Detection and avoidance of low probability phenomena using probabilistic graphical models in electromechanical actuators
Due to its interaction with the environment and internal changes due to wear, every machine is bounded to evolve over time. This evolution might entail several unwanted consequences such as production loss, unexpected outages, etc. Actuators constitute the driving force of every machine. They convert energy into a form that is required by the application. Electromechanical actuators, that convert electrical energy into motion, take up the lionU+2019s share of all actuators (>30%), and like other machine components, they are subject to changes over time. The State-of-Health (SoH) parameters that characterize this evolution can often not be directly measured such as the quality of the conductor insulation in electromechanical actuators. In case the evolution exhibits a low probability, in the sense that from all possible changes only a few will occur once over the entire lifetime of a specific actuator (e.g. short-circuit due to failed insulation), no methods exist that are able to estimate various SoH parameters from measurable quantities in order to detect changes.To fill this gap, this PhD project proposes to augment existing models, that use operational data to estimate SoH parameters, with a probabilistic model. This allows exploit historical and recent operational data to estimate various SoH parameters and extend the estimation capabilities by exploring unseen data. Additionally, the registered evolution of SoH parameters will be used to learn to avoid these changes.
- See also: undefined