Hidden semi-Markov Models for Predictive Maintenance of Rotating Elements

The dissertation at hand introduces a novel algorithm to predict the remaining useful life (RUL) of rotating components such as grooved ball bearings or gear boxes. The focus of the implemented method is on so-called Hidden semi-Markov Models (HsMM), which are suitable for the modeling of sequential events. The selected method is designed to support maintenance processes based on advisory generation in the context of predictive maintenance. In this regard, the goal of predictive maintenance is the generation of predictions about the RUL of examined components based on their current state. Thus, maintenance events can be scheduled more precisely, downtime is reduced, and the actual useful life of components can be exploited. After an initial classification of the proposed method in the context of Prognostics and Health Management (PHM), the concept is described. The algorithm is based on methods, which apply historical data of the component’s degradation process for the estimation of the current and future damage state. A novel concept in the field of HsMM permits the identification of similar damage states within different degradation datasets to obtain more information about the damage process of the examined components. A further research question analyzes, whether the consideration of available information about the component’s endured load increases the prognostic performance. First results in the context of verification conclude the concept description. Subsequently, the design and realization of a new test rig, which permits an accelerated bearing aging for induction machines due to a so-called bearing current, is presented. Here, an alternating current flows through the tested bearing and reduces its life cycle significantly. By means of these data, the concept is evaluated. In comparison to state-of-the-art methods in field of bearing PHM, the validation is executed by examining the motor current of the induction machine instead of the widespread analysis of the resulting vibration signal. The results indicate that the precise and accurate prediction of the bearing’s RUL is possible. In addition, the consideration of already endured load for the generation of life cycle predictions is beneficial. A selected state-of-the-art algorithm, also based on HsMM, permits a realistic evaluation of the achieved prognostic performance. The dissertation ends with the estimation of possible cost savings in an exemplary aircraft maintenance scenario. For this, the obtained prognostic results are assessed with a state-of-the-art cost-benefit analysis tool. The outcome indicates that the application of the proposed algorithm leads to savings due to e.g. decreased downtimes