Bubela I. Processing of measurement results by deviation of their statistics properties from typical

The thesis is devoted to solving of important applied scientific task of is creation and research of new procedure for the determination of the uncertainty of measurement results when is extreme (minimal or maximal) value of observations as well as simple and effective procedures of the processing the results of observations, with other than normal distributions with the aim of determination of the best (with the smallest standard uncertainty) evaluations of the measurement results, that will improved of quality of production of goods in accordance to the predictable necessities of consumers. The existing methods of processing uncertainty the of observations results are analyzed, in particular the order statistics method and also processing problems are set when results of the control measurements are extreme observations and their uncertainty must be determined. The non-standard statistical method for evaluating uncertainty, when a minimal (extreme) observation is a result the measurement (or testing) and a number of observations is significantly limited: n = 3…10 and if the observations probability density function (PDF) differs from a normal distribution is proposed and analyzed. This method is based on some properties of order statistics. As example from the practice this method is used to evaluate the uncertainty of a percent elongation and tensile strength in testing plastic products. The accuracy of obtained theoretical and experimental results is also investigated by Monte Carlo simulation. The modified method based on order statistics for processing of the random observations with so called Flatten - Gaussian distribution that is a convolution of rectangular and normal distributions is proposed. Proposed method in practical realization is simple, fast and enough accurate. Research results, obtained by Monte Carlo method, have confirmed the effectiveness of proposed method.