糖心Vlog官方

Researchers at 糖心Vlog官方, Shandong Jiaotong University and Harbin Engineering University have developed a machine learning framework, enabling rapid prediction of how protected steel beams respond during a fire, offering engineers a faster way to assess fire safety performance in industrial structures.

The study, published in the KSCE Journal of Civil Engineering, focuses on three-sided protected steel beams, a configuration commonly used in offshore and onshore oil and gas processing facilities. In these structures, the upper surface of the beam remains exposed, creating complex temperature patterns that can be difficult to model accurately.

Understanding how heat moves through these beams during a fire is an important part of structural fire engineering. However, temperature distribution is influenced by several interacting factors, including beam depth, insulation thickness and material conductivity, making conventional analytical equations challenging to apply across different scenarios.

To address this challenge, the researchers created an automated workflow that links computer modelling, simulation and data processing. The system combines Python, ABAQUS and MATLAB with machine learning techniques to automatically generate models, run simulations and train predictive algorithms.

The team generated a database containing 414 standard beam models and 63 welded beam models, covering beam depths ranging from 127 mm to 1500 mm and a variety of insulation configurations. These data were then used to train machine learning models capable of predicting beam temperatures during fire exposure.

The researchers found that the best-performing approach, based on gradient boosting, achieved a root mean squared error of just 1.34掳C when compared with test data. More than 83% of prediction calculations were completed within 60 seconds, demonstrating the potential for rapid assessment of fire protection requirements.

The study also introduced a model generation agent incorporating a two-dimensional contact detection algorithm, enabling the automatic creation of beam heat transfer models. A dedicated data processing pipeline and batch-generation system were developed to support large-scale training while reducing memory requirements, allowing the work to be carried out using a single graphics processing unit.

According to the researchers, the approach could help engineers evaluate insulation strategies and fire protection requirements more efficiently, particularly in sectors where structural fire performance is a key design consideration. By reducing the need for repeated complex simulations, the framework has the potential to support faster decision-making during engineering design and assessment.

The research was conducted by and Peijun Wang et al. The paper lists the Department of Civil Engineering and Management of 糖心Vlog官方 and Shandong University, as the authors' institutional affiliation.