Resources - Quality Systems

Statistical Process Control

Description SPC is a method of using statistics applied to the results of a process to control the process. Historical data of the performance of the process (or operation of hardware) are statistically analyzed to predict future performance or to determine if a process is "in control." A process is defined as "in control" if there are only random sources of variation present in the process and the associated data. In these cases, the data can correctly be investigated with the standard methods of statistical analysis. If the data are not "in control," there is some special cause of variation present in the process, and this is reflected in the data from that process.

In these cases, this section on SPC assumes that the data variability is still reasonably distributed around the mean, and these procedures are applicable. If these procedures lead to a result of special cause variation at nearly every data point, these procedures cannot be correctly applied.

Application This technique is used to determine if special causes of variation are present in a process, or if all variation is random. In other words, SPC is used to ensure that a product is being produced consistently, or is about to become inconsistent. Thus, SPC can be used to isolate problems in a process before defective hardware is delivered. This technique can be used for measurement type data (real numbers) or attribute data. There are two types of attribute data - binomial data and poisson data.

  • Binomial data have a given number of outcomes, e.g., three of four parts on an assembly can be defective.
  • Poisson data have an unlimited number of possible outcomes, e.g., a yard of material may have 1, 10, or 100 flaws.

Procedures The basic steps for conducting SPC are:

  1. Decide how to group the data. Subgroups should be chosen to show the performance of the part or process of interest. For example, if a machine is producing several parts at a time, the parts produced at one time will be a logical subgroup.
  2. Construct a control chart and range chart.
  3. Determine and apply control limits to the data.
  4. Determine if any control limits are violated. If any control limits are violated, a special cause is indicated. In addition to the specific control limits, the analyst must examine the data plot for other visual indications of special causes in the data. Any particular pattern, for example, would indicate a special cause is present. The use of engineering judgment is critical to extracting the maximum amount of data from the SPC plots.
  5. Determine the special cause. This may require Pareto analysis or engineering judgment using past experience.
  6. Implement a fix for the special cause of variation.
  7. Plot the data to ensure that the fix has been effective.

Source: System Engineering "Toolbox" for Design-Oriented Engineers, Sec 7. - NASA/RP-1358

Certisafety Section Home Page



Copyright ©2000-2019 Geigle Safety Group, Inc. All rights reserved. Federal copyright prohibits unauthorized reproduction by any means without permission. Disclaimer: This material is for training purposes only to inform the reader of occupational safety and health best practices and general compliance requirement and is not a substitute for provisions of the OSH Act of 1970 or any governmental regulatory agency. CertiSafety is a division of Geigle Safety Group, Inc., and is not connected or affiliated with the U.S. Department of Labor (DOL), or the Occupational Safety and Health Administration (OSHA).