Scholarlink

<p>Allomorphic transformation of cellulose can alter chain packing, hydrogen-bonding arrangements, accessibility, and chemical reactivity. However, the crystallinity and phase purity of cellulose III are not intrinsic properties of the allomorph itself, but depend strongly on preparation conditions. In this study, we developed an infrared (IR) spectroscopy–based machine learning approach for estimating cellulose III<sub>I</sub>-equivalent content on a Rietveld-informed calibration scale. An ethylenediamine-treated cellulose III<sub>I</sub>-rich reference sample was first reassessed by Rietveld refinement using cellulose III<sub>I</sub> and cellulose Iβ structural models. The refinement indicated that the reference sample was not fully converted to cellulose III<sub>I</sub>, giving fitted phase fractions of 49.7% cellulose III<sub>I</sub> and 50.3% cellulose Iβ. The calibration targets were therefore calculated from the Rietveld-derived cellulose III<sub>I</sub> fraction rather than from the conventional XRD peak-height index. Feature importance analysis using random forest (RF) identified the 1400–900 cm<sup>−1</sup> fingerprint region as the most informative spectral domain for cellulose III<sub>I</sub> estimation. Restricting the model to this region reduced the number of spectral variables from 1764 to 260 while improving RF performance from R<sup>2</sup> = 0.955 and RMSE = 0.035 to R<sup>2</sup> = 0.975 and RMSE = 0.025. Application to ethylenediamine-treated commercial cellulose samples further demonstrated that the optimized RF model can provide physically reasonable estimates of cellulose III<sub>I</sub>-equivalent content without requiring explicit XRD peak separation. This study shows that Rietveld-validated reference calibration combined with IR-based machine learning offers a practical and interpretable framework for rapid evaluation of cellulose allomorphic transformation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning detection and quantification of cellulose III using infrared spectroscopy

  • 期刊论文

  • 发表日期：

    2026年7月1日

查看全文
  • Yong Ju Lee,
  • Geon-Woo Kim,
  • Soon Wan Kweon,
  • Hyoung Jin Kim

摘要

Allomorphic transformation of cellulose can alter chain packing, hydrogen-bonding arrangements, accessibility, and chemical reactivity. However, the crystallinity and phase purity of cellulose III are not intrinsic properties of the allomorph itself, but depend strongly on preparation conditions. In this study, we developed an infrared (IR) spectroscopy–based machine learning approach for estimating cellulose IIII-equivalent content on a Rietveld-informed calibration scale. An ethylenediamine-treated cellulose IIII-rich reference sample was first reassessed by Rietveld refinement using cellulose IIII and cellulose Iβ structural models. The refinement indicated that the reference sample was not fully converted to cellulose IIII, giving fitted phase fractions of 49.7% cellulose IIII and 50.3% cellulose Iβ. The calibration targets were therefore calculated from the Rietveld-derived cellulose IIII fraction rather than from the conventional XRD peak-height index. Feature importance analysis using random forest (RF) identified the 1400–900 cm−1 fingerprint region as the most informative spectral domain for cellulose IIII estimation. Restricting the model to this region reduced the number of spectral variables from 1764 to 260 while improving RF performance from R2 = 0.955 and RMSE = 0.035 to R2 = 0.975 and RMSE = 0.025. Application to ethylenediamine-treated commercial cellulose samples further demonstrated that the optimized RF model can provide physically reasonable estimates of cellulose IIII-equivalent content without requiring explicit XRD peak separation. This study shows that Rietveld-validated reference calibration combined with IR-based machine learning offers a practical and interpretable framework for rapid evaluation of cellulose allomorphic transformation.