To use SPC effectively, understand the concept of variation. When a product characteristic is measured repeatedly, each measurement is likely to differ from the last. This is because the process contains sources of variability.
When the data is grouped into a frequency histogramA bar graph that shows frequency of occurrence versus value. Quite often the data is fitted to a distribution such as a normal distribution. ., it will tend to form a pattern. The pattern is referred to as a probability distribution and is characterized in three ways:
Note: Most SPC techniques assume that the collected data has a normal distributionAlso known as a `bell' curve, the normal distribution is the best known and widely applicable distribution. The distribution is symmetrical and popularly represents the laws of chance. 68.27% of the area lies between -1 sigma and +1 sigma, 95.45% between -2 sigma and+2 sigma, and 99.73% between -3 sigma and +3 sigma. The values of skewness and kurtosis are used to provide quantitative measures for normality. Assuming that at least 20 samples are used to construct a distribution, a good rule of thumb is to accept the data as a normal distribution when, -1.0 = skewness = 1.0 2 = kurtosis = 4..
Variation is generally categorized into one of two types:
Statistics indicate that common variations account for about 85% of departures from process quality requirements. Usually these departures require solution at the management level.
Statistics indicate that special variations account for about 15% of departures from process quality requirements. Typically these departures require local action (equipment repair and so on) for solution.