| dc.description.abstract | Image-based simulation plays a pivotal role in diverse engineering applications, integrating both image and numerical variables as inputs to predict design performance, understand system behaviors, and drive discovery. Uncertainty, inherent in these simulations, must be quantified and managed as it arises in numerical variables due to randomness in materials, manufacturing processes, and operations. Similarly, images exhibit uncertainty stemming from the inherent variability of the quantities they represent and the involved image processing. Addressing image uncertainty presents a unique challenge, primarily due to the high dimension and the limited availability of image samples, imposing constraints on conventional uncertainty quantification (UQ) techniques. To overcome this challenge, this study introduces a new concept—uncertainty separation, designed to disentangle the impacts of uncertainties associated with image and numerical inputs, particularly in scenarios with limited image samples. The proposed method decomposes a simulation model into two distinct submodels: one handling image inputs and the other managing numerical inputs. While image samples directly inform the analysis of the image submodel, existing uncertainty quantification approaches are applied to assess the submodels with numerical input. This concept has proven to be efficient, achieving satisfactory accuracy through two practical examples, demonstrating its potential to enhance engineering analysis and design in scenarios involving image and numerical uncertainties. | |
