The heat treatment of large steel forgings plays a crucial role in achieving the desired mechanical properties essential for various industrial applications, including aerospace, automotive, and heavy machinery. The tempering process, a key stage in heat treatment, significantly influences material performance by modifying microstructures to enhance toughness, hardness, and wear resistance. However, maintaining uniform temperature distribution within large-scale industrial electric furnaces remains a critical challenge due to complex thermal interactions, large temperature gradients, and energy inefficiencies. Traditional empirical approaches for optimizing the heat treatment process are often inadequate, as they fail to account for the intricate interplay between furnace design, loading configurations, and process parameters. In recent years, computational fluid dynamics (CFD) and machine learning (ML) have emerged as powerful tools for analyzing and optimizing industrial heat treatment processes. Despite their potential, existing research primarily focuses on small-scale systems or gas-fired furnaces, leaving industrial-scale electric furnaces underexplored. Addressing these gaps, this study integrates experimental measurements, CFD simulations, multi-objective optimization, and ML-based predictive modeling to enhance temperature uniformity, reduce processing times, and improve energy efficiency in industrial electric heat treatment furnaces. By systematically evaluating stacking patterns, heating element layouts, and predictive loading strategies, this research provides a comprehensive framework for improving process control and achieving consistent mechanical properties in large steel components.
A transient three-dimensional CFD model was developed to simulate the thermal and flow characteristics within a 112 m³ car-bottom industrial electric furnace. The furnace, located at Finkl Steel (Sorel, Quebec), was equipped with multiple electrical heating elements and axial fans to ensure forced convection, and its thermal behavior was analyzed under various loading configurations. The CAD geometry of the furnace and forging blocks in different scenarios were designed using CATIA V5 to ensure high-fidelity representation of industrial conditions. To validate the model, experimental temperature measurements were obtained from industrial forging blocks using embedded thermocouples strategically placed at critical surface positions. These measurements captured real-time thermal gradients, allowing direct comparison with simulation results. ANSYS® (versions 2022 and 2023) was employed for CFD simulations and meshing, while JMatPro® was employed to determine the temperature-dependent mechanical properties of the forging material, ensuring accurate thermophysical inputs for CFD simulations. The CFD analysis revealed significant temperature non-uniformities of up to 300 K in conventional stacking patterns, particularly in multi-block arrangements where airflow obstructions and inadequate radiative exchange contributed to localized overheating and underheating. The integration of CFD simulations with real-time industrial data provided the foundation for optimizing furnace loading strategies and heating element placements. For visualization and post-processing of simulation results, Tecplot 360 EX was utilized to interpret temperature fields and flow characteristics effectively.
Multi-objective optimization techniques were employed to refine the furnace operating parameters, focusing on minimizing temperature differentials while maintaining processing efficiency. A genetic algorithm (GA) and Pareto-based optimization framework were implemented to explore the impact of heating element layout on overall temperature uniformity. MATLAB 2021 was used to implement the optimization routines, enabling efficient execution of genetic algorithms and Pareto searches. The surrogate modeling approach, using polynomial regression, enabled rapid evaluation of various design configurations without requiring full CFD simulations for each iteration. Results demonstrated that optimized heating element layouts could reduce surface temperature variations by 8%, while improved stacking configurations reduced core-to-surface temperature differentials by 35%. These findings emphasize the importance of systematic furnace design improvements in achieving uniform heating across large-scale steel forgings.
To further enhance operational efficiency, a machine learning predictive model was developed using an extensive dataset of over 1,100 industrial tempering logs. Python 3.1 was employed to develop and train a predictive model, utilizing robust preprocessing techniques, including feature engineering, transformation methods, and hyperparameter tuning. The XGBoost regressor was trained to predict optimal loading cycles based on key operational variables, including forging dimensions, material composition, and furnace thermal history. By analyzing historical process data, the model provided real-time recommendations for loading configurations, significantly reducing reliance on empirical heuristics and manual scheduling. The ML model achieved an R² of 0.78–0.89, demonstrating its effectiveness in accurately predicting optimal loading parameters while reducing energy consumption and processing time. The integration of ML-based predictive analytics into furnace operations enhances adaptability, allowing operators to optimize throughput without requiring extensive trial-anderror adjustments.
The findings of this research underscore the potential of combining experimental validation, high-fidelity CFD modeling, multi-objective optimization, and data-driven machine learning to improve the efficiency of industrial electric heat treatment furnaces. The validated CFD framework serves as a powerful tool for scenario analysis, enabling precise evaluations of different furnace configurations and loading strategies. Meanwhile, the ML model enhances real-time decision-making by providing rapid, data-driven insights into optimal furnace operations. The integration of these methodologies establishes a holistic approach to improving heat treatment processes, offering a scalable and computationally efficient solution for industrial applications.
Future research directions include extending machine learning frameworks to predict final mechanical properties based on thermal history, further refining CFD turbulence and radiation models to improve simulation accuracy, and exploring real-time adaptive optimization techniques for dynamically adjusting furnace operating conditions. By leveraging the synergy between physics-based modeling and data-driven analytics, this research contributes to the advancement of predictive and prescriptive control strategies in metallurgical heat treatment, facilitating energy-efficient, high-quality production in industrial settings.
| Date | 3 Sept 2025 |
|---|
| Original language | American English |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Mohammad Jahazi (Supervisor) & Farzad Bazdidi-Tehrani (Co-supervisor) |
|---|