A Statistical Approach to Causality Analysis in a Distributed Design Framework

Numerous collaborative design tools have been developed to accelerate the product development, and recently environments for building distributed simulations have been proposed. For example, a simulation framework called DOME (Distributed Object-oriented Modeling and Evaluation) has been developed in MIT CADLAB. DOME is unique in its decentralized structure that allows heterogeneous simulations to be stitched together while allowing proprietary information an simulation models to remain secure with each participant. While such an approach offers many advantages, it also hides causality and sensitivity information, making it difficult for designers to understand problem structure and verify solutions. The purpose of this research is to analyze the relationships between design parameters (causality) and the strength of the relationships (sensitivity) in decentralized web-based design simulation. Algorithms and implementations for the causality and sensitivity analysis are introduced. Causality is determined using Granger’s definition of causality, which is to distinguish causation from association using conditional variance of the suspected output variable. Sensitivity is estimated by linear regression analysis and a perturbation method, which transfers the problem into a frequency domain by generating periodic perturbations. Varying Internet latency and disturbances are problematic issues with these methods. Thus, new algorithms are developed and tested to overcome these problems.