Why the evaluation has "so few" values is often seen as a problem and is one of the recurring questions on the hotline. Defined as outliers, the measured values can be excluded from the evaluation.
How to deal with outliers is an ongoing issue. It is up to the customer to decide whether and how to handle it. This is a multi-faceted and complex issue, which is covered in detail in the description of the evaluation strategy, but the focus here should be on general understanding. This explanation may need to be read iteratively several times. And it is recommended that it is discussed in the project or in a training session.
Outliers in the machine and the process capability analysis
The technical possibilities are the same in both modules, the Sample Analysis and the Process Capability Analysis. But the philosophical approach is different.
Outliers in process capability analysis are perfectly normal for many customers. Continuous data flow, automatic recording mechanisms, problems with incorrect measured values can occur anywhere. But not when it comes to sample analysis, to the analysing machine capabilities! A machine capability should be performed under control! So, can there be outliers? Not really. But these options are there for a reason. If there are outliers in the sample analysis, it can be assumed that the data recording is not under control. In this case, the customer can specify in the evaluation strategy whether the process should be output as limited capable or not capable. As an indication that the recording process is not OK.
Types of outliers
The outliers can be excluded from the evaluation automatically during an evaluation or manually by selection. Outliers can be defined based on limit values or mathematical considerations.
Outliers based on limit values
A value "x" is above or below a "limit" and is therefore defined as an outlier. Various options are available in the "Preparation" section of the evaluation strategy workflow.
These are explained in the evaluation strategy and are therefore only briefly mentioned here. Plausibility limits and scrap limits must be entered manually in the characteristics mask. Natural boundaries cannot be technically exceeded. The "X% tolerance" option should be used with caution. While measured values of 500% out of tolerance might be considered as an outlier in classical industry, it can be considered as a valid measured value in the electrical industry. However, all this information depends to a greater or lesser extent on what is entered in the characteristics mask. In other words, a conscious action.
Link to: Evaluation Strategy
Mathematical outliers
How to exclude outliers manually
The term "selection" is used here instead of "outliers". This refers to the explanation of "removing" values from the value chart using the "Select" function. Link to: Working with the Graphics - Select Functions
The use of select functions should also be reconsidered, with a concept of authorisation as to who should have the right to do this.
User rights for manual removal of statistical outliers
The outliers can be removed statistically using the test procedure, for example with the Hampel test. Link to: Test Procedures
The use of test procedure should also be controlled by setting user permissions.
