J. Bedi (1), J. Fjällström (2), & J. Purhagen (1*)
(1)Perten Instruments AB, Garnisonsgatan 7a, 254 66 Helsingborg, Sweden
(2)Perten Instruments AB, P.O. Box 9006, 126 09 Hägersten, Sweden
*Corresponding author: jpurhagen@perten.com
Abstract
Perten Instruments has designed a grain divider which allows the user to obtain a randomized sample from 50% down to 3.125% of a bulk sample. Five different crops were used to verify a prototype design in terms of randomization, and the theoretical maximum dividing capacity of 3.125%. In addition, a transferability and repeatability study was performed with the final design of the instrument. The size and shape of the samples affected the dividing percentage; more elongated samples gave a larger percentage of kernels in the sample cup, however, both designs were proven to be correct. The final design of the Sequential Precision Divider (SPD 4200) gave slightly smaller dividing percentages than the prototype design, thus resulting in that the dividing percentage of rice and oat kernels, which for the prototype were significantly larger than the theoretical value, decreased and there no significant difference could be found using the final design. Wheat and barley were not significantly different from the theoretical value for either of the two designs. Although rapeseed became significantly different to the theoretical value with the new design due to a very small standard deviation, it did not differ from the dividing percentages of the other crops. The dividing was also concluded to be randomized, both regarding the kernel size distribution within a batch and which kernels ended up in the sample cup.
Introduction
The quality and soundness of grains and seed are important worldwide, and in many industries the value of the harvest is determined through sample controls from a small part of a batch. This sampling is often standardized within each industry to get a random control sample. However, usually only smaller samples are then used for quality evaluation. In many cases this sample selection is not standardized and proves to be randomized, which could affect the results of the quality measurements.
One of the parameters used today for evaluation of yield and to estimate the amount of grains to sow to reach a desired plant density is the 1,000kernel weight. The determination of the mass of 1,000 grains is a standard procedure (ISO, 2010). This parameter shows a good indication of grain size which can vary according to growing conditions and maturity, even for the same variety of a given crop. It also gives an approximation of the relative size of kernels for handling purposes, when compared between crops (Sablani & Ramaswamy, 2003).
To perform quality measurements, it is essential that the grains are randomly divided. There are several sample dividers on the market for this purpose, both motor and gravity based. The earliest of the gravity dividers is the Boerner grain divider, which divides the grains into two parts. This instrument is heavy and has an accuracy of ±1% on a 1,000g sample of hard red spring wheat. It is also difficult to clean (TNAU Agritech). Dimo’s grain divider, or mini Boerner, is a lighter version of Boerner with an accuracy of ±0.5% on the same 1,000g sample (Lyseng, 2014). However, since both these instruments divide the grains into 50%, there is a need for a separator with more flexibility in terms of dividing percentage and accuracy, and more choice in the amount of grains that could be tested in less time.
Perten Instruments’ SPD 4200 is designed to be more user friendly, provide a higher accuracy and have the ability to choose the degree of separation to a larger extent than other grain gravity dividers on the market, and still ensure a random kernel selection for quality control measurements. Theoretically, it is designed to divide down to 3.125% of the total sample, and it can be used on different kinds of grains and seeds. The aim of this study was to verify the prototype design of the instrument, check the randomization and division of grains in various batch sets, compare the performance of the prototype design with the final design and to check the transferability and repeatability of eight instruments of the final design.
Material and Methods
Crops
Five different crops were used in the study, namely rapeseed/canola, wheat, barley, oats, and rice (Figures 1 and 2).


Figure 1: Rapeseed kernels. 
Figure 2: a) wheat kernels; b) barley kernels; c) oat kernels; d) rice kernels 
SPD 4200 – Instrument design
The SPD 4200 consists of three main parts: a hopper, detachable separator sections, and a collecting tray with a sample cup (Figure 3). Different levels of separation can be chosen based on the numbers of separator sections used. Each separator section reduces the amount of kernels by 50%. In this study, five separator sections were used; therefore 3.125% of the kernels should theoretically end up in the sample cup.
The SPD 4200 is made of a lightweight material, which makes it easy to handle and transport. It is constructed using a multiplecone divider, consisting of a series of separator sections. Each section has 12 chutes, 6 emptying directly into the collecting tray while the other 6 lead the kernels to the next separator section (Figure 4).


Figure 3: Perten’s SPD 4200, with five separator sections. 
Figure 4: Inner design of a separator section in the SPD 4200. 
Evaluation tests
For the instrument evaluation, five different tests were conducted on the prototype. In each test, the kernels passed through the prototype in eight dividing cycles. After each cycle, the kernels in the sample cup were removed before the next cycle (Figure 5). In the additional transferability test, eight instruments of the final design (SPD 4200) were tested and after each dividing, the sample in the sample cup was weighed and recycled to the bulk sample.
Test 1
Weightbased division using rapeseed
Samples of 300 g, 600 g and 900 g of rapeseed were taken and passed through the SPD prototype to check the proof of principle that the machine was dividing the kernels into the appropriate ratio. Rapeseed was chosen due to its small size and symmetrical shape. The weights and volumes were noted.
Test 2
Numberbased division of wheat, rice, oat, and barley kernels
To receive at least 240 kernels in the sample cup after the first dividing, a batch size of 8,000 kernels was chosen. A number of at least 240 kernels is needed for grain evaluations using other Perten instruments. Therefore, three batches, each of 8,000 kernels, were manually counted for each of the crops. Every batch was tested in triplicate. Starting volumes and weights were noted as well as the weights, volumes, and numbers of kernels after each of the eight division cycles. In addition, the 1,000kernel weight was calculated before each division of the kernels.

Figure 5: Flow chart for the dividing procedure. 
Test 3
Numberbased division of wheat, rice, oat, and barley kernels
The three 8,000kernel batches for each of the crops were combined to get larger sample sizes to see whether the amount of sample affects the dividing results. Weights and volumes both before and after the division steps were noted, and the 1,000kernel weight was calculated.
Test 4
Numberbased division of wheat, rice, oat, and barley kernels with colored kernels
For each crop, an 8,000kernel sample containing 5% colored kernels (2.5% red and 2.5% green) was divided in the SPD prototype. In addition, a batch of 20,000 wheat kernels with 6% colored kernels (3% red and 3% green) was also counted and divided in the SPD prototype. The percentage of colored kernels ending up in the sample cup after the dividing was calculated. The colored kernels symbolized infected kernels and the aim of the test was to see whether the colored kernels were represented in the sample cup.
Test 5
Numberbased division of a batch with mixed kernels of wheat, rice, oats, and barley
Approximately 2,000 kernels (based on the individual 1,000kernel weights) from each crop were mixed in one batch and run through the SPD prototype. The test was performed to see whether the shapes and sizes of the grains affected the outcome of the sample in the cup. The kernels in the sample cup were separated and counted.
Test 6
Weightbased division for repeatability and transferability between 8 SPD 4200 units
Samples of 300 g were used as a starting weight for all five crops. After dividing, the collected sample was weighed and recycled to the 300 g sample. Each 300 g sample was measured five times on each of the eight units to determine the repeatability and transferability.
Results and Discussion
The chosen crops all had different physical properties in terms of weight, shape, and size distribution. Therefore, the volume and weight for both 8,000 kernels and 1,000 kernels were calculated and evaluated. Table 1 gives the mean values and standard deviations for the starting volumes and weights of the 8,000 kernel batches for each crop used for testing the prototype. It can be seen here that oats has the largest volume for 8,000 kernels, followed by barley, wheat, and rice. On the other hand, if the 8,000 kernels were collected by weight, barley was the heaviest kernel followed by oats, wheat, and rice. The 1,000kernel weights for the crops were calculated and can be seen in Table 2 together with values in the literature of 1,000kernel weights for the same type of crops. All 1,000kernel weights were in the same range as the reference values.
Table 1: Mean* starting volumes and weights with std dev. for crops before first division 
Crop 
Volume8.000 [dl]

Std dev. 
Weight8.000 [g]

Std dev. 
Wheat 
3.5 
0.16 
292 
13 
Rice 
3.4 
0.05 
211 
1 
Oats 
5.7 
0.12 
320 
5 
Barley 
4.8 
0.11 
345 
1 
*Means of three 8.000 kernel batches 
Table 2: 1,000kernel weights for wheat, rice, oats, and barley 
Crop 
1.000kernel weight [g]



Reference values 
Wheat 
36.5 
30–40* 
Rice 
26.4 
14.4–32.7** 
Oats 
40.0 
30–45* 
Barley 
43.1 
30–45* 
* (Bauder, 1999);**(Bhattacharya, 2011) 
To see how well the size randomization worked, the difference between the 1,000kernel weight before the first division and before the last (eighth) division was calculated. This was to check whether the instrument was passing smaller kernels earlier than larger kernels. An increase in 1,000kernel weight for an increasing number of dividing circles would indicate that the smaller kernels of a batch were favored during the division. Figure 6 shows that the batches Oat III and Wheat I had this behavior, however, for the rest of the batches for these crops and for barley, no clear trends or significant variations were seen. Furthermore, a slight increase in 1,000kernel weight was also seen for the rice samples, but this increase was between 0.008–0.1 g which is <4 kernels and therefore assumed to be negligible. The difference in 1,000kernel weights between the batches for each crop was larger than the variation within each dividing series.
The first division is the most important since it is going to be used by the customers. Therefore, the number and/or weight of the kernels collected in the sample cup was counted/weighed after the first division, and the dividing percentage was calculated on both weight and number basis. According to theory, it was expected that 3.125% of the sample should end up in the sample cup. This was verified by the result obtained from Test 1 using rapeseed which gave a mean value of precisely 3.125 % (weight basis). It could therefore be confirmed that the design of the SPD 4200 is working. The dividing percentages for the other crops are displayed in Table 3. It can be seen here that the dividing percentage for wheat and barley did not differ significantly from the theoretical value, while oats and rice gave a significantly larger dividing percentage. This is believed to be related to the size and shape of the grains, since both rice and oats have a more elongated shape than the other crops. In addition, the rice hull has a hairy surface. The colored kernels in the batches were supposed to represent defect kernels. These kernels were found in the sample cup for all dividing cycles for all crops and batches, indicating that an infected batch will be found.

Figure 6: 1,000kernel weights before each dividing cycle for all batches. 
Table 3: Dividing percentage for the first division cycle for 8,000 kernel batches 
Grains 
Kernels in sample cup 
Std dev. 

[%Weight] 
[%Number] 
Weight 
Number 
Rapeseed(1) 
3.13 
 
0.03 
 
Wheat(2) 
3.03 
3.06 
0.19 
0.17 
Oats(2) 
3.22* 
3.27* 
0.13 
0.12 
Rice(2) 
3.23* 
3.30* 
0.15 
0.17 
Barley(2) 
3.03 
2.99 
0.23 
0.21 
*Significantly different from the theoretical value of 3.13% (P<0.05). (1)Mean value of 3 samples; (2)Mean value of 9 samples

When increasing the batch sizes to 24,000 kernels, no significant differences were seen compared to the 8,000kernel batches. It was therefore concluded that the batch size does not affect the total dividing percentage.
Furthermore, a mixed batch containing approximately 2,000 kernels from each crop (not rapeseed) was divided to see if the kernels that ended up in the sample cup were still equally represented or not. However, no clear conclusions could be drawn from this batch size.
In the transferability test, twoway Anova was used to calculate the significance between the eight instruments using five measurements per crop and instrument. No significant differences (P<0.05) were found between the instruments. Therefore, all measurements from each crop were used in the calculation of the mean [%_{weight}] and standard deviation for the divided samples, see Table 4.
Table 4: Dividing percentage for the division of 300g samples1using the final design

Grains 
Kernels in sample cup 
Std dev. 

[%Weight] 

Rapeseed 
3.03 
0.06 
Wheat 
3.01 
0.17 
Oats 
3.15 
0.20 
Rice 
3.12 
0.17 
Barley 
3.05 
0.22 
(1)Mean values of 40 samples 
When comparing the performance between the prototype and the final design of the SPD 4200 (Tables 34) it is clear that the performance of the SPD 4200 was improved by the final design. No crop, except rapeseed, was found to be significantly different to the theoretical value of 3.125% in the final design of the SPD 4200.
Conclusions
The design of the SPD 4200 prototype was proven to be correct, dividing an optimal shapedsample (rapeseed) into the theoretical dividing percentage of 3.125. However, the size and shape of the samples affected the dividing percentage; more elongated samples gave a larger percentage of kernels in the sample cup. The new and final design for the SPD 4200 gave slightly smaller dividing percentages for all crops compared to the prototype. Thus, the dividing percentage of the rice and oat kernels, which for the prototype were significantly larger than the theoretical value, decreased and therefore did not display any significant differences. Wheat and barley were still not significantly different from the theoretical value. Rapeseed became significantly different to the theoretical value with the new design due to a very small standard deviation, however, it did not differ from the dividing percentages of the other crops. The dividing was also concluded to be randomized, both regarding the kernel size distribution within a batch and which kernels ended up in the sample cup.
References
Bauder, J. (1999). Calculating the Amount of Seed to Plant http://www.montana.edu/cpa/news/wwwpbarchives/ag/baudr182.html  3rd July 2015
Bhattacharya, K. R. (2011). Rice quality  A guide to rice properties and analysis. USA: Woodhead Publishing.
ISO. (2010), Cereals and pulses  Determination of the mass of 1000 grains. ISO 520:2010
Lyseng, R. (2014). Old Technology Given New Lift With Unique Design http://www.producer.com/2014/02/oldtechnologygivennewlifewithuniquedesign/  6th July 2015
Sablani, S. S., & Ramaswamy, H. S. (2003). Physical and Termal Properties of Cereal Grains. In Handbook of Postharvest Technology: Cereals, Fruits, Vegetables, Tea, and Spices).
TNAU Agritech, P. Seed Technology http://agritech.tnau.ac.in/seed/Seed_seedtesting.html  7th July 2015