Real-Time Simulations Using Learning Algorithms for Immersive Data Visualization in Buildings

Abstract

Computational Fluid Dynamic (CFD) simulations are used to predict indoor thermal environments and assess their response to specific internal/external conditions. Although computing power has increased exponentially in the past decade, CFD simulations are still time-consuming and their prediction results cannot be used for real-time immersive visualization in buildings. A method that can bypass the time-consuming simulations and generate “acceptable” results will allow such visualization to be constructed. This paper discusses a project that utilizes a supervised Artificial Neural Network (ANN) as a learning algorithm to predict post-processed CFD data to ensure rapid data visualization. To develop a generic learning model for a wide range of spatial configurations, this paper presents a pilot project that utilizes an unsupervised Reinforcement Learning (RL) algorithm. The ANN technique was integrated with an interactive, immersive Augmented Reality (AR) system to interact with and visualize CFD results in buildings. ANN was also evaluated against a linear regression model. Both models were tested and validated with datasets to determine their degree of accuracy. Initial tests, conducted to evaluate the user's experience of the system, indicated satisfactory results.