GH PLUG-IN: INTRODUCING BI-DIRECTIONAL INFERENCE FOR PARAMETRIC SIMULATION ANALYSES

Year: 2018-2019

Role: Doctoral thesis work​

 

Description: 

The following introduces an explanatory approach to engineering simulation in the parametric design environment. 

 

Most simulation tools used in building design produce an exact output, which can be difficult to interpret at times, especially if engineering domain knowledge is limited, and/or if considering many design parameters. Therefore, instead of utilising simulation tools in a one-directional, input to output fashion, our approach builds a probabilistic representation of the simulation inputs and outputs using Bayesian networks that enables bi-directional prediction between the two. 

 

For example, we can use the probabilistic representation to quickly identify the parameter ranges required to produce target simulation outputs of interest (as shown in video). Furthermore, the shapes of the predicted input distributions provide insight into the sensitivity of the parameters in effecting the responses. The approach is not limited to a single output; one can set multiple output targets of interest, and observe the causal input distributions.