To Calculate with Averages : What is a Good Result in the Absence of the Correct?
Niklas Molin · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2008
This thesis specifies the foundation for a computer system that operates on averages of measured values provided by automatic measurements systems. This consists of specifying a data object for the average of measured values and fusion methods for that object. The data object should, apart from the value, contain a time and quality description for the average. The fusion methods considered are elementary arithmetic operations to combine different objects and aggregates to yield summarized information on homogenous objects. The quality of the data objects are described by three attributes; uncertainty, confidence and completeness. The study provides propagation methods for all quality attributes but focuses on the uncertainty of the value. Three different methods for uncertainty propagation are tested. The first method is based on the framework of the corresponding ISO-standard. The others are standard Monte Carlo and quasi-Monte Carlo simulations. The findings can be used directly as material for an implementation of a system that operates on averages or as material for further discussion. Some of the issues concerning such a system remain unresolved, while others might be optimized.