Assessment Concept for Transfer Learning Methods in Context of Lifetime Prognosis of Systems

Transfer Learning uses information from comparative problems for modeling and is therefore a promising method for lifetime prognosis under a lack of data. A literature review is conducted to provide an overview of existing Transfer Learning methods and their requirements and properties. On this basis, a procedure is developed with which the suitability of collected Transfer Learning methods can be assessed stepwise, considering the requirements and the available data for the implementation of a lifetime prognosis. The developed procedure takes particular account of different approaches of lifetime modeling and the similarity evaluation of systems.