How Precise Has Fault Detection to Be? Answers from an Economical Point of View

Abstract

Condition monitoring as part of an effective condition based maintenance has long been accepted in the academic and industrial world, 25 COMADEM conferences are proof of that. Various methods, algorithms, and tools have been developed to monitor the condition, detect faults, or predict the remaining useful life. Furthermore, there are methods available supporting the engineer to choose the appropriate approach to his problem at hand. One question that often comes along with the implementation of fault detection and failure prediction is: How accurate has the condition monitoring to be and what is the right balance between false alerts and undetected faults which may lead to failures? This paper presents a method to calculate the required accuracy in terms of an economical efficient condition monitoring. It gives the parameters necessary to calculate the right balance between undetected failures and false alerts. The diagnosis of a railway point machine is used to illustrate the application of the proposed method. Points are critical to the railway operation and their breakdown has a high impact on delays and hence costs. Moreover, the structure of the railway network makes them a distributed system, sometimes hard to reach for maintenance staff, which also makes the maintenance expensive. This and other facts make the decision complex, but give a good example on how the question for the required accuracy can be answered taking into account the goal of a maximum of cost efficiency.