ESTIMATING AND OPTIMIZING THE EFFICACY OF PREDICTIVE SYSTEMS IN PROACTIVE MAINTENANCE

This article presents a statistical methodology that may be used in qu antifying and optimizing the efficacy of predictive systems used in pr oactive maintenance. The methodology was developed with regard to a pa rticular tool employed by Bellcore's clients, but is believed to have application to the use of other similar predictive systems. As there a re a large number of factors driving trouble reports and typical data sets tend to be highly noisy, until recently there were no clear stati stics about how much predictive systems reduce trouble reports. Moreov er, there were no clear guidelines about how these systems should be u sed in an optimum manner. The quantification of the performance of suc h systems requires careful modeling of the process governing the troub le reports. In this article, by using Generalized Linear Models (GLM), we studied the effect of these tools on Code 4 trouble reports. We al so discuss the use of experimental designs in this context to determin e the optimum use of the tool under different environmental conditions .