Abstract 

 

In recent years, virtual testing has played an increasingly important role in the design and evaluation of engineered products. However, it is challenging to build the highly accurate computational models for virtual testing. Blind and recognized uncertainties are often unintentionally incorporated. These uncertainties consequently decrease the predictive capability of the models. To this end, this paper proposes a systematic approach for model refinement that minimizes the impact of unrecognized blind and recognized epistemic uncertainties in computational modeling. The approach consists of three steps: model invalidity analysis (MIA), development of an invalidity reasoning tree (IRT), and invalidity sensitivity analysis (ISA). First, in the MIA, possible causes that lead to discrepancies between the experimental and simulation responses are identified through brainstorming. Next, the IRT is built using the affinity diagram. It sequentially lists and screens potential candidate issues for model refinement at the stages of conceptual, mathematical, and computational modeling. Finally, the ISA quantifies the effect of incorporating updates in the model to address potential candidate issues with the goal of reducing the impact of the blind and recognized uncertainties. The most critical candidates are determined by using a weighted decision matrix. To demonstrate the effectiveness of the proposed approach, a case study examining a smartphone liquid crystal display fracture is presented.