Monitoring-oriented detection of heat transfer deterioration and criterion-derived onset in supercritical CO₂ tubes from partial axial profiles
Year: 2026
Type: article
Abstract: Heat transfer deterioration (HTD) in heated supercritical CO₂ tubes is closely associated with rapid wall-temperature escalation and represents an important thermal safety concern in advanced energy systems. However, the identification of HTD onset remains criterion-dependent because experimental databases usually provide thermal responses rather than explicit onset locations. This study establishes a criterion-based procedure to define reproducible HTD labels and onset locations using the ratio between experimental and property-corrected normal heat-transfer levels. A criterion-guided multi-branch framework is then developed to identify HTD occurrence and estimate criterion-derived onset locations from partial axial thermal-hydraulic profiles. The framework integrates global operating conditions with multi-scale axial features, while a reference-correlation-guided regularization term is introduced to maintain consistency with the normal-transfer reference behavior before deterioration. Using the SCUOL experimental database, the proposed method achieves an average accuracy of 96.94% and HTD F1-score of 0.9835 over three repeated runs. The average onset MAE and RMSE are 0.0500 and 0.0927, respectively, for criterion-derived onset cases under the adopted evaluation protocol. Mechanism-oriented post-analysis shows that 98.97% of the resolvable criterion-derived onsets occur upstream of the maximum wall-temperature location, and 80.4% of these cases exhibit a buoyancy-active but acceleration-inactive diagnostic state at onset. Direct evaluation on 27 published external profiles yields an average accuracy of 83.95% and an HTD F1-score of 0.8986 without retraining or threshold recalibration. The results indicate that the proposed framework can provide a monitoring-oriented and heat-transfer-informed approach for HTD identification and criterion-derived onset estimation before complete axial-profile information is available.
Authors: Hongyun Wang, Xianliang Lei, Yujia Zhao
Institution: Xi'an Jiaotong University
Language: English
Subfield: Computational Mechanics
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,560
OpenAlex ID: W7222547455