Level: Intermediate
Paul Deen, Program Manager, Data Science, ASM International
We expect most processes to behave normally. If there is a target and people are making an effort to hit that target, we expect the results from a process to follow a normal distribution. We then know that if things deviate from normal, the process has changed. Fix the process and stay on target to get the best version of your process. This is called statistical process control and the deviations are detected through control charts. What happens when there is a constraint? Quite often the data does not follow a normal distribution and no control chart platform can deal with this. The standard solution is to do a transformation on the raw data to make it more normal and then apply classical control charts.
Process control systems only work if they are executed live on the line and done by the operators.Transformed data does not work in that application. So instead of transforming the data we need to transform the control chart. By using JMP scripting, I was able to build a custom control chart script that allows the user to select a distribution and spec limits, and then generate a chart, including alarms and rates that can be used for process control on the line.
