采用热重分析仪研究了空气气氛中废弃医用外科口罩的热氧化降解特性,并运用3种无模型方法以及1种模型拟合方法进行了动力学和热力学分析。结果表明:废弃医用外科口罩在空气气氛中的热氧化降解是一种吸热的非自发反应,且整个热氧化降解...采用热重分析仪研究了空气气氛中废弃医用外科口罩的热氧化降解特性,并运用3种无模型方法以及1种模型拟合方法进行了动力学和热力学分析。结果表明:废弃医用外科口罩在空气气氛中的热氧化降解是一种吸热的非自发反应,且整个热氧化降解过程可被认为是单步反应。热氧化降解过程的活化能和指前因子的平均值分别为63.83 k J·mol^(-1)和1.27×10^(5) min^(-1),最概然反应模型为g(a)-[-ln(1-a)]^(2/3),所获得的动力学参数可用于较好地预测废弃医用外科口罩在空气中的热氧化降解行为。研究结果可以为预防废弃医用外科口罩可能引发的火灾事故提供参考。展开更多
本文将机器学习与化工安全设计课内闪点测定实验有机结合,基于人工智能技术构建了可燃液体闪点测定实验。首先,通过文献和数据库收集到1800组闪点实验值和简化分子线性输入规范(SMILES)数据,对盐类、闪点缺失和SMILES缺失的数据进行剔...本文将机器学习与化工安全设计课内闪点测定实验有机结合,基于人工智能技术构建了可燃液体闪点测定实验。首先,通过文献和数据库收集到1800组闪点实验值和简化分子线性输入规范(SMILES)数据,对盐类、闪点缺失和SMILES缺失的数据进行剔除得到1300组可用数据。然后查阅文献,选择分子指纹作为分子描述符用于提取分子结构特征参数,基于Python内的RDKit工具包完成由溶剂SMILES到分子指纹的计算。通过调用sklearn内的随机森林算法建立溶剂闪点预测模型,引入网格搜索函数对决策树数量、深度和样本量进行优化并进行5次交叉验证防止模型过拟合。采用上述模型对测试集数据进行预测得到溶剂的闪点数据并与实验数据进行比对得到模型的准确率。该仿真实验融合了机器学习前沿学科知识和危险化学品管理、化工热力学等专业课程内容,强化了化工安全工程专业学生的基本功,激发了学生科研兴趣,实现了“智能 + 安全”新工科课程的建设。This article combines machine learning with chemical engineering safety experiments and builds a flammable liquid flash point determination experiment based on artificial intelligence technology. Firstly, 1800 sets of flash point experimental values and Simplified Molecular Input Line Entry System (SMILES) data were collected from the reference and databases. Data with salts, missing flash points, and missing SMILES were removed, resulting in 1300 sets of usable data. Then, the molecular fingerprints as descriptor were selected to extract structural characteristic parameters of flammable liquid molecules, as suggested by literature. The calculation of solvent SMILES to molecular fingerprints was performed using the RDKit toolkit in Python. A solvent flash point prediction model was established by invoking the random forest algorithm in the sklearn package. The grid search function was used to optimize the number of decision trees, depth, and sample size, and 5-fold cross-validation was performed to prevent model overfitting. The above model was used to predict flash point data of the test set and compared with experimental data to obtain the accuracy of the model. This simulation experiment integrates cutting-edge knowledge of machine learning with professional courses in chemical management and thermodynamics, strengthening the basic training of students in chemical engineering safety engineering, stimulating students’ research interests, and realizing the construction of “intelligent + safety” new engineering courses.展开更多
文摘采用热重分析仪研究了空气气氛中废弃医用外科口罩的热氧化降解特性,并运用3种无模型方法以及1种模型拟合方法进行了动力学和热力学分析。结果表明:废弃医用外科口罩在空气气氛中的热氧化降解是一种吸热的非自发反应,且整个热氧化降解过程可被认为是单步反应。热氧化降解过程的活化能和指前因子的平均值分别为63.83 k J·mol^(-1)和1.27×10^(5) min^(-1),最概然反应模型为g(a)-[-ln(1-a)]^(2/3),所获得的动力学参数可用于较好地预测废弃医用外科口罩在空气中的热氧化降解行为。研究结果可以为预防废弃医用外科口罩可能引发的火灾事故提供参考。
文摘本文将机器学习与化工安全设计课内闪点测定实验有机结合,基于人工智能技术构建了可燃液体闪点测定实验。首先,通过文献和数据库收集到1800组闪点实验值和简化分子线性输入规范(SMILES)数据,对盐类、闪点缺失和SMILES缺失的数据进行剔除得到1300组可用数据。然后查阅文献,选择分子指纹作为分子描述符用于提取分子结构特征参数,基于Python内的RDKit工具包完成由溶剂SMILES到分子指纹的计算。通过调用sklearn内的随机森林算法建立溶剂闪点预测模型,引入网格搜索函数对决策树数量、深度和样本量进行优化并进行5次交叉验证防止模型过拟合。采用上述模型对测试集数据进行预测得到溶剂的闪点数据并与实验数据进行比对得到模型的准确率。该仿真实验融合了机器学习前沿学科知识和危险化学品管理、化工热力学等专业课程内容,强化了化工安全工程专业学生的基本功,激发了学生科研兴趣,实现了“智能 + 安全”新工科课程的建设。This article combines machine learning with chemical engineering safety experiments and builds a flammable liquid flash point determination experiment based on artificial intelligence technology. Firstly, 1800 sets of flash point experimental values and Simplified Molecular Input Line Entry System (SMILES) data were collected from the reference and databases. Data with salts, missing flash points, and missing SMILES were removed, resulting in 1300 sets of usable data. Then, the molecular fingerprints as descriptor were selected to extract structural characteristic parameters of flammable liquid molecules, as suggested by literature. The calculation of solvent SMILES to molecular fingerprints was performed using the RDKit toolkit in Python. A solvent flash point prediction model was established by invoking the random forest algorithm in the sklearn package. The grid search function was used to optimize the number of decision trees, depth, and sample size, and 5-fold cross-validation was performed to prevent model overfitting. The above model was used to predict flash point data of the test set and compared with experimental data to obtain the accuracy of the model. This simulation experiment integrates cutting-edge knowledge of machine learning with professional courses in chemical management and thermodynamics, strengthening the basic training of students in chemical engineering safety engineering, stimulating students’ research interests, and realizing the construction of “intelligent + safety” new engineering courses.
