A comparative study on methods of reduction and selection of information in technical diagnostics

The problem of knowledge acquisition is one of the most important problems connected with applications of expert systems in the domain of machinery diagnostics. The knowledge acquisition from databases by means of machine learning methods is one of the methods that could be applied to solve the problem. In most cases, collected values of attributes (diagnostic symptoms) are real numbers. The usage of machine learning methods usually requires a conversion of quantitative values to qualitative ones - this requires an estimation of cutting points. Limitation of a number of attributes used in knowledge acquisition process is also crucial. In the paper, a methodological process of estimation of cutting points as well as attributes selection for a machine state assessment is presented. The concept of an indirect assessment of the considered methods is also presented. Then an example of verification is shown.