Effect of Hammer Mill Particle Size on the Performance of NIR Prediction Models for Whole-Plant Corn Constituents

This study used a portable NIRS analyzer to explore the feasibility of modeling NIRS dry milling using hammer mill-prepared samples and determine the optimal particle size. 180 whole-plant corn samples from across China underwent four treatments: milling to cyclone 1.0 mm (control), hammer 2.0 mm, 3.0 mm and 4.0 mm. Samples underwent NIRS scanning followed by wet chemistry for DM, starch, and ash. PLS models were built for each constituent and particle size. Results showed the hammer 2.0 mm sample had significantly lower mean spectral absorbance than hammer 3.0 mm and 4.0 mm samples ( P < 0.05). Its spectral repeatability was similar to hammer 3.0 mm but significantly lower than hammer 4.0 mm ( P < 0.05). Among the prediction models, those for DM and starch at hammer 2.0 mm and 3.0 mm particle sizes performed best (R 2 c > 0.9, R 2 cv > 0.9, R 2 p > 0.9, RPD > 3), demonstrating practical a p plication potential. The Ash prediction model performed poorly overall (R 2 c < 0.8, R 2 cv < 0.6, R 2 p < 0.8, RPD < 3) and requires optimization. Considering spectral quality and model performance, hammer 2.0 mm was determined optimal for NIRS sample preparation. • Portable infrared device tested for corn milling analysis. • Found best corn particle size: 2.0 mm using hammer mill. • Accurately predicted corn starch and dry matter content. • Ash content prediction needs improvement for practical use. • Smaller corn particles gave clearer infrared signals.