Search Results

Current alert setting methodologies based on setting defect detection alert thresholds for vibration and other Health & Usage Monitoring Systems (HUMS) indicators have many limitations, principally there is necessarily a compromise to be achieved between the true negative and false positive diagnostic metrics. This is true for all alert detection techniques from fixed thresholds through to Support Vector Machines. In this paper we describe techniques, validated using helicopter HUMS data, which do not invoke this compromise and independently minimise both the true negative and false positive rates. This paper will also demonstrate how the alert processing can be made more robust and overcome the problems introduced by HUMS data being both non-stationary, non-ergodic i.e. characteristics that change both with time and from platform to platform.