Experimental data-based analysis and intelligent prediction of smoke production characteristics for typical PMMA fire loads in substations
Year: 2026
Type: article
Abstract: Accurate characterization and early prediction of smoke production are essential for fire-risk assessment in substations, where polymethyl methacrylate (PMMA) is a typical fire load. Cone-calorimeter tests were conducted on 5 and 25 mm PMMA specimens, with three replicate trials each. Seven time-resolved combustion variables, including smoke production rate (SPR), heat release rate, and gas concentrations, were recorded throughout the full combustion cycle. Using a 30 s historical window, K-nearest-neighbor (KNN), long short-term memory (LSTM), and Transformer models were trained to forecast SPR at horizons of 5, 15, and 30 s, with one complete trial held out for independent cross-trial testing. The 5 mm specimens showed a short, concentrated smoke-production process, whereas smoke production from the 25 mm specimens persisted for more than 1800 s; despite a 26.8% lower peak SPR, their total smoke production (68.73 ± 15.88 m²) was approximately 10.3 times that of the 5 mm specimens. Among the three PMMA smoke models developed from the experimental data, Transformer achieved R² greater than 0.8 under both scenarios and was the better performing of the three models. Error analysis revealed an increasing tendency to underpredict SPR at longer horizons, and SHAP analysis showed that model reliance shifted from historical SPR toward material consumption and combustion-state features.
Source: Scientific Reports
Authors: Menghan Jin(Conceptualization, Methodology, Writing – original draft), Wenbing Zhu(Formal analysis, Methodology, Writing – review & editing), Yidong Zhang(Investigation, Validation), Qing Liu(Conceptualization, Project administration, Writing – review & editing), Tiancheng Meng(Methodology, Writing – original draft), Chunpeng Jiang(Visualization, Writing – review & editing), Zhihong Tang(Data curation, Visualization), Yujia Li(Investigation), Hui Wang(Funding acquisition), Bingyang Li(Investigation, Validation)
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7234355630
