Model-based Prognostics for Prediction of Remaining Useful Life

Abstract: Prognostics and healthmanagement (PHM) is an engineering discipline that aims to maintain the systembehaviour and function, and assure the mission success, safety andeffectiveness. Health management using a proper condition-based maintenance (CBM)deployment is a worldwide accepted technique and has grown very popular in manyindustries over the past decades. These techniques are relevant in environmentswhere the prediction of a failure and the prevention and mitigation of itsconsequences increase the profit and safety of the facilities concerned.Prognosis is the most critical part of this process and is nowadays recognizedas a key feature in maintenance strategies, since estimation of the remaininguseful life (RUL) is essential.PHM can provide a stateassessment of the future health of systems or components, e.g. when a degradedstate has been found. Using this technology, one can estimate how long it willtake before the equipment will reach a failure threshold, in future operatingconditions and future environmental conditions. This thesis focuses especiallyon physics-based prognostic approaches, which depend on a fundamentalunderstanding of the physical system in order to develop condition monitoringtechniques and to predict the RUL.The overall research objective of thework performed for this thesis has been to improve the accuracy and precisionof RUL predictions. The research hypothesis is that fusing the output of morethan one method will improve the accuracy and precision of the RUL estimation,by developing a new approach to prognostics that combines different remaininglife estimators and physics-based and data-driven methods. There are two waysof acquiring data for data-driven models, namely measurements of real systemsand syntactic data generation from simulations. The thesis deals with two casestudies, the first of which concerns the generation of synthetic data andindirect measurement of dynamic bearing loads and was performed atBillerudKorsäs paper mill at Karlsborg in Sweden. In this study the behaviourof a roller in a paper machine was analysed using the finite element method(FEM). The FEM model is a step towards the possibility of generating syntheticdata on different failure modes, and the possibility of estimating crucialparameters like dynamic bearing forces by combining real vibration measurementswith the FEM model. The second case study deals with the development ofprognostic methods for battery discharge estimation for Mars-based rovers. Herephysical models and measurement data were used in the prognostic development insuch a way that the degradation behaviour of the battery could be modelled andsimulated in order to predict the life-length. A particle filter turned out tobe the method of choice in performing the state assessment and predicting thefuture degradation. The method was then applied to a case study of batteriesthat provide power to the rover.