⚡ electrical engineering

Heat-transfer-constrained virtual assessment of thermal penetration in industrial aluminum-alloy soaking furnaces without workpiece-temperature measurement

This study introduces PC-VHSA, a physics-constrained virtual analysis method that estimates thermal penetration and identifies boundary anomalies in industrial aluminum-alloy soaking furnaces without direct workpiece-temperature measurements by utilizing multi-zone furnace data, thermophysical properties, and a novel heat-through risk index validated through energy consistency and sensitivity analysis.

Original authors: Jie Hou, Xiaohui Li, Li Li, Jianping Tan

Published 2026-09-30

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Jie Hou, Xiaohui Li, Li Li, Jianping Tan

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the world of heavy industry, aluminum is rarely used in the form it comes from the smelter. To turn raw metal into the strong, uniform sheets needed for airplanes, cars, or machinery, engineers must first heat large cylindrical blocks of aluminum, known as billets, to a precise temperature. This process, called soaking, is critical. If the metal is not heated evenly throughout its entire volume, the internal structure remains weak and inconsistent, leading to cracks or failures later on. The goal is to ensure that the very center of the massive metal cylinder reaches the same temperature as its outer surface. However, a significant problem plagues this industry: while factories can easily measure the temperature of the air inside the furnace, they cannot practically stick a thermometer into the middle of a glowing, multi-ton block of metal while it is being processed. For decades, operators have had to guess whether the center of the metal was ready, relying on timers and furnace readings that might not reflect the true state of the material inside.

Researchers at Central South University have developed a new way to solve this guessing game without ever needing to measure the metal directly. They created a method that acts like a virtual thermometer, using the laws of physics to calculate what is happening inside the aluminum based on what is happening outside. Instead of trying to predict the future or rely on complex computer simulations that require perfect data, their approach builds a digital model of the heat moving through the metal. They take the recorded temperatures from the furnace walls, the size and weight of the metal blocks, and the known properties of the specific aluminum alloy, and they run these numbers through a model that mimics how heat naturally travels from the surface to the core. This allows them to estimate the temperature at the center of the billet and determine if it has been heated enough, all without a single physical sensor touching the metal.

The team tested this method, which they call a physics-constrained virtual thermal-state analysis, using real production data from 2024 involving hundreds of batches of aluminum. They found that for the vast majority of cases, the method worked with high precision. In the batches where the data was clean and reliable, the virtual model showed that the center of the metal was almost exactly where it needed to be. The average difference between the estimated center temperature and the target temperature was less than one degree Celsius. This tiny gap suggests that the current industrial processes are generally doing a very good job of heating the metal through, a fact that might have been missed if operators were only looking at the furnace air temperature. The researchers also discovered that the method could spot when something was wrong with the data itself. In about one-fifth of the batches, the model flagged the records as unreliable, identifying issues like missing temperature readings or inconsistent energy logs, rather than falsely blaming the heating process. This ability to separate real thermal problems from bad data is a crucial step forward, as it prevents factories from wasting time trying to fix a process that is actually working fine.

One of the most valuable findings was that the method could detect subtle differences in how different types of aluminum or different furnaces performed. While all the reliable batches were deemed sufficiently heated, the model showed that some specific combinations of furnace and alloy had a slightly smaller safety margin than others. This means that while the metal was safe to use, those specific conditions were closer to the edge of being insufficient. By identifying these subtle risks, the method gives factory managers a way to fine-tune their operations, ensuring that every single block of metal is perfectly prepared, regardless of its size or alloy type. The study confirms that it is possible to understand the hidden thermal state of industrial materials by combining simple physical laws with the data already available in factory logs. This approach offers a clear, reliable path to ensuring quality in metal production, turning invisible heat into visible, actionable information without the need for invasive or impossible measurements.

Technical Summary: Physics-Constrained Virtual Thermal-State Analysis (PC-VHSA)

1. Problem Definition

Industrial aluminum-alloy soaking furnaces require homogenization treatments to reduce segregation, dissolve low-melting phases, and improve mechanical properties. The critical quality metric is whether the billet center achieves sufficient thermal penetration (reaching the target temperature). However, in real-world industrial settings, continuous online measurement of the workpiece center temperature is unavailable.

Current assessment methods face significant limitations:

Furnace Temperature Compliance: Relying solely on furnace thermocouples ignores the radial heat-conduction lag inherent in large-diameter billets, potentially masking insufficient center heating.
Manual Records: Using manually recorded soaking durations fails to account for dynamic furnace fluctuations, stage-switching criteria, and zone non-uniformity.
Energy Consumption: Total energy usage is influenced by loading mass, heat loss, and control strategies, making it an unreliable direct proxy for thermal penetration.
Data Scarcity: Without measured workpiece temperatures as ground-truth labels, conventional supervised machine learning models for temperature prediction cannot be validated or trained effectively.

The core challenge is to establish a virtual thermal-state assessment method that is constrained by heat-transfer mechanisms, bounded by reliability rules, and interpretable for engineering use, despite the absence of direct material temperature measurements.

2. Methodology: PC-VHSA

The authors propose PC-VHSA (Physics-Constrained Virtual Thermal-State Analysis), a framework that transforms available industrial data into a physically constrained virtual thermal state.

2.1 Boundary Construction

Equivalent Furnace Temperature: Instead of using a single sensor, the method constructs a robust boundary by taking the median of multi-zone furnace temperatures. This reduces sensitivity to local sensor noise and outliers.
Geometric and Mass Consistency: Billets are modeled as cylinders. The method calculates total volume from recorded diameter, length, and quantity, then back-calculates an equivalent density. A consistency ratio ( $C_m$ ) is used to verify that geometric records match the recorded mass; batches with inconsistent records are flagged.
Thermophysical Properties: Material-specific thermal conductivity ( $k$ ), specific heat ( $c$ ), and density ( $\rho$ ) are incorporated to define thermal diffusivity ( $\alpha$ ), ensuring the model accounts for alloy-specific heat transfer characteristics.

2.2 Virtual Thermal State Modeling

A one-dimensional radial heat conduction model in cylindrical coordinates is solved using the equivalent furnace temperature as a convective-radiative boundary condition.

Outputs: The model estimates the virtual center temperature ( $T_c$ ), virtual surface temperature ( $T_s$ ), and the surface-center temperature difference ( $G$ ).
Key Metrics:
- Center-Temperature Deficit ( $D$ ): The difference between the soaking set temperature and the virtual center temperature.
- Soaking Sufficiency Ratio ( $A$ ): The proportion of the soaking stage where the deficit is within an allowable threshold.
- Heat-Through Ratio ( $\phi$ ): The ratio of the terminal center temperature to the set temperature.

2.3 Heat-Through Risk Index (HRI)

To comprehensively characterize thermal margins, a composite index (HRI) is developed. It integrates normalized values of:

Terminal center-temperature deficit.
Average deficit in the final 30 minutes.
Maximum surface-center temperature difference.
Insufficient soaking ratio.
Furnace-zone temperature difference (non-uniformity).

2.4 Reliability Control Mechanism

Since direct error validation is impossible, the method employs a multi-layered reliability filter to distinguish between interpretable thermal sufficiency and data/boundary anomalies:

Energy Consistency: Compares estimated material heat absorption against total energy consumption to ensure the virtual state falls within a physically reasonable energy boundary.
Stage Consistency: Analyzes the alignment between manually recorded soaking times and SCADA-identified effective soaking stages.
Boundary Anomaly Identification: Flags batches with missing furnace temperature data, severe stage-record anomalies, or unreasonable energy ratios. These batches are excluded from thermal sufficiency conclusions to prevent misinterpreting data quality issues as process failures.

3. Experimental Results

The method was validated using 2024 industrial production data from aluminum-alloy homogenization furnaces.

Data Scope: 385 valid batches were cleaned; 369 were modelable (satisfying geometric and boundary requirements).
Reliability Filtering:
- 290 batches (78.59%) met all reliability conditions and were classified as the "reliable subset."
- 79 batches (21.41%) were identified as data or boundary anomalies and excluded from thermal sufficiency conclusions.
Thermal Sufficiency Findings (Reliable Subset):
- All 290 reliable batches were assessed as thermally sufficient.
- Median Terminal Center-Temperature Deficit: 0.77 °C.
- 90th Percentile Deficit: 1.19 °C.
- Median HRI: 1.59 (indicating a low-risk range).
- The results suggest that under current operating regimes, the center temperature generally approaches the target state with a sufficient margin.
Furnace and Alloy Variations:
- Differences in HRI were observed across different furnaces (F1–F5) and alloy categories (4A, 6B, 6C, etc.).
- Furnaces F4 and F5, and alloys 6B and 6C, exhibited relatively higher HRI values, indicating lower heat-through margins compared to others, though still within the sufficient range.
Parameter Sensitivity:
- Sensitivity analysis perturbing heat-transfer coefficients, thermal conductivity, specific heat, and initial temperature showed that the HRI ranking correlation for the reliable subset remained above 0.9992. This confirms the method's robustness to parameter uncertainties.

4. Key Contributions

Methodological Innovation: Proposed PC-VHSA, a physics-constrained virtual thermal-state analysis method specifically designed for scenarios where workpiece temperature measurement is unavailable. It converts multi-zone furnace trajectories and static records into a radial heat-transfer constrained virtual state.
Quantitative Framework: Developed a framework to quantify thermal penetration using virtual center-temperature deficits, surface-center gradients, and the composite HRI, extending furnace compliance metrics into material-oriented thermal interpretations.
Reliability Control: Established a mechanism to separate reliable thermal assessments from data anomalies. By filtering out batches with boundary or energy inconsistencies, the method avoids misclassifying data quality problems as actual thermal insufficiency.

5. Significance and Claims

The paper claims that PC-VHSA provides a practical, interpretable route for confirming homogenization sufficiency and quantifying thermal margins in industrial settings without direct workpiece temperature sensors.

Shift in Perspective: The method moves beyond simple furnace temperature compliance to a material-centric view, acknowledging that furnace setpoints do not guarantee center temperature attainment due to radial conduction lag.
Data Integrity: By explicitly identifying and excluding data-boundary anomalies (21.41% of the dataset), the method prevents the misinterpretation of industrial data quality issues as process failures.
Engineering Utility: The HRI allows operators to compare relative heat-through margins across different furnaces and alloy types, identifying specific operating conditions (e.g., F4/F5 furnaces or 6-series alloys) that may require process optimization, even if they are currently "sufficient."

The authors maintain a modest stance, noting that the virtual thermal state is a physically constrained approximation rather than a direct replacement for measured temperatures. They acknowledge limitations regarding 3D airflow, billet stacking, and end effects, which are not explicitly modeled. Future work is suggested to incorporate limited discharge temperature measurements or mechanical property data for direct validation and to extend the model to account for spatial loading effects.