Fleet knowledge for prognostics and health management-identifying fleet dimensions and characteristics for the categorization of fleets

Abstract

Current prognostics and health management approaches areoften not able to meet expectations due to their limited abilityto accurately detect abnormal machine conditions, identifyfailures and estimate the remaining useful life. This is inmany cases attributed to the lack of real data and knowledgeabout the component or machine under consideration.Instead, experimental data is often used for algorithmtraining, which is not able to reflect the complexity of realworldsystems. To improve prognostics and healthmanagement approaches condition data from fleets ofmachines rather than single units can be taken intoconsideration. Therefor machine conditions are assessedagainst situations encountered by machines in the same fleetand knowledge is transferred to allow algorithms tointelligently learn and improve their capabilities.Several approaches have recently been presented in theliterature, which make use of the fleet knowledge forcondition-based maintenance. These approaches are designedfor specific fleet compositions and characteristics.Therefore, in order to incorporate fleet knowledge intodiagnostic and prognostic approaches the fleet underconsideration and resulting requirements have to be analyzed.With this information, it is possible to determine whetherfleet-based approaches are applicable in general to thespecific case as well as facilitate the selection of a suitablefleet-based approach. Three types of fleets are distinguishedin the literature, namely identical, homogeneous andheterogeneous fleets. This distinction makes reference to thestructural dimension of fleets. For fleet-based approaches,however additional dimensions should be taken into account.These include among others the operating condition in thefleet (e.g. identical, different, or dynamically changing) and the type of available data (e.g. sensor reading, context data,textual description). This paper aims at identifying andanalyzing different dimensions and respective characteristicsof fleets to be considered in the context of prognostics andhealth management. The results are synthesized in aclassification structure to support the categorization of fleets.