In the last blog post I demonstrated a method to give clues for root cause analysis using the Feature Selection tool in STATISTICA Data Miner. In essence I was trying to predict if any of the inputs could explain changes in the output. This process has been referred to by others as Predictive Quality Control.
Today I want to talk about in what situations we would want to use this Predictive Quality Control method. Obviously if the data collected from the process are good then there is no need to investigate small fluctuations in the metric of interest. So this begs the question of when does it become necessary to perform root cause analysis?
To help answer this question I would like to refer back to the machine data from the last blog post. Remember that the output was a Quality Score and the inputs consisted of various machines, machine users, and lots of materials. Let’s assume that the data came from one day’s worth of production. Let’s also assume that there were 19 days of production data that we have collected previously.
Here is what the SPC chart looks like for this hypothetical situation:
The following video takes into account this hypothetical scenario and motivates the use of statistical process control charts to signal when to use the predictive process control method discussed last time.