The digital transformation leads to technical systems with enhanced functionality. Modern technical systems are equipped with sensor networks, which augment the already available operating data and enable the monitoring of the overall system. Condition monitoring often comprises diagnosis of the current condition and the prediction of future conditions of the technical system or the product quality in a production line. Condition Monitoring methods rely on models, either developed by engineers or trained via machine learning aproaches, to estimate health states of technical systems or predict their remaining useful lifetime. Sometimes a combination of both approaches, i.e. a hybrid solution is utilized to achieve this aim. Based on this, the maintenance strategy predictive maintenance enables a ressource-saving and cost-efficent operation of the monitored system.
A current research field is the development of robust prognostics approaches for systems operating under non-stationary conditions for which future system conditions is challenging due to the increased uncertainty. To harness the potential of a condition monitoring process to a reasonable effort for maintenance engineers, different approaches are been developed to automate these methods. Furthermore, approaches to explain data-driven black-box models are been developed, to increase the explainability and the acceptance of these methods in the industry. In this context, the combination of a priori engineering knowledge and data-driven algorithms to a hybrid approach represents another research field.