Optimization of temperature distribution in continuous annealing furnaces using CFD and machine learning to achieve a more uniform temperature field

• Integrated CFD and ML approach optimizes temperature distribution in continuous annealing furnaces. • MLP neural network trained on 200 CFD samples enables rapid prediction of thermal behavior. • Improved NSGA-II algorithm with amplification factor and retention strategy enhances multi-objective optimization. The continuous annealing furnace is a key piece of equipment in steel strip processing, and its inner liner is crucial for the efficient, stable, and safe operation of the annealing process. However, in actual production, many steel manufacturers face challenges related to internal temperature field non-uniformity in their natural gas-fired continuous annealing lines. This research integrates numerical simulation and machine learning techniques to optimize the temperature distribution in continuous annealing furnaces for steel strips, specifically targeting the mitigation of internal temperature field non-uniformity. A three-dimensional geometric model of the combustion chamber liner was constructed, and the Realizable k - ε model, Discrete Ordinates (DO) radiation model, and Finite Rate/Eddy Dissipation Model (EDM) were used for the simulation. Building upon the CFD-derived thermal field characteristics, this study employs a data-driven Multi-Layer Perceptron (MLP) model trained on 200 CFD-sampled combustion scenarios to establish high-fidelity mappings between operational parameters. Key parameters such as the fuel equivalence ratio, fuel inlet flow rate, exhaust outlet pressure, and inner liner height were trained, and the Latin Hypercube Sampling (LHS) method to generate training samples for developing a MLP neural network model. Moreover, the MLP model was combined with an improved Non-dominated Sorting Genetic Algorithm II (NSGA-II), and an amplification factor and retention strategy were introduced to optimize the key parameters. Through strategic adjustments: this configuration achieves remarkable improvements—reducing the average bottom temperature of the inner liner by 55.45 K while shortening the steady-state attainment time by 486 s, thereby ensuring enhanced thermal uniformity across the system. The research offers a robust, data-driven framework for steel manufacturers to optimise annealing furnace operations, resulting in substantial reductions in energy consumption, increased production throughput, and enhanced strip quality consistency.