Pitfalls in control charting and how to solve them, Per Vase, Senior Specialist, Manufacturing Intelligence, Novo Nordisk Engineering

To set control limits on individual observations right many conditions need to be fulfilled. Data must be normal distributed, and must be described by a single mean and one variance component. In addition a large historical data set is needed to minimize standard error on mean and especially standard deviation. 

If not all of these conditions are fulfilled, control limits can be very misleading. However, if you can describe your data set with a model, and calculate tolerance limits from this model, proper control limits can be calculated even though not all conditions are fulfilled.

From the event where The Danish Chemometric Society and IDA Kemi focused on process monitoring.