Data Driven Acoustic Design

Abstract

Acoustics are rarely included as a design driver in the early phases of design due to the multi-faceted nature of sound and the complex and time-consuming analysis process of room acoustics software. Inevitably this results in architectural spaces with poor acoustics, where treatment is either disregarded or focuses only on noise prevention using absorbent materials. However, most commonly used construction materials have sound-reflecting properties and can be configured into sound-diffusive surfaces. These surfaces can help reduce unwanted flattered echoes, colourisation, and image shift and create a more pleasant and comfortable environment without needing additional elements (e.g. absorption panels). Faster and simpler analysis tools are required to harness the potential of diffusion in architectural design. This dissertation presents a new data-driven approach to designing and evaluating the acoustic properties of architectural surfaces. It investigates the use of machine-learning techniques to study the mutual relationship between geometry and sound diffusion. It introduces a new acoustic dataset meant as a basis for training predictive machine-learning models. These models enable the creation of fast, less cumbersome, and reasonably accurate acoustics analysis tools. It proposes and implements a new automated multi-robotic data-acquisition method for collecting impulse responses from scale-modelled surfaces. It also develops computational tools to design and generate three-dimensional wall-like surface geometries. The geometrical characteristics of these surfaces are based on commonly used construction materials and techniques. A computational framework is developed in parallel to process the collected data and generate customisable and interactive visualisations for low- and high-dimensional data. This framework caters to both expert and non-expert users in acoustics, providing expert users with familiar descriptors and visualisations and introducing non-experts to simpler ones. Furthermore, to address users with no programming knowledge, it develops a web-based application enabling easy access to the collected dataset, the acoustic descriptors, and visualisations. It introduces a new workflow to the performance-driven acoustic design of sound-diffusing wall surfaces, allowing architects and designers to explore alternative wall designs with sound-diffusing properties, given a set of desired acoustic performance criteria. The proposed workflow has the potential to bring acoustics closer to the early phases of architectural design and enable a more integrative acoustic and architectural design exploration. Providing architects and acousticians with comprehensive and user-friendly tools for acoustics analysis can help integrate acoustics into the design process from the beginning rather than as an afterthought.