A Novel Classification of Grading Sizes on Production Processes

A Santhakumaran

Indian Institute of Food Processing Technology

Abstract

Classification of post production objects are most preferable measure in the market while

purchasing products and it helps consumers to get a right value for right product based on their

acceptance. Product grading into different classes is one of the safety measures of products in

transporting, marketing, handling and storing objects. The economic value of a post harvesting

objects depends on the safety measures of handling materials. They are packaging, transporting

from one place to another place and marketing the products lead to heavy losses. In such

situations grading sizes of objects helps to reduces loss of objects. This paper describes the

methods to implement the grading sizes in production processes. The methodology is applicable

to all shorting post production objects.

1. Introduction

During Second World War the country Japan has heavily affected. Now it led to

produce more reliable products than USA and Western Countries. Japan has utilized the

scientific quality control techniques on production processes to establish its name in the

world. Statistical Quality Control can through more light on an issue than a committee of

experts for making decisions on production processes. Quality control methods are used in

designing a product to increase the production usually in the range of 10% to 100% without

further investments for the products. Product classification based on quality and its sizes is an
 effective methodology for monitoring a production process through the use of control charts.

This is the emerging trends methodology of mechanization in the modern world to build the

global industrialization changes for reducing the losses in the post production processes.

Quality is one of the most important customer acceptance factors in the selection

among the competing products. Understanding and improving quality is a key factor leading to

business success, growth and an empowerment enhanced position. Hence grading products lead

to quality improvement which is an integral part of the over all business strategy. The effect way

to save the quality of the product is sorting the product into homogenous sizes.

Statistical process control is widely accepted for analyzing quality problems and

improving the performance of the production processes. It has been developed by Walter

Shewhart [9] of the Bell laboratories. The control charts have found favor with practitioners to

monitoring the large process shift. These charts are still popular because they are easier and

simpler techniques. The control chart is an online process monitoring technique widely used for

the occurrence of assignable causes of process shifts. The chart contains a centre line that

represents the average value of the quality characteristic corresponding to in control state. Two

other horizontal lines are called upper control and lower control limits. These control limits are

chosen so that if the process is in control, all of the sample points will fall between them. Any

one sample point falls outside the limits, and then the production process is an out of control.

Several new charts have been proposed by many researchers in the past few decades.

They are a good alternative when the production processes have different types of shifts. Some
of them are CUSUM [6], EWMA [7] control charts for Markov dependent sample sizes in which

an optimum way is to detect all types of process shifting (large, moderate and small) by taking

both the small and large samples under a switching rule of Markov dependent sampling[10].

Semi-circle control chart for variable [1] uses on 2D (Two – dimension) chart to control

location and variability simultaneously. A multivariate semi-circle control chart for variables

data [2] deals an alternative single variables control chart for multivariate data.

In this paper, Walter Shewhart Control Chart limits are used to sort out as small, medium

and large with reference to 2𝜎 and 3𝜎 control limits on the variable data. Neural network is used

to evaluate the accuracy of the product grading sizes. Data normalization to normalizes the data

[3,4,] has used. The feasibility of the product grading system is also physically verified. Results

show that grading sizes meet the market flexibility of consumers, increase the economic value of

producers and ultimately leads to wastage reduction. The combination of Shewhart Control

Charts and ANN provides a rapid, consistent and objective inspection technique which can be

expanded into many diverse area of sorting sizes. An example is illustrated the feasibility of eggs

grading sizes.

2. Sorting method

Shewhart control charts [8] work best for numeric data with normal assumptions. When there is

no special cause are present in the system, then there is approximately a 0.0027 likelihood of an

observation exceeding 3𝜎 control limits. For Shewhart control using 3𝜎 limits indicates that

false alarm occurs on average once every 1 / 0.0027 or 370.4 observations. The in control
average run length of the Shewhart chart is 370.4 observations. Similarly the processes are in

control, then 95.45% probability of an observation including 2𝜎 control limits.

Suppose if the underlying distribution is normal distribution, the 3𝜎 limit will fall short of the

upper 0.00135 chance while the lower 3𝜎 chance will fall below the 0.00135 chance. If chance

causes were present in the production processes, the probability of a observation falling either

above or below would be one out of a thousand. It means that the expectation of risk of looking

for assignable causes is positive variation. But the risk of searching for an assignable cause of

negative variation will be reduced under additive, normal and homogenous conditions.

Shewhart quality control chart 3𝜎 control limits are,

̅+3 𝝈
Upper control limit, UCL = 𝑿
√𝒏

̅−3 𝝈
and Lower control limit, LCL = 𝑿
√𝒏

∑ 𝑋̅
Where 𝑋̅ = 𝑛 𝑖 where 𝑋̅𝑖 is the average of ith sample with subgroup size n. Subgroup size

can be n =2, 3, 4,… and 2𝜎 control limits are

𝝈
̅+2
Upper specification limit, USL = 𝑿
√𝒏

̅−2 𝝈
and Lower specification limit, LSL = 𝑿
√𝒏

The classification is based on the reduced risk of Shewart control chart. They are small, normal

and large. Small grade object is chosen as small chance of observations falling in the interval

LCL to LSL with low risk probability 99.73%. Normal grade size is selected with probability of

an observation falling in the interval LSL to USL is slightly high risk probability 95% as

compared to extreme tails. Large grade size is sorted as the interval USL to UCL of low risk for
probability 99.73%.The sorting range of three

sizes is proposed in Table1.

Table 1. Sorting three grade sizes

Grade size under normal Range Number of items

assumption

Small (In extreme left an LCL to LSL N1

observation falls out side 3𝝈

limit of the normal

distribution is 0.00135 )

Medium (In middle part an LSL to USL N2

observation falls out side

either above or below 2σ

limits of the normal

distribution is 0.0455)

Large (In extreme right an USL to UCL N3

observation falls out side

𝟑𝝈 limit of the normal

distribution is 0.00135)

Total - N =N1 + N2 + N3
The grading size of objects is illustrated for considering egg procured from open market of

Salem. The geometry of the egg major and minor axes, weight and volume are measured. Section

3 describes these measures.

3. Geometry of Egg

Individual eggs are measured using digital weighing machine and the diameter has been

determined. Dimensions of the eggs are taken manually using Vernier Calipers. The

measurements are illustrated in Figure 1

b c

Figure 1 Specimen egg showing major and minor axes

Two dimensional shape of an egg is drawn by joining two ellipses at their minor axes.

The eggs equatorial radius (a cm) short polar radius (b cm) and long polar radius (c cm) are

known, then the surface area of the eggs are calculated based on the formula,

𝐛𝟐 𝐚 𝐜𝟐
i.e., Surface area = 2𝝅𝒂𝟐 + 𝝅𝒂 ( 𝐜𝐨𝐬 −𝟏 ( ) + 𝐬𝐢𝐧−𝟏 )
√𝐛 𝟐 −𝐚 𝟐 𝐛 √𝐛 𝟐 −𝐚𝟐

in cm2

Similarly the volume of the eggs are found by the formula,

𝟐𝝅
i.e., Volume of Egg = 𝒂𝟐 (𝒂 + 𝒄) in cm3
𝟑

The weight, surface area and volume of the eggs are given in Table 2
 Table 2 Geometrical measures of weight, surface area and volume of Eggs

S Weight(gm) Surface Volume(cm3)

a b c
No. Area(cm2)

1 2.20 2.5 3.3 61.34 74.11 58.76

2 2.60 2.7 4.2 61.85 104.21 97.64

3 2.30 2.7 3.2 67.82 79.29 65.33

4 2.15 2.4 3.7 60.77 74.99 59.02

5 2.15 2.6 3.1 57.25 71.03 55.15

6 2.10 2.5 3.4 57.29 70.92 54.46

7 2.15 2.4 3.4 58.25 72.07 56.12

8 2.15 2.6 3.4 61.37 73.92 58.05

9 2.25 2.6 3.5 62.11 79.20 65.64

10 2.15 2.7 3.2 58.74 72.93 57.09

11 2.05 2.5 3.2 50.69 67.01 50.14

12 2.10 2.2 3.1 52.48 65.39 48.92

13 2.25 2.5 3.5 66.62 78.23 63.58

14 2.10 2.3 3.2 50.84 67.22 50.77

15 2.20 2.4 3.3 59.54 73.17 57.75

16 2.05 2.2 3.5 53.33 67.15 50.14

17 2.25 2.6 3.8 69.27 82.25 67.82

18 2.20 2.5 3.4 61.74 75.10 59.77

19 2.25 2.4 3.5 65.47 77.27 62.52

20 2.15 2.3 3.7 61.89 74.08 58.05

21 2.10 2.5 3.4 56.52 70.91 54.46

22 2.10 2.4 3.1 51.72 67.18 50.77

23 2.30 2.4 3.4 67.24 78.39 64.22

24 2.25 2.3 3.3 62.11 74.31 59.34

25 2.25 2.3 3.6 62.13 77.34 62.52

26 2.25 2.6 3.5 67.90 79.20 64.64

27 2.10 2.4 3.1 52.33 67.18 50.77

28 2.25 2.6 3.2 60.85 76.18 61.46

29 2.20 2.5 3.3 60.39 74.11 58.76

30 2.15 2.5 3.6 62.12 74.94 59.02

31 2.15 2.6 3.2 58.54 71.99 56.12

32 2.10 2.4 3.4 56.47 70.01 53.54

33 2.25 2.7 3.3 66.76 78.15 63.58

34 2.1 2.4 3.4 53.95 70.01 53.54

35 2.15 2.30 3.4 56.44 71.16 55.15

36 2.20 2.30 3.4 61.57 73.22 57.75

37 2.10 2.20 3.4 51.55 68.22 51.69

38 2.15 2.40 3.3 56.28 71.10 55.15

39 2.10 2.20 3.5 53.84 69.17 52.62

40 2.35 2.50 3.6 71.85 83.64 70.51

41 2.20 2.40 3.2 59.56 72.19 56.73

42 2.10 2.50 3.1 57.07 68.08 51.69

43 2.05 2.50 3.2 52.67 67.01 50.14

44 2.15 2.60 3.2 58.17 71.99 56.12

45 2.25 2.60 3.2 62.08 76.18 61.46

46 2.20 2.70 3.2 62.11 75.03 59.77

47 2.15 2.70 3.2 60.61 72.93 57.09

48 2.15 2.50 3.0 55.72 69.15 53.22

49 2.20 2.30 3.2 58.76 71.25 55.72

50 2.15 2.70 3.4 62.12 74.86 59.02

51 2.15 2.60 3.3 57.97 72.96 57.09

52 2.15 2.60 3.1 57.45 71.03 55.15

53 2.1 2.40 3.1 55.40 67.18 50.77

54 2.15 2.70 3.6 61.12 76.80 60.96

55 2.10 2.30 3.2 50.35 67.22 50.77

56 2.10 2.4 3.0 53.14 66.24 49.85

57 2.30 2.70 3.1 62.12 78.28 64.22

58 2.20 2.60 3.2 60.14 74.07 58.76

59 2.15 2.7 3.1 58.52 71.97 56.12

60 2.15 2.5 3.5 62.11 73.97 58.05

61 2.25 2.3 3.4 62.09 75.31 60.40

62 2.20 2.7 3.0 58.45 73.07 57.75

63 2.20 2.9 3.4 65.21 78.92 63.82

64 2.10 2.7 3.2 55.64 70.85 54.46

65 2.20 2.3 3.6 61.99 75.21 59.77

66 2.25 2.5 3.1 61.47 74.22 59.34

67 2.25 2.6 3.5 65.50 79.20 64.64

68 2.05 2.2 3.3 50.50 65.29 48.38

69 2.2 2.6 3.4 62.12 76.04 60.79

70 2.0 2.2 3.1 46.36 61.52 44.37

71 2.2 2.5 3.5 63.70 76.09 60.79

72 2.05 2.2 3.1 48.21 63.44 46.62

73 2.25 2.6 3.5 65.43 79.20 64.64

74 2.15 2.5 2.9 56.37 68.20 52.25

75 2.15 2.5 3.3 58.13 72.03 56.12

76 2.35 2.6 3.2 68.06 80.46 67.05

77 2.15 2.6 3.1 56.46 71.03 55.15

78 2.15 2.8 3.4 60.87 75.80 59.99

79 2.0 2.4 2.9 47.09 61.45 44.37

80 2.2 2.7 3.1 56.45 74.05 58.76

81 2.1 2.5 3.1 54.17 68.08 51.69

82 2.2 2.7 3.1 59.32 74.05 58.76

83 2.1 2.6 3.2 57.30 69.93 53.54

84 2.15 2.7 3.3 62.12 73.89 58.05

85 2.15 2.6 3.5 62.11 74.89 59.02

86 2.05 2.5 3.3 52.76 67.93 51.02

87 2.10 2.4 3.2 55.31 68.12 51.69

88 2.20 2.8 3.3 62.11 76.97 61.80

89 2.25 2.8 3.3 67.56 79.13 64.64

90 2.10 2.5 3.1 55.28 68.08 51.69

91 2.10 2.7 3.1 59.15 69.91 53.54

92 2.30 2.8 3.6 73.99 84.40 70.87

93 2.25 2.7 3.3 67.32 78.15 63.58

94 2.20 2.6 3.5 65.69 77.03 61.80

95 2.30 2.6 3.2 65.60 78.31 64.22

96 2.10 2.5 3.1 55.62 68.08 51.69

97 2.20 3.0 3.2 66.38 77.92 62.81

98 2.15 2.4 3.2 56.29 70.14 54.18

99 2.05 2.4 3.2 50.86 66.12 49.26

100 2.0 2.5 3.1 51.02 64.11 46.89

 Eggs are sorted into three groups such as small, normal and large based on the interval

LCL to LSL, LSL to USL, and USL to UCL respectively. After predicting the small, normal and

large grading size ranges, Neural Networks Back Propagation algorithm is used for evaluating

the accuracy of the product grading sizes using MATLAB Software Version 7.7.0.47 (R 2008b).

4. Construction of Control Charts

For the construction of control charts, 100 Eggs have been considered. Control charts

present graphic display of process stability or instability over time. The important reason for
using control chart is to get the process stability. Process stability is the state in which a process

has displayed a certain degree of consistency in the past and is expected to continue to do so in

the future.

An outlier observation can be made based on the ratio as

𝐋𝐚𝐫𝐠𝐞𝐬𝐭 𝐯𝐚𝐥𝐮𝐞 𝐨𝐟 𝐭𝐡𝐞 𝐬𝐞𝐭

≥𝟐
𝐒𝐦𝐚𝐥𝐥𝐞𝐬𝐭 𝐯𝐚𝐥𝐮𝐞 𝐨𝐟 𝐭𝐡𝐞 𝐬𝐞𝒕

or an observation falls outside the 3𝜎 control limits (UCL/LCL). If any outliers are in the

observations, then they can be removed in the set and then the modified Shewhart control chart

limits are used to find out the grading sizes. The outliers in the data violate the sample comes

from the normality conditions.

The Shewhart Control Charts with standard deviation are separated into different grades

based on the weight, surface area and volume. Using SPSS software the Shewhart Control Charts

with standard deviation for sample subgroup size n = 4 are calculated. The charts corresponding

control limits are shown in Table 3.

Table 3 Shewhart Control Chart Limits

Control Limits Weight Surface Area Volume

UCL 63.55 79.8 64.67

USL 62.13 77.31 62.67

CL 59.30 74.09 58.68

LSL 56.47 70.87 54.69

LCL 55.05 69.26 52.70

 5. Determination of Grade Sizes

Control charts [5] are used to classify the eggs with the help of 3𝜎, 2𝜎 control limits. For

the weights of the eggs, the UCL, LCL and CL are 63.55, 55.05 and 59.30 respectively followed

by surface area 78.9, 69.76 and 74.09 and volume 64.67, 52.70 and 58.68 respectively. These

limits are used to sort out the grading sizes. Table 4 shows the grade sizes of the eggs respective

geometrical measures by physical verification.

Table 4 Grade Sizes of Eggs based on Weight, Surface Area and Volume

Surface Volume
Weight of eggs
area of eggs of eggs

Grading size Range in gm No. of Range in cm2 No. of Range in cm3 No. of

eggs Eggs Eggs

Small 55.05 - 56.46 33 69.26 - 70.86 34 52.70 - 54.69 32

Medium 56.47 - 62.13 47 70.87 - 77.31 47 54.70 - 62.67 48

Large 62.14 - 73.99 20 77.32 - 104.20 19 62.68 - 97.64 20

Total 100 100 100

6 . Evaluation of the manual classification

Machine vision is a powerful tool in its ability to remember, recognize and efficient, but it is

unable to think and perform like human brain. Researchers understand these and possibly

develop computer vision system for its operation as similar function of the human nervous

system. Neural networks refer to computing systems utilizing a large number of interconnected

neurons which perform simple computations and convey a signal from one neuron to another

through the connections as in the biological system. An artificial neural work is characterized

by its architecture, activation function for node output, and learning algorithm for determining

the weights of the connections. It can be classified according to their supervised or unsupervised

mode of learning. The introduction of supervised learning procedures based on the back –

propagation of error. There are many different variations of the back-propagation algorithm

which continue to be the most widely used supervised learning method for classification. The

data available for the development of a neural network are divided into training and test set data.

The training set is used in the model development, and network performance is evaluated both

for training and test set data. Product grading into different classes have used the feed forward

multilayer neural networks trained with back propagation algorithm. Product classification

accuracy is evaluated using Leven berg Marquordt a back propagation algorithm 1 – 1 – 1

neural network with the input neuron corresponding to small, medium and large sizes of the

physical measures of the Egg. One neuron in the hidden layer and a single neuron in the output

layer indicating the numerical value of physical characteristics are found to be more appropriate.

The accuracy of the manual classification is valuated using artificial neural network learning

algorithm Levenberg Marquordt . The table shows the coefficient of determination R 2 [11]
for evaluating the grading sizes of weight, surface area and volume of the egg. The manual

physical verification and the learning process evaluation is almost same. The unceternity of the

response variable of the grading size is least explained by the independent variables weight,

surface area and volume.

Table 5 Accuracy of the manual verification using neural net work algorithm

Egg Grade size Accuracy Error Number of eggs

Evaluation

𝑅2

Weight Small size 55.05 - 56.46 98.9% 1.1 % Small - 33%

gm Medium 56.47 - 62.13 Medium – 47 %

Large 62.14 - 73.99 Large – 20 %

Surface Small 69.26 - 70.86 99.6 % 0.4% Small -34 %,

Area Medium 70.87 - 77.31 medium- 47 %

Cm2 Large 77.32 - 104.20 large -19%

Volume Small 52.70 - 54.69 98.7% 1.3 % Small -32%

Cm3 Medium 54.70 - 62.67 Medium –48%

Large 62.68 - 97.64 Large – 20%

7. Conclusion

All the three geometric measures, the egg grading size is consistent. They are small size egg 33

%, medium size egg 47 % and large size egg 20 %. The results are assured about 99.07% by the

Neural Net Work algorithm as well as manual classification. Classification of grade sizes are the

most preferable measure in the market while purchasing products and it helps consumers to get a

right value for right product based on their acceptance. The grading sizes into three groups

reduce the loss of items considerably while transporting, storing and handling the objects. It

helps to maintaining a homogenous size in single grading for reduces the loss items in costs and

material wises. If adding computer vision for sorting quality and there by grading sizes can be

enhanced customer good will and loyalty, but also in terms of costs for scrap and rework.

Grading sizes of an object is decided to keep the process within specification for

extreme items 99.73 % and normal items 95.45 % of the time. The novel technology ±2σ

and ±3σ normal distribution specification limits for classifying grading egg sizes are same to the

American Egg Board classification [11]. Under this method the small size egg range is

The combination of Shewhart control charts techniques, the ANN learning processes and

physical verification can be used into much diverse area for reducing the losses of production

processes. This is an approach similar to Six Sigma program developed by Motorola in the late

1980s as response to the grading size on production processes. The grading size of Small size

egg is classified as weight 1.8 ounces(51.0291 gm), Medium size egg is stated as weight

2.1 ounces (59.535 gm) and Large size egg weight is 2.4 ounces ( 68.045 gm). On a

mission with a vision, the novel method of grading includes the American Board of

Classification of eggs. It helps to develop for designing automatic machines to grade the egg
sizes into three groups small, medium and large as weight < 56.47gm, 56.47 – 62.13gm and >

62 .13 gm respectively. Production processes are the emerging trends for developing automation

machineries in modern industries. Product grading into different classes helping hands to achieve

the goal of production processes of automatic mechanizations.

References

1. Cho, M. T . and Cheng, S , W. (1996). Semi circle control chart for variable data.

Quality Engineering, 8(3), pp. 441-446.

2. Clyton, (2001) . ARL comparisons between Neural Network Models and X bar

control charts for quality characteristics that are non- normality distributed,

Economic Quality, 16, pp.5 – 16.

3. Jayalakshmi, T. and Santhakumaran, A. (2010). Statistical Normalization and back

Propagation for classification, International Journal of Computer Theory and

Engineering , 3(10), 89,

4. Jayalakshmi, T. and Santhakumaran, A. (2010). A novel classification method for

diagnosis of diabetes mellitus using artificial neural networks, Proceeding of the

International conference on data storage and data engineering , IEE computer

society, pp. 159 – 163.

5. Montgomery, D.C. (2009). Introduction to statistical quality control, 6th. Edition,

John Wiley & Sons. Inc.

6. Page, E.S.(1954). Control inspection schemes, Biometrika, 41, pp.100 – 115.

7. Patel, H. I.(1993). Quality control methods for Multivariate Binomial and Poisson

distributions, Technometrics, 15, pp. 103-112.

8. Shewhart, W.A.(1926), Quality control chart, Bell Syst. Tech. J., 5,pp. 593 – 602.
9. Sivasamy, R, Santhakumaran, A. and Subramanian, C.(2000). Control charts for

Markov dependent sample sizes, Quality Engineering, 12 , pp. 593 – 601.

10. Santhakumaran,A.(2016), Governing principles of mathematical modeling, Apple

Academic Press, Inc. [AAP], USA.

11. https:/www.southernstates.com/articles/size-grade-eggs.appx/ How to grade and

size eggs?- Southern States Cooperative .