Statistical process control (SPC) is a form of feedback process control that does for discrete products what automatic process control does for the chemical process industry. Its purpose is to distinguish special or assignable causes from random or common cause variation. SPC is often part of the process’ control plan that is required, for example, by IATF 16949 (clause 8.5.1.1).
The successful application of SPC requires not only control charts, but also proper selection of the rational subgroup; a sample that reflects all the variation in the process. The normal distribution or bell curve assumption on which traditional SPC relies is meanwhile far more common in textbooks than in real processes. This presentation will include off the shelf methods for handling non-normal distributions.
Seminar Agenda:
- Common cause or random variation cannot be adjusted out of a process, while special or assignable cause variation can. Overadjustment, or attempts to adjust common cause variation out of the process, actually makes variation worse.
- SPC charts are visual controls that tell us when to adjust the process and when to leave it alone. SPC charts can detect (1) undesirable changes in the process mean and (2) undesirable increases in process variation.
- SPC is a hypothesis test that begins with the null hypothesis (starting assumption) that the process is in control. A point outside the control limits proves the alternate hypothesis (that the process is not in control) beyond a quantifiable reasonable doubt.
- This quantifiable reasonable doubt is 0.27% for each sample in the traditional Shewhart chart (X or x-bar) for process mean. If the process is centered on the nominal, there is only a 0.00135 chance of going over either control limit, for a combined risk of 0.0027.
- It is vital to define the rational subgroup properly. Control charts with inexplicable out of control signals outside the lower and upper control limits are evidence that there are variation sources that are not accounted for by the subgroup.
- Control charts with inexplicable out of control signals can also indicate that the underlying distribution is not a bell curve. It is however possible (e.g. as done by both StatGraphics and Minitab) to create charts whose limits are based on the non-normal distribution, and therefore function properly for control purposes.
- The probability plot and histogram can be used to test the assumption that the data fit the normal (or other selected) distribution.
- The central limit theorem mitigates the effect of non-normality on charts for sample averages (x-bar charts) but it is nonetheless mandatory to identify and fit the underlying distribution to get accurate process performance index (Ppk) results because individual measurements, as opposed to averages, are in or out of specification.
- Control chart setup (including for non-normal distributions) will be covered.
- Use of control charts on the shop floor; out of control signals require action, as defined by the work instruction or out of control action plan (OCAP), which is part of the control plan.
Attendees will receive a pdf copy of the slides and accompanying notes for the presentation, as well as a simulator (works in Windows 7) that places simulated gun targets side by side with control charts to illustrate the manner in which the charts reflect shifts in the process mean or increases in process variation.
Who Should Attend:
- Quality Assurance Departments
- Quality Control Departments
- Research and Development Departments
- Manufacturing Departments
- Engineering Departments
- Operations Departments
- Production Departments
- QA/QC Technicians
- Manufacturing Technicians
ST3084
