Identification of dynamic diagnostic models with the use of methodology of knowledge discovery in databases

This paper deals with an approach to knowledge discovery in databases applied in order to identify a dynamic model of a real-existing machine. The problem considered within the paper is how to identify dynamic models suitable for model-based diagnosing of a physical object. A special attention is paid to identification on unsupervised way, while big databases collected by a SCADA system is handled.In the paper a method of identification of dynamic models of objects and processes is presented. The usefulness of the method in technical diagnostics are shown. The elaborated method of analysis of quantitative dynamic data is based on applications of accessible methods of knowledge discovery in databases. The essence of the method is to project values of considered set of attributes into the so-called multidimensional space of regressors. In order to select the subset of relevant features the genetic algorithm was used. Knowledge was induced using the support vector machines (SVM) method. The AIC measure as well as our own heuristic function were applied as evaluation criteria. The method was applied in a process of discovery of a model of changes of temperature of a pump. Within framework of the research, data gathered by means of an industrial system registering data on a peculiar object, which was deep-well pumping station, was analyzed.