Process Capability Analysis Of Extremely Non-normal Data
Reliability Plotting is a graphical technique that is a standard method described in some reliability textbooks. The method is used primarily for data that is problematic in one or more of the following ways: non-normal (e.g., a Fatigue-Life distribution), a mixture of distributions (e.g., the distribution looks bi-modal when arranged into a histogram), low precision (e.g., a large number of identical readings in a small sample size), and/or incomplete (e.g., when a study is terminated before all on-test devices can be measured, due either to measurement equipment limitations or due to time limitations). Reliability plotting can easily handle all such situations.
This method involves first creating a probability plot (Y = %cumulative vs. X = raw data). That step and all subsequent ones can easily and automatically be performed using an Excel spreadsheet.
Why should you Attend: The most informative method for analyzing the data that results from QC, Validation, or Engineering activities is the calculation of the product's or lot's "reliability" at a chosen "confidence" level (where "reliability" means "in-specification").
Such calculations are relatively simple when data is "normally distributed"; but if the data is non-normal and cannot be transformed to normality, then there is typically no simple way to calculate a reasonably accurate level of reliability. In such a situation, the best method for determining reliability is called "Reliability Plotting".
The output of reliability plotting is a definitive statement that the given product or lot has a specific % in-specification, which conclusion can be stated with a specific level of confidence (e.g., 95% confidence of 99% reliability, or 90% confident of 93% reliability"). Reliability plotting can be performed using an Excel spreadsheet and formulas found in almost any introductory statistics textbook.
Areas Covered in the Session:
- Definitions
- How to create a reliability plot
- How to use it to determine reliability
- Example, using typical data
- Exact vs. Interval plotting
- Examples using data from: mixed distributions, highly replicated values, or censored studies
- Comparison to use of K-tables, etc.
Who Will Benefit:
- QA/QC Supervisor
- Process Engineer
- Manufacturing Engineer
- QC/QC Technician
- Manufacturing Technician
- R&D Engineer
His experience as an instructor in applied statistics includes having given annual 3-day seminars for many years at Ohlone College (San Jose CA), and previously having given that same course for several years for Silicon Valley ASQ Biomedical. He's given numerous statistical seminars at ASQ meetings and conferences. And he creates and sells validated statistical software programs that have been purchased by more than 110 companies, world-wide.