Development and Validation of a Powder Mixture Set for Performance Testing of Miniaturized Spectrometers using Machine Learning Models

This paper presents the development and validation of a powder mixture set designed for performance testing of miniaturized spectrometers using Partial Least Squares (PLS) regression models. The validation powder mixture set is specifically tailored to exploit PLS regression models in order to assess and evaluate the performance of miniaturized spectrometers. The powder mixtures within the set contain known contents of various materials including Polytetrafluoroethylene (PTFE), polypropylene, polystyrene, talc, starch, protein, ascorbic acid, and MgSO
4
salt, aiming to emulate real-world applications for Near-Infrared (NIR) spectrometer measurements. To prepare the samples, powder materials were mixed with specific concentrations, generating three sets, each containing four different contents. The concentration values for each material were generated using the Latin hypercube sampling (LHS) method, ensuring low correlation among the materials within the powder samples. Robust PLS models were successfully built using all three sets, demonstrating their effectiveness in testing the performance of miniaturized spectrometers compared to other spectrometers. Furthermore, the backward compatibility of variations introduced to a spectrometer design was examined by measuring the prediction bias of the new spectrometer using the PLS model developed with reference spectrometer measurements. The results indicate that the developed powder mixture set not only enables the evaluation of the performance of miniaturized spectrometers, but also facilitates the assessment of their compatibility with existing models. Overall, this research contributes to the advancement of miniaturized spectrometry and offers a reliable method for validating and testing their performance using PLS regression models.