Rule mining in maintenance: Analysing large knowledge bases

Association rule mining is a very powerful tool for extracting knowledge from records contained in industrial databases. A difficulty is that the mining process may result in a huge set of rules that may be difficult to analyse. This problem is often addressed by an a priori filtering of the candidate rules, that does not allow the user to have access to all the potentially interesting knowledge. Another popular solution is visual mining, where visualization techniques allow to browse through the rules. We suggest in this article a different approach: generating a large number of rules as a first step, then drill-down the produced rule base using alternatively semantic analysis (based on a priori knowledge) and objective analysis (based on numerical characteristics of the rules). It will be shown on real industrial examples in the maintenance domain that UML Class Diagrams may provide an efficient support for subjective analysis, the practical management of the rules (display, sorting and filtering) being insured by a classical Spreadsheet.