Quantification of cellulose II using infrared spectroscopy: Machine learning approaches

Area of Science:

• Materials Science
• Spectroscopy
• Machine Learning

Background:

• Cellulose II allomorph is commercially significant in regenerated and mercerized fibers like lyocell and viscose.
• Cellulose III and IV allomorphs are primarily of scientific interest.
• Accurate quantification of cellulose allomorphs is crucial for material characterization.

Purpose of the Study:

• To develop and validate a machine learning approach for quantifying cellulose II content in cellulose-based materials.
• To leverage infrared spectroscopy for rapid and reliable cellulose II quantification.
• To compare the performance of different regression models for this task.

Main Methods:

• Utilized infrared spectroscopy combined with second-derivative preprocessing for enhanced feature detection.
• Applied feature importance analysis to identify the informative spectral region (1400-900 cm⁻¹).
• Developed and compared four regression models: random forest, decision tree, partial least squares regression, and multilayer perceptron.

Main Results:

• Second-derivative preprocessing significantly improved the detection of cellulose II features.
• The 1400-900 cm⁻¹ spectral region proved highly informative for distinguishing cellulose allomorphs.
• The random forest model achieved superior performance with R² = 0.994, RMSE = 0.022, and MRE = 0.052.
• The best model successfully quantified cellulose II in NaOH-treated wood pulp, CNC, and lyocell.

Conclusions:

• The combination of infrared spectroscopy and machine learning offers a powerful tool for cellulose II quantification.
• This method provides a rapid and reliable alternative to conventional X-ray diffraction.
• The developed model demonstrates high accuracy and robustness for analyzing diverse cellulose-based materials.