other method for rating production quality must be used. Often the most practical method of checking a part or a characteristic may be to classify it as being either good or bad. That is, it either conforms to the specification or it does not conform. We use the term "attribute" to describe a characteristic which is classified in this manner. There are two types of attributes:
1. Where measurements are not possible-- visually inspected items for color, missing parts, scratches and other damage. 2. Where measurements can be made but are not because of time, cost or need. At this point we should probably distinguish between the terms "nonconformity" and "nonconforming." A nonconformity is a quality characteristic which does not conform to specifications. A nonconforming product is an item which is not useable as intended because it has one or more nonconformities. VARIABLE CHART LIMITATIONS
Variable control charts are excellent for controlling and improving quality but they have certain limitations. One obvious limitation is that they must be used only for characteristics which are measurable. They cannot be used for characteristics which are attributes. A further limitation is that they can be used for only one characteristic. This means that there must be a control chart for each characteristic to be controlled. It can be seen that this could prove costly and burdensome. The attribute chart can minimize this limitation by providing overall quality information economically. TYPES
OF ATTRIBUTE CHARTS
The following types of attribute charts find application where control charts for variables are not considered advisable. 1. p-charts -- These record the proportion of nonconforming parts in a sample. This proportion is expressed as a decimal fraction. A variation of the p-chart which is somewhat easier understood by others is the 100p-chart. This actually represents the percent nonconforming. Yet another variation of the p-chart is the np-chart which is based on the number of nonconforming parts found in each sample of material. 2. c-charts -- these charts are based on the number of nonconformities. 3. u-charts -- these charts are based on the number of nonconforitiesing per unit. Care should be taken to detect the differences in the above definitions with respect to the terms nonconforming and nonconformity. Generally, the charts for nonconforming are the p, 100p and np charts, while the charts for nonconformities are c and u charts. p-CHART CONS
TRUCTION This chart is probably the most popular and widely used of all the attribute charts. The number of nonconforming parts in the sample is divided by the number of items in the sample. The quotient is the fraction nonconforming. Then multiplied by 100 it becomes the percent nonconforming. The charts are usually kept as fraction nonconforming charts. Preliminary Considerations As with the variable charts, the success of fraction nonconforming charts depends to a great extent on the care and thought put into its conception. The characteristic, or characteistics, to be studied must be defined and carefully selected. The frequency of check and the sample size should be selected to provide adequate information. S
ome very beneficial aid may be obtained by analyzing the Military S
tandard S
ampling Procedures and Tables for Inspection by Attributes. No simple rule can set the frequency of check or the sample size, but these tables can provide a guide. Provision should be made, where necessary, to facilitate the selection of a random sample of the product. Normally this may be accomplished quite easily, but occasionally a rather elaborate technique is necessary. Testing fixtures, gauges and equipment must be analyzed to insure accurate and reliable results. If a visual inspection is performed, marginal standards should be developed to provide a verifiable comparison. Limiting shades of color, or a limiting roughness of surface may be obtained and these parts retained as a master against which to check production. Unless the inspection data is capable of close duplication, it is of little value and no quality control chart can enhance its reliability even a little bit. THE p-CHART DATA S
HEET As in the data sheet for variable control charts, the data sheet for a p-chart should provide all the information necessary to completely identify the product, the operation, the inspection procedure and the characteristics to be checked. An example of the form such a data chart might take is provided on the following page. (INS
ERT TWO PAGES
OF PDATA S
HEET AND PCHART)
In addition, the data sheet should provide a means for recording the inspection information and transforming it into usable form. This usually means that a column for listing the number of nonconformities in each sample and a parallel column for recording either the fraction or percent nonconforming for that sample. The example is set up for fraction nonconforming The example also has provisions for additional analysis of the number nonconforming due to each of the various characteristics. This provides useful information when process changes are required. The sample chart has been filled in with inspection data on the number nonconforming found in each of twenty-five samples. Let's calculate the fraction nonconforming for each sample and record it in the appropriate column. Fraction nonconforming = sample size / number nonconforming The p-Chart
Here again the chart form should provide the same
identification information as that on the data sheet. This provides a chart which fairly adequately describes the operation without reference to additional forms. An example of such a form is shown on the following Page. The vertical scale should be constructed to permit plotting of all values which might be anticipated in the data. For our purposes we can use a scale of 1 inch = .05 fraction nonconforming. The plottings are then made, and as is common practice, they are joined by straight lines to avoid "loss" of any data. ANALYS
IS
OF THE p-CHART The first step in analyzing a p-chart is to calculate the center line or average line. This line is labeled p (p-Bar)
and is the average fraction nonconforming. The average fraction nonconforming may be calculated by the following formula. Total number of nonconforming parts observed p = -------------------------------------------- Total number of parts inspected For the percent nonconforming chart this value is obtained by multiplying the fraction by 100. (100)
(Total number of nonconforming parts)
100p = ------------------------------------------ Total number of parts inspected This value should be calculated for the sample problem, the p- Bar value drawn on the chart and properly labeled. After the center line has been calculated, the control limits can be determined. The formulas are shown for both fraction and percent nonconforming charts Fraction Percent Nonconforming Nonconforming Charts Charts Upper Control Limit p + 3 p(1-p)
100p + 3 100p(100-100p)
UCL(P)
n n Lower Control Limit p - 3 p(1-p)
100p - 3 100p(100-100p)
LCL(P)
n n Let's now calculate the limits for our example and draw them in place. It should be noted that often the lower control limit will turn out to be negative. When this occurs, the lower limit is usually not drawn on the chart. The converse is of utmost importance:
if no lower limit is drawn on the chart it means that the value is negative. It is important that the lower control limit be drawn on the control chart whenever it is positive or whenever it is zero. After the control limits have been drawn they should be labeled. INTERPRETATION OF THE p-CHART There are two questions which require an answer when a p-chart is analyzed. Is the process in control? Is the process satisfactory? If all the plottings are within the control limits and are relatively evenly dispersed about the center line in a pattern similar to those of the Xbar and R charts discussed previously, the process is said to be in a state of control. In more general terms, we might say that the variation among the plottings is such that it could be expected to occur if all the plottings were based on samples taken from the same lot of material. The variation is apparently due only to sampling fluctuation. There is no indication of a process change. The process has been producing in a consistent fashion throughout the period in which the samples were taken. If the process is not in control, there is very little that can be said about the process capabilities. If the process has been operating in an erratic manner, it is nearly impossible to say anything about how good the process actually is. The source of erratic variation, the assignable causes, must be eliminated before any valid conclusions can be reached. If the process is in control, the average fraction nonconforming may be considered the expected quality level of the process. It has been producing material of a certain fraction nonconforming in the past and unless some process change is made, it can be expected to continue at approximately the same level. When this has been established, we should determine whether this level is satisfactory or not. No simple rule exists for setting an acceptable quality level. The level must be selected on the basis of knowledge concerning the function of the part, assembly difficulties, further processing, and the possibility of a nonconforming part actually getting into an assembly or a finished part. If the process is in control at a satisfactory level, the center line and control limits on the chart may be extended and used to maintain ongoing control. As with other control charts, it is important to observe the p-charts to learn of any chaotic condition. A plotting above the upper control limit would probably indicate that the change in the process is undesirable. A plotting below the lower limit would probably indicate that the change in the process is desirable. S
imilarly, production should be alert to detect trends which might occur as the process continues to be monitored.