Digital Twins: Neural-Networks for the implementation of digital twins of gearboxes

Digital Twins (DTs) are widely used for the design and prognostic analysis of mechanical devices. For the implementation of optimized, effective DTs of gearboxes, engineers often lean on the ISO Standards and Codes. However, the use of Standards could be not trivial to setup, slow and with license problems for its portability. Neural Networks (NNs), also implemented on open-source software, has been proved to be able to create links between input and output quantities without the need of knowing the underlying laws if trained properly. Here we trained a NN with a huge dataset created by using Standards for a simple gearbox. By comparing with Standards, we found that the accuracy of a NN depends on the safety factor, the physical characteristics of the gearbox and correct setup of the NN. This result is of paramount interest since it reveals that NNs can be used for the implementation of accurate digital twins when a proper training on a wide dataset is carried out.