Rapid Identification of Textile Fiber Composition Using Microtube Plasma Ionization Mass Spectrometry Combined with Random Forest Algorithm
10.19756/j.issn.0253-3820.251124
- VernacularTitle:微管等离子体电离质谱法结合随机森林模型快速鉴别纺织品纤维成分
- Author:
Yu-Han SHANG
1
;
Yue-Guang LYU
;
Xian-Shuang MENG
;
Qing LYU
;
Xiang-Yu GUO
;
Qing ZHANG
Author Information
1. 中国质量检验检测科学研究院,国家市场监督管理总局消费品质量安全监测与风险评估重点实验室,北京 100176
- Keywords:
Textiles;
Microtube plasma ionization mass spectrometry;
Random forest;
Fiber composition identification
- From:
Chinese Journal of Analytical Chemistry
2025;53(8):1331-1341,中插93-中插95
- CountryChina
- Language:Chinese
-
Abstract:
A rapid and accurate method for textile fiber identification was developed for quality control and consumer protection.This method utilized electric soldering iron burning-mesh collision enhanced microtube plasma ionization mass spectrometry(ESIB-MC-μTP-MS)to acquire textile fiber MS data and used a random forest(RF)prediction model to identify fiber composition based on these MS data.The MC-μTP device involved in the method was a homemade low-temperature plasma ionization device constructed using cost-effective and readily available components.The system was applicable for direct analysis of small amount of textile samples without any complex sample pretreatment processes.Characteristic thermal decomposition products of different fibers were generated via soldering iron burning(350℃)in ambient atmosphere,and were subsequently analyzed by a mass spectrometer,with each analysis completed within 5 s.Raw MS data underwent noise reduction,normalization,and global binning steps to form a dataset,and its intrinsic class separability was evaluated using principal component analysis(PCA)combined with k-means clustering.Then,the RF model was trained based on the dimensionality-reduced textile fiber dataset.After grid search optimization,this model demonstrated robust performance with a 0.9762 out-of-bag score,a 0.9683 cross-validation accuracy(5-fold),and a 0.9636 test accuracy,supported by precision,recall,and F1-scores exceeding 0.889 for all fiber classes.The method was applied to analysis of 30 luxury apparel samples from eight brands,among which 20 samples achieved 100%prediction confidence,aligning with labeled compositions.The identification result of two low-confidence samples was further confirmed using attenuated total reflection Fourier transform infrared spectroscopy(ATR-FT-IR).The method has been proven to be simple,portable and with minimal sample requirements for on-site customs inspections,providing a viable tool in the fight against counterfeit products,therefore supporting regulatory enforcement and consumer trust in the textile goods market.
