Structural Health Monitoring (SHM) refers to a paradigm enabling maintenance activities to be scheduled
based on the forecast of the system degradation. This forecast is primarily derived from the analysis of sensor
data. This dissertation presents two novel contributions, which enhance the precision of SHM, through the
combination of physics-based simulations and data-driven models.
The first contribution is a robust approach, that finds an optimal configuration of heterogeneous sensors, which maximizes the damage estimation confidence. This approach’s novelty lies in three aspects. First, the mathematical definition of the damage estimation process using a Kalman filter is modularly incorporated in the optimization problem’s objective function. Second, benefiting from such a definition, the Jacobian of the objective function with respect to the design variables is derived, permitting the usage of a gradient-based method. Third, within the proposed approach, a systematic algorithm for a-priori identifying the optimal number of sensors is derived. The combination of the three aspects in one approach make it customizable according to the subsequent estimation requirements and makes it applicable to complex industrial structures.
The approach’s sensitivity to the number of sensors, their types, and the constraint enforcement approaches is rigorously investigated on two structures with ascending physical complexity. The proposed approach precisely estimated the accumulated damage, and consistently surpassed existing methods in literature. The robustness of the approach to complex cases is evaluated by applying it to two industrial structures under realistic operating conditions.
The second contribution of this thesis is a hybrid model for fatigue damage estimation in fleets of engineering systems. Mainly two novelties are presented in this model. The first novelty is the unprecedented utilization of physical degradation models for fleet estimation. This yields an interpretable damage estimation model, in comparison to a conventional purely-data-driven model. The second novelty is the robust transferability of physics-based degradation models of one engineering system to other unidentical systems. This allows more accurate damage estimation, even where only limited physics-based models and operation data are available.
The hybrid model utilizes the availability of scarce, yet accurate, physics-based degradation models, and combines them to approximate the degradation behaviour of other homogeneous structures in a fleet. The combination is performed by a data-driven weighted-mean filter. Based on the extent of similarity between fleet structures, and the current operating conditions, a data-driven algorithm assigns the suitable weights to each physics-based model.
The model is tested on a fleet of beams, which included a deviant range of damage severities. To assess its robustness, the constituents and parameters of the hybrid model are vigorously varied between nominal and extreme cases. To further evaluate the applicability in an industrial environment, the model is evaluated on a realistic fleet of servomotors. On both use-cases, with only a few physics-based models, and very limited operation data, the hybrid model quantified accurately and precisely the fatigue damage accumulation in other structures of the fleet.
«
Structural Health Monitoring (SHM) refers to a paradigm enabling maintenance activities to be scheduled
based on the forecast of the system degradation. This forecast is primarily derived from the analysis of sensor
data. This dissertation presents two novel contributions, which enhance the precision of SHM, through the
combination of physics-based simulations and data-driven models.
The first contribution is a robust approach, that finds an optimal configuration of heterogeneous sensors,...
»