Webinar: Using Average Run Length to Optimize Advanced Control Charts
One of the issues for Six Sigma and Quality practitioners using basic or advanced control charts has been what tests for special causes to use and what settings to use with those tests, or what parameters to use for an EWMA or CUSUM chart. This presentation introduces the concept of the average run length (ARL) as an approach to help the practitioner make these decisions. The average run length is the average number of runs before an out-of-control signal is given on the control chart. We desire the “In-Control” average run length (ARL 0) to be as large as possible and the “Out-of-Control” average run length (ARL 1) to be as small as possible. The economic consequences are clear, if ARL 0 is too low, we will be frequently chasing false alarms. If the ARL 1 is too high, we will fail to detect a significant shift in the process mean.
The calculations for ARL are quite complex, involving either Markov Chains or Monte Carlo simulation to solve. ARL Templates will be demonstrated that take care of these calculations and are easy to use.
The EWMA and CUSUM will be compared to the Shewhart Individuals chart with tests for special causes.
We will also introduce the Poisson and Binomial EWMA and CUSUM as alternatives to the classical attribute C and P charts.
The problem of robustness to non-normality will be considered by using the Pearson family to simulate any specified value of skewness and kurtosis and estimate the ARLs.
- Introduction to Average Run Length (ARL)
- Shewhart Charts with Tests for Special Causes
- EWMA Binomial Proportions
- EWMA Poisson Counts
- CUSUM Binomial Proportions
- CUSUM Poisson Counts
- Robustness to Non-normality
- Limitations of Average Run Length