High fidelity mechanics modeling has resulted in the ability to design robust structures without having to go through extensive and expensive testing. However, a common problem is the use of existing structures to new demands that they were not tested for. Our goal in this project is to quantify the uncertainty and confidence in system model predictions, by linking models of the system and its components that exist in multiple scales and multiple disciplines. This multi-scale multi-physics uncertainty quantification will use a Bayes network approach, allowing the updating of the system parameters as more observed data becomes available. The Bayes network will be used to investigate the sensitivity of the final solution uncertainty to each of the sub-models, their parameters and inputs. This would allow decision making for model fidelity, i.e., allocating resources to more extensive investigation where the errors affecting the solution uncertainty are the most significant.
The above methodology is being developed along the following four steps:
1. Quantification of uncertainty in model prediction from multiple sources, i.e., natural variability, data uncertainty, and model errors
2. Integration of multiple error and uncertainty sources and their propagation using a Bayes Network
3. Extension of objectives 1 and 2 to multi-scale/multi-fidelity and multi-physics problems
4. Optimization of analysis fidelity in various disciplines
APPLICATION EXAMPLE
The proposed methodology is being applied to a part of the blended fuselage and wing structure in hypersonic aircraft. It consists of a stiffened panel subjected to acoustic, thermal, aerodynamic and structural loads. The work systematically includes the sources of variability, data uncertainty, and modeling errors in the response prediction (vibration, deformation) as well as life prediction (crack propagation) of the panel, while taking into consideration the coupling between the different disciplines in question.
Fig.1:Experimental temperature distribution on panel