Application of data mining in a maintenance system for failure prediction

In industrial environment, data generated during equipment maintenance and monitoring activities has become increasingly overwhelming. Data mining presents an opportunity to increase significantly the rate at which the volume of data can be turned into useful information. This paper presents an architecture designed to gather data generated in industrial units on their maintenance activities, and to forecast future failures based on data analysis. Rapid Miner is used to apply different data mining prediction algorithms to maintenance data and compare their accuracy in the discovery of patterns and predictions. The tool is integrated with an online system which collects data using automatic agents and presents all the results to the maintenance teams. The purpose of the prediction algorithms is to forecast future values based on present records, in order to estimate the possibility of a machine breakdown and therefore to support maintenance teams in planning appropriate maintenance interventions. increasing productivity. Manufacturing data collected in real-time contains valuable information and knowledge that could be integrated within prediction systems to improve decision making and enhance productivity (Elovici & Braha 2003). The main objective of the on-going project is to develop a recording and prediction system to foresee machine failures in manufacturing units, globally dispersed. A crucial and core component of the functionality of this system lies in the ability to collect and interpret dispersed data, creating predictive models allowing maintenance teams to act before problems occur. In this paper a Rapid Miner prototype that complies with the conceptual specifications previously defined for the interpretative module on the global maintenance system, is presented. Into the system will flow real-time data gathered by intelligent agents from different machines in world scattered production facilities, allowing after data mining activities to produce warnings guiding future maintenance actions. Such a system could improve overall maintenance operations, reducing breakdowns and maintenance costs, assisting managers for a better planning of maintenance activities. An empirical study was conducted by using real data from production