Although Statistical Process Control is applicable to many areas it is most commonly linked with manufacturing processes and the need to ensure high quality in manufactured goods.
All manufacturing processes exhibit an intrinsic variation from one part to the next. No two parts are ever exactly the same size or have quite the same appearance. Variations in size mean that, while many parts will be produced close to the desired dimensions, some may be big or small enough to exceed the tolerance limits. Sometimes this can be corrected by re-setting but if the process has too great an inherent variability then it may produce both parts that are too big and parts that are too small. In such cases no amount of setting will make it produce parts that are reliably within tolerance.
Inspecting out bad parts is time consuming and wasteful since it does nothing to prevent bad parts from being made in the first place. Either the process itself must be changed to reduce its variability or a different process must be used.
An American engineer, W Edwards Deming, studying the problems of process variability in the 1930s concluded that variation came from two distinct sources. Special Causes, he reasoned, were particular identifiable problems in the process that, with a little effort on the part of the process operator, could be largely eliminated. Common Causes, on the other hand, were inherent in the process, e.g. vibration in machine tools, backlash in gears, temperature effects, and the operator could do little to affect them. Common causes needed management action.
To distinguish between special and common cause variations Deming developed the concepts we know as process control charts. Using carefully controlled conditions, Deming worked out the level of variation he could expect from a process. He plotted the limits of this variation as lines on a chart and then measured samples of parts from the process as it ran. By plotting the measured dimensions on his chart he could see when the process produced parts that were beyond his expected limits of variation. Investigation revealed an identifiable ‘special cause’ in every case and by eliminating these causes he was able to improve the process and produce parts more reliably and closer to the desired dimensions. He described a process running without special causes as being ‘in statistical control’ but admitted that this was not a natural state of affairs!
Process control ensures that a process is running efficiently within its own limits of variability. But are these limits good enough? Is the process producing parts sufficiently close to the desired dimensions as to be consistently usable?
Put simply, if the limits of process variation fall within the tolerance limits for the part being produced it may be said to be ‘capable’ of producing parts to that tolerance. Traditionally the process spread is taken to be the width of three standard deviations either side of the mean dimension, a total width of six times the standard deviation (sigma). By dividing the tolerance bandwidth by this 6 sigma value we can calculate a figure of merit for the process: the process capability index – often known as Cp. A Cp of greater than 1 indicates a generally capable process.
Current trends are moving towards higher capability requirements and it is common for Cp values of 1.33, 1.67 or even greater to be requested. This is part of a drive towards continuous quality improvement to reduce the loss due to poor quality.
A study, in the 1950s, of gearboxes for Ford trucks, determined that one plant had a significantly higher warranty returns rate than another producing the same gearboxes. Further investigation revealed that the plant with the higher rate was producing parts with dimensions spread over the entire tolerance band while the other plant was using typically only one third of the tolerance band. This lead to a change of thinking, away from the idea that ‘anything in tolerance is good,’ to one where the goal was to produce parts closer and closer to the required dimensions in order to reduce the costs of poor quality.