Quantitative Optical Identification of Fiber Composition in Mixed Textile Bales
Diffuse reflectance spectroscopy provides rapid non-destructive fiber identification in mixed bales when multivariate models correct for moisture and density.
Root Mean Square Error
Statistical Metric
Quantitative analysis of textile blend ratios using spectroscopy requires a reliable measure of model accuracy. The root mean square error represents the primary metric for evaluating how well a calibration model predicts fiber percentages. This value represents the average discrepancy between the laboratory reference chemistry and the spectral prediction.
A lower value indicates a more reliable calibration model for mill use.
Calculation Mechanism
Computing this metric involves squaring the individual differences between predicted and measured values across a validation set. By taking the average of these squared differences and calculating the square root, the root mean square error is expressed in the same units as the fiber blend percentage. This squaring penalizes large prediction errors more heavily than small ones.
This weighting is particularly useful for detecting samples with unusual dye or finish coatings. It provides a conservative assessment of the model’s performance on production samples.
Calibration Standard
Calibration software utilizes this metric to compare different mathematical pretreatments of NIR spectra. When optimizing a model for cotton-polyester blend determination, technicians adjust scatter corrections to minimize the root mean square error on a test set. This process ensures that the model is robust against differences in fabric thickness and dye type.
It helps avoid overfitting the spectral data to a single fabric batch.
Accuracy Assessment
Acceptance limits for this metric are established based on the commercial tolerance of the fiber blend. For cotton-polyester yarns, a root mean square error below one percent is required to pass audits. This high accuracy ensures that shipments comply with legal labeling regulations.