融合无监督学习的差示扫描量热特征峰分析方法  

Differential scanning calorimetry characteristic peak analysis method based on unsupervised learning

作　　者：王晓东 许金鑫 丁炯[1] 王晓娜[1] 叶树亮[1] WANG Xiaodong;XU Jinxin;DING Jiong;WANG Xiaona;YE Shuliang(Institute of Industry and Trade Measurement Technique,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]中国计量大学工业与商贸计量技术研究所,浙江杭州310018

出　　处：《中国测试》2025年第3期53-58,共6页China Measurement & Test

基　　金：国家自然科学基金(22173087);浙江省基础公益研究计划(LGG22B030002)。

摘　　要：为解决传统的差示扫描量热法(differential scanning calorimetry,DSC)信号分析需要手动选点来构造基线,存在人为因素引入误差且操作繁琐等问题,将机器学习方法应用于DSC信号分析中,提出一种融合无监督学习的DSC自动基线构造及特征峰信号分析方法。首先使用改进的聚类算法将特征峰两侧的基线与特征峰信号初步分离;其次对特征峰两侧的基线信号结合迭代多项式拟合进行基线重构;最后将原始信号减去重构的基线信号得到净特征峰信号,进行热力学分析。对多组实验数据分析表明,基于机器学习的DSC信号分析方法可自动实现良好的基线构造与峰信号分析,提高DSC信号分析的速度和精度,有效减少人为因素引起的分析误差。Machine learning method was applied to differential scanning calorimetry(DSC)signal analysis to solve the problem that the traditional DSC signal analysis method needed to manually select points to construct the baseline,there was a human factor to introduce repeatability error and cumbersome operation,an automatic method of DSC baseline construction and characteristic peak signal analysis based on unsupervised learning was proposed.Firstly,the baseline on both sides of the characteristic peak was initially separated from the characteristic peak signal by the improved clustering algorithm.Secondly,the baseline signals on both sides of the characteristic peak were combined with iterative polynomial fitting to reconstruct the baseline.Finally,the original signal was subtracted from the reconstructed baseline signal to obtain the net characteristic peak signal for thermodynamic analysis.The analysis of several sets of experimental data shows that the DSC signal analysis method based on machine learning can automatically achieve good baseline construction and peak signal analysis,improve the speed and accuracy of DSC signal analysis,and effectively reduce the analysis errors caused by human factors.

关 键 词：差示扫描量热法 基线重构 峰识别 聚类 迭代拟合 信号分析 

分 类 号：TB9[一般工业技术—计量学] TH81[机械工程—测试计量技术及仪器] TP181[自动化与计算机技术—控制理论与控制工程]