Using the Scikit-Learn Library and Our Own Clustering Program to Process Statistical Data on Failures of Double-Circuit Gas Boilers

Abstract

At the present stage of technology development, a large number of industrial enterprises and service organizations often have the problem of processing arrays of statistical data on emergency situations and equipment failures. The increase of information quantity and its complexity, the multifactorial nature of the occurred failures and the risks, which are associated with it, do not allow us to use primitive mechanisms of statistical analysis and accounting. Clustering of statistical information is one of the ways to stratify incoming data, identify certain patterns in equipment failures and establish the weight of influencing factors. For small and medium-sized enterprises involved in the maintenance and repair process, this is especially important, because along with identifying the causes of malfunctions and failures, the application of clustering to an array of statistical data provides information for strategic planning - the volume of repair work, the volume of materials and components procurement, time management, and in general, prognostics. Processing of an array of data requires the involvement of appropriate software. For this purpose, both freely available packages (for example, the scikit-learn library with the Jupyter Notebook interactive computing environment used in the work) and proprietary software products designed for investigation of specific problem can be used. In the study, we conducted sequential processing and normalization of the initial data on boilers equipment failures for cluster analysis. Then the cluster analysis itself is performed and its results are compared, obtained on the basis of the scikit-learn library with Jupyter Notebook public interactive computing environment and the data received from clustering program of our own design.