CONTROL CHARTS FOR V
ARIABLES SOME STATISTICAL PRINCIPLES
As was mentioned before, the normal curve has certain properties centered around two identifying characteristics -- the average and the standard deviation. Ordinarily the values (variables) measured for a given process which is operating under a chance cause system will generate a normal distribution. This can be observed in the frequency histogram. There are tests which can be conducted to measure normality. If there is any suspicion that the distribution is not normal, a test should be conducted. The average we usually refer to is the arithmetic mean. There are, however, other "averages" which indicate the central tendency of a distribution of measures. Two of these measures which are used in quality control are the median and the mode. The mean is calculated by adding all of the values and dividing by the number of values added. The equation is:
Xbar = sumX/N Median refers to the middle value of a set of ranked values. As an example, if we have the values 3,2,4,1,5 and we rank them we obtain 1,2,3,4,5. The median is 3 because it is the middle value. The median for the following set of values, 2,3,4,5 is 3.5 because that is the value which we would find in the middle. Mode is the most frequently occurring value. Thus in the following data, 2,3,5,1,4,3,6, the mode is 3 because it occurs two times. V
ariation refers to the way the data is spread or dispersed. If we have 1+2+3+4+5 we have a mean of 3 but we could also have the same mean if the values were 2+3+3+3+4. One way of expressing the measure of variability is to use the range. This is the highest value minus the lowest value. In our first example the range is 4 (5 - 1 = 4). In the second case the range is 2. It can be seen that in both cases the mean is the same, but we gain additional information about the nature of the data when we learn something about its variability. The second case is grouped more closely about the mean than is the first. Another measure of the spread is the standard deviation which is usually identified by the lower case Greek sigma ( ). We will use the letters SD to indicate Standard Deviation. It represents the concentration of values about the mean. One way to look at the standard deviation is to think of it as representing the average difference between the mean and each individual value. There are several forms of the equation for calculating the standard deviation. The basic form is:
SD = sumx2/N. Where x = X-Xbar. It would follow then that another form is:
SD = sum(X - Xbar)2/N. For calculators without the standard deviation function there is yet another form:
SD = sumX2/ - (sumX/N)2. Many calculators today provide two different standard deviation values. This revolves on statistical theory. Technically, when we have all possible values for a set of data we use the form of SD = sum(X-Xbar)2/N and when we have a collection of data which is only representative of the total population (That is, when we have only a sample of all the values) we use the form SD = sum(X-Xbar)2/N-1. Generally we do not use the latter form for quality control calculations even though we are looking at data which represents a sample of what the total would be. We will not know what the true number of items will be until we have produced the last item. The advantage of the standard deviation over the range as a measure of variability is that it provides a better insight into the pattern of variability. As was stated earlier, statistical theory tells us that 68% of the population will fall between one standard deviation unit above the mean and one standard deviation unit below the mean. Three sigma limits (standard deviation units) will include about 98% of the values. Knowing this then, it is possible to estimate the percent of values falling between any set of values. The equation for permitting us to make this estimate is:
z = X-Xbar/SD. Lets look at an example. The average length of a shaft being manufactured is 9.07
inches and the standard deviation of the process is 0.40 inches. (A) What percent of the shafts will measure less than 8.30 in.?
(B) What percent of the shafts will measure more than 10.00 in.?
(C) What percent of the shafts will measure between 8.00 and 10.10 kg.?
First let's calculate the z-values for each. 8.30 - 9.07
(A) z = ----------- = -1.925 .4 10.00 - 9.07
(B) z = ------------ = 2.325 .4 8.00 - 9.07
(C) z(8.00) = ----------- = -2.67
5 .4 10.10 - 9.07
z(10.10) = ------------ = 2.57
5 .4 We now look up the z-value for (A) in the tables and find that for 1.92 the proportion between the mean and 9.07
is .47
26 or 47
.26%. Since the question was the percent less than 8.30 kg. we must subtract this percent from 50. This chart is constructed to represent one side of the distribution. Since a normal distribution is symetrical about the mean, 50% of the area is above the mean and 50% is below the mean. Thus we find that:
50 - 47
.26 = 2.7
4 . We say that 2.7
4 percent of the bicycles will weigh less than 8.30 kg. For the part of the problem we find that a z-value of 2.32 yields a value of .4898. Subtracting this from .50 tells us that .0102 or 1.02% will weigh more than 10.00 kg. To determine the percent between 10.10 and 8.00 we look up each z-value and add them together. Following the accepted rounding rules we raise the odd values to evens when we drop the 5. Thus we will use the values of 2.68 and 2.58. For 2.68 we get the value of .4963 and for 2.58 we get .4951. Adding the two values gives a total of .9914. We can say that 99.14% of the weights will fall between 10.10 and 8.00 kg. LIMITS We need to consider three different types of limits as we prepare to move into the onstruction of control charts. These are SPECIFICATION LIMITS, CONTROL LIMITS, AND PROCESS LIMITS. Specification limits refer to the maximum and minimum dimensions permitted by the design of the product. These are the dimensions set by engineering. The only relationship this bears to what is being produced is that which pertains to attempts to meet those specifications. Process limits are the maximum and minimum sizes of the parts being produced. This is usually represented by three sigma limits either side of the mean generated by measurements of the parts. If the process limits are less than the specification limits we have a satisfactory process because all of the parts we make will meet specification. On the other hand if the process limits are larger than the specification limits we will be making some product out of specification. Some action will need to be taken. Some adjustment will need to be made in the chance cause system. If nothing can be done in this area then either the specification will need to be changed or it will be necessary to run and sort out the faulty parts. Control limits are employed in control charts and represent variation based on the averages of samples taken to monitor the process. Because these figures represent averages they will naturally be smaller than the actual values would be. They represent three sigma limits of the AV
ERAGES. PURPOSE OF CONTROL CHARTS We have been discussing how we collect data, analyze it, make assumptions about the process and compare the process limits to the specification limits. This is all very good, but it only tells us what has happened to completed material. We would like to know what is happening to the process while it is running. We would also like to run the process as close to the mean specification dimension as possible. This can be accomplished through the application of control charts. No other tool of statistical quality control has received the recognition which has been accorded the Shewhart control chart for variables. This chart has found wide usage and continues to be the "bread and butter" of nearly every statistical quality control application.