Weibull parameters estimation from uncensored test data contaminated with outliers

摘要

Experimental results relevant to dielectric strength tests on electrical components insulated with solid materials can be affected by outlying observations, i.e. values that stand apart from the bulk of the data. In such cases, the application of consolidated methods for the Weibull parameters estimation can sometime induce to erroneous assessment. Contaminated samples generated through the Monte Carlo method have been submitted to different Weibull parameters estimators with and without the application of data pre-processing techniques. Results have been compared with the aim to verify whether such pre-processing techniques can improve the performance of the estimation methods in presence of outlying observations.