Method development and evaluation scheme for cross-life linkage of structural health monitoring data and exiting knowledge via deep transfer learning

In the project "Method development and evaluation scheme for cross-life linkage of structural health monitoring data and exiting knowledge via deep transfer learning", the foundations for a generally valid approach to structural health monitoring (SHM) are being methodically investigated to take into account the degradation of sensors and systems, as well as the only insufficiently identifiable boundary conditions of unique structures through artificial intelligence. This prepares the way for the life-time accompanying linking of condition information and measurement data with a "digital twin" to a predictive maintenance, see Figure 1.

Basic investigations are carried out on AI models learned by means of Deep Transfer Learning. Deep Transfer Learning reuses existing "knowledge" from simulations and real measurements in the form of general AI models to create a suitable AI model with high accuracy using a much-reduced inventory of training data from the individual structure and less training effort. By re-training, new classes, of features not previously known and considered, can be added during the life of the structure to account for sensor effects and structure responses for the particular SHM. The project investigates how perturbations and inaccuracies in the training data affect the knowledge transfer from initial model to sensor model or building model. The finite element method is used to generate general knowledge about a structure.

In the subsequent deep transfer learning process, real world monitoring data are used, which are obtained, from the demonstrator structure, the Nibelungen Bridge crossing the river Rhine near the city Worms, see Figure 2.