人工智能
机器学习
计算机科学
支持向量机
人工神经网络
可解释性
特征(语言学)
数据挖掘
特征选择
工程类
训练集
作者
Kang Yang,Yulong Ma,Yi Zhang,Anlan Zhong,Lijuan Zang,Juheng Wang,Xiaoxiao Huang
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-04-24
卷期号:: 1-7
被引量:1
标识
DOI:10.1109/iciscn67954.2026.11566577
摘要
In modern times, prediction of mechanical properties of polyethene become increasingly essential because of growing demand for high-performance polymer products in packaging, automotive and structural applications. Though, existing Firefly-optimized Extremely Randomized Trees (FAET) approach obtained high prediction accuracy but struggled with challenges like restricted interpretability of ensemble prediction, absence of physical aware learning strategies and high computational cost because of metaheuristic optimization. To address these limitations, Physics-Aware Interpretable Bayesian-Optimized Multi-Output Extremely Randomized Trees (PAI-BO-MOET) framework is proposed to predict the mechanical properties of polyethene. Primarily, data is collected from previously published study on moulded LLDPE which comprises of 25 samples with oven residence time as input parameter then tensile, impact and flexural strength as measured outputs. Next, physics-aware constraint mechanism is incorporated for ensuring predictions follow realistic polymer melting and behaviour of degradation. Further, multi-output Extremely Randomized Trees approach is adopted for simultaneously predict tensile, impact and flexural strength which preserves inter-property correlation. Furthermore, model transparency is improved through SHAP analysis and Partial Dependence Plots which makes the framework more interpretable. Additionally, Bayesian Optimization (BO) is employed which significantly lowers optimization time. Finally, proposed PAI-BO-MOET framework delivers enhanced predictive stability, reduced computational time, improved interpretability and physically consistent prediction of mechanical properties of polyethene.
科研通智能强力驱动
Strongly Powered by AbleSci AI