C. K. Spillman

REVIEW: Near infrared reflectance spectroscopy for online particle size analysis of powders and ground materials

The cross-sensitivity of near infrared (NIR) reflectance to the particle size of powders or ground materials has long been documented but not fully exploited for particle size estimation. Diffuse reflectance of a powder sample is dependent on light scattering within its layers, and a powders absorption and scattering coefficient are related to its particle size. This is the basis of NIR reflectance–particle size calibrations. The availability of fibre optic probes and the speed of NIR spectrometers make them suitable for remote and online sensing of particle size, in addition to providing chemical information of a powder sample. The basics of NIR reflectance spectroscopy relevant to particle size determination and its relation to sample preparations, methods of presentation, reference methods, calibration development and validation are reviewed in this paper.

Mechanical Properties of Wheat

Fundamental mechanical properties of the wheat kernel were characterized for five different wheat classes. An Instron testing machine was used to perform a compression test on wheat slices. Wheat slices were prepared by cutting off both ends of the wheat kernel. A cross-sectional area of the wheat slices was measured using a commercial image-processing system. An offset yield of 0.2% was introduced for determining the yield point for wheat and used to determine yield stress, yield strain, modulus of deformability, and energy to yield point. To determine the effect of sample location, each kernel of hard red spring and soft red winter wheats was cut into two slices (germ end and brush end). No significant differences, a = 0.05, were found in values of the physical properties for germ-end and brush-end samples. In general, the four properties increased as loading rate increased up to a 100 mm/min and decreased thereafter. The mean values of yield stress for all five wheat classes decreased as moisture content increased. Soft wheat samples had generally lower modulus of deformability values.

Neural network modeling of physical properties of ground wheat

ABSTRACT Physical properties of ground materials from roller mills are affected by the characteristics of wheat and the operational parameters of the roller mill. Backpropagation neural networks were designed, trained, and tested for the prediction of three physical properties of ground wheat: geometric mean diameter (GMD), specific surface area increase (SSAI), and break release (BR). Eight independent variables were used as input data. Compared to conventional statistical models, the accuracy of prediction was improved substantially, as reflected by the significant reduction in root mean squared error (RMS), relative error (RE), and the increase in coefficient of determination R2 (>0.98). The neural network models are, therefore, capable of predicting the physical properties of the ground wheat.

COMPARISON OF ENERGY EFFICIENCY BETWEEN A ROLLER MILL AND A HAMMER MILL

AIt is generally believed that roller mills utilize energy more efficiently than hammer mills (Silver, 1931; Puckett and Daum, 1968; Appel, 1987). To verify this, a completely randomized factorial experimental design (CRD) with two replications was constructed: a total of 72 grinding tests was conducted to compare the energy efficiency between a roller mill and a hammer mill. Energy data were collected using a computerized data acquisition system and a watt-hour meter for the roller mill and the hammer mill, respectively. A mixture of hard red winter wheat (HRW) varieties was used to conduct the experiment. To make the comparison meaningful, the parameters for the two machines were set so that the ground material characteristics from both were almost identical. For roller mill grinding, roll gap was the most significant factor affecting energy requirement, followed by the roll speed differential. For hammer mill grinding, screen opening size had the most significant effect. For both mills, feed rate was not significant; and energy efficiencies were about the same.

ENERGY REQUIREMENTS FOR SIZE REDUCTION OF WHEAT USING A ROLLER MILL

An experimental two-roll mill was developed and instrumented for computerized data acquisition. Milling tests were performed on three classes of wheat. Included in the study were six independent variables each with three levels, namely, class of wheat, moisture content, feed rate, fast roll speed, roll speed differential, and roll gap. Two covariates, single kernel hardness and single kernel weight, were also included in the statistical analysis. Prediction models were constructed for five dependent variables (fast roll power, slow roll power, net power, energy per unit mass and specific energy). The prediction models fitted the experimental data well (r2 = 0.88 ~ 0.95). The power and energy requirements for size reduction of wheat were highly correlated with the single kernel characteristics of wheat. Feed rate affected fast roll power, slow roll power and net power significantly. Roll gap had a significant effect on roller mill grinding. Additional milling tests were conducted by randomly selecting independent variables and covariates to verify the robustness and validity of the prediction models.

MODELING THE SIZE PROPERTIES OF FIRST–BREAK GROUND WHEAT

The potential of single kernel properties of wheat (mean and standard deviation of hardness, size, and mass) and roller mill’s roll gap in modeling the size properties of first–break ground wheat was investigated. Full (seven variables) and reduced models (six variables) for break release (BR in log scale, or L_BR) and geometric mean diameter (GMD) were developed based on milling data from six wheat classes ground at five roll gaps. The models explained most of the variability in the experimental data and relationships obtained were consistent with previous research and millers’ experience. Roll gap and single kernel hardness had the most significant effects on L_BR and GMD. Reduced models that used a variable combining single kernel size and roll gap reduced some collinearities among single kernel properties and isolated the effect of single kernel mass from that of single kernel size. The effect of single kernel mass became significant in the reduced models. The models performed well in validation tests. Predictions using the L_BR model were better than an equation based only on roll gap. L_BR was linearly related to GMD. Because the L_BR model was wheat class–specific, it has potential for online roll gap control.

NEURAL NETWORK MODELING OF ENERGY REQUIREMENTS FOR SIZE REDUCTION OF WHEAT

Power and energy requirements for size reduction of wheat are affected by the physical and mechanical characteristics of wheat, and the operational parameters of the roller mill. Wheat milling tests were conducted using 204 samples of six classes and various varieties collected from around the United States. Backpropagation neural network models were designed, trained and validated for the prediction of power and energy requirements of wheat milling using a roller mill: fast roll power (Pf), slow roll power (Ps), net power (Pn), energy per unit mass (Em), and specific energy (Ea). Nine variables including physical properties of wheat samples and operational parameters of the roller mill were used as inputs of the networks. Each of the networks had only one layer of hidden neurons. Sensibility studies also were carried out to investigate effects of input variables on the output variables. The developed network models performed well during validation. Compared to the experimental data, the values of the root mean square error (RMS) and relative error (RE) of the predicted values were small, ranging from 0.34 to 0.73 for the RE values. The r2 values were higher than 0.98 for all five networks. The prediction accuracies of the neural network models were significantly improved compared to statistical models.

Evaluation of a near-infrared reflectance spectrometer as a granulation sensor for first-break ground wheat : Studies with six wheat classes

ABSTRACT In flour milling, a granulation sensor for ground wheat is needed for automatic control of a roller mills roll gap. A near-infrared (NIR) reflectance spectrometer was evaluated as a potential granulation sensor of first-break ground wheat using offline methods. Sixty wheat samples, ground independently, representing six classes and five roller mill gaps, were each used for calibration and validation sets. Partial least squares regression was used to develop the models with cumulative mass of size fraction as the reference value. Combinations of four data pretreatments (log (1/R), baseline correction, unit area normalization, and derivatives) and three wavelength regions (700–1,500, 800–1,600, and 600–1,700 nm) were evaluated. Unit area normalization combined with baseline correction or second derivative yielded models that predicted well each size fraction of first-break ground wheat. Standard errors of performance of 4.07, 1.75, 1.03, and 1.40 and r2 of 0.93, 0.90, 0.88, and 0.38 for the >1,041...

Evaluation of a near-infrared reflectance spectrometer as a granulation sensor for first-break ground wheat: Studies with hard red winter wheats

ABSTRACT A single wheat class or blended wheats from two wheat classes are usually milled in a flour mill. A near-infrared (NIR) reflectance spectrometer, previously evaluated as granulation sensor for first-break ground wheat from six wheat classes, was evaluated for a single wheat class, hard red winter (HRW) wheat, using offline methods. The HRW wheats represented seven cultivars ground by an experimental roller mill at five roll gap settings (0.38, 0.51, 0.63, 0.75, and 0.88 mm) which yielded 35 ground wheat samples each for the calibration and validation sets. Granulation models based on partial least squares regression were developed with cumulative mass of size fractions as a reference value. Combinations of four data pretreatments (log 1/R, baseline correction, unit area normalization, and derivatives) and subregions of the 400–1,700 nm wavelength range were evaluated. Models that used pathlength correction (unit area normalization) predicted well each of the four size fractions of first-break gro...

Modeling the energy requirements of first-break grinding

The influence of roller mill gap and single kernel properties of wheat on the energy requirements of first–break grinding was studied. Multiple linear regression models for energy per unit mass (EM), new specific surface area (ANSS, in log scale), and specific energy (EA = EM /ANSS) were developed based on milling data from six wheat classes ground at five roll gaps with an experimental roller mill. The models, which were functions of roll gap, single kernel properties, and wheat class as a classification variable, explained most of the variability in the experimental data. Roll gap and single kernel hardness had the greatest influence on EM, ANSS, and EA. Milling ratio, a variable that combined single kernel size and roll gap, reduced some collinearities among single kernel properties and isolated the effect of single kernel mass from that of single kernel size. The effect of single kernel mass became significant in the models based on milling ratio. Good agreement between predicted and measured values was observed from 100 validation samples. The EA model is a potential alternative to break release as a basis for online roll gap control.