(96a) Predicting Loss-in-Weight Feeder Performance Based on a Reduced Set of Material Property Measurements

Tianyi Li, Yi Tao, Fernando J. Muzzio, Benjamin J. Glasser*

Purpose: Loss-in-weight feeders plays a vital role in continuous powder processing. Material properties have a significant impact on loss-in-weight feeder’s performance. Previous work has built up a correlation between loss-in-weight feeder’s gravimetric feeding and refill performance utilizing a statistical model, partial least squares regression (PLSR), with 30 material flow indices. However, based on research on correlation between material properties and the weight of each property in the prediction model, a reduced set of measurement can be developed with less material characterization/properties, which can save time, powder and effort for the material characterization stage. Moreover, deeper understanding of material properties and material flow behavior can be established by the analysis in this work. Method: The proposed work includes techniques to characterize material flow properties, method to quantify loss-in-weight feeders feeding performance under gravimetric mode and during hopper refill, principal component analysis (PCA) to build up a material library with material flow properties, cluster analysis to classify materials investigated into groups to compare the difference between different selection of reduced set of measurements, partial least squares regression to build up a predicting model to validate the prediction made with reduced set of measurement by correlating feeding performance with material flow properties. Results: Experiment results shows feeder’s feeding deviation is highly dependent on material properties and can be predicted with reduced set of measurement only 4 measurements of compressibility, permeability, shear cell at 6 kpa and particle size distribution. Conclusion: The work shows, based on the correlation between material properties and the weight of each property in the prediction of feeder performance, a good predicting model to correlate feeder performance and material flow properties, can be set up with a reduced set of measurements with 7 material properties instead of original 30. With this work, half of the time and materials can be saved from material characterization tests. Moreover, a deeper understanding between material properties and their correlation with feeding performance and be established.