay finalists for the 2024 INFORMS Franz Edelman Award and inductees into the Franz Edelman Academy.

Predict the furnace a minute ahead instead of reacting to it

The dull method was tuning PID loops harder — they still only react after the damage is done.

MPC exploits the furnace's predictability: transitions are scheduled, so the disturbance is known before it arrives.

Quality and fuel were treated as one optimization, not two departments trading blame.

The benefits are physical and audited: 8% less fuel per ton and 20% of output at premium grade.

Slow industrial processes beat reactive control: when the plant lags minutes behind the dial, the winner plans ahead — predict the transition, don't correct it afterward.

Fully operational since January 2023, the MPC system keeps the annealing line within constraints during grade changes. The work won a 2024 Franz Edelman finalist award, inducted Tata Steel and IIT Bombay into the Franz Edelman Academy, and strengthened Tata Steel's position in automotive steel.