P
A COMPUTATIONAL DESIGN FRAMEWORK OF AN ADAPTIVE SOUND BARRIER SYSTEM FOR NOISE REDUCTION IN URBAN CONTEXT
Master of Science in Sustainable Design Thesis
Carnegie Mellon University
Advisors:
Dana Cupkova
Dina El-zafanly
Louis Suarez
40°26'36.5"N 79°56'43.6"W
Preliminary simulations indicate that the final product of the adaptive sound barriers can significantly mitigate urban noise pollution, demonstrating variability and potential for broad application in diverse urban environments. Future research will aim to optimize system designs, explore materiality and scalability, and integrate these adaptive systems with urban infrastructure to enhance their long-term impact.
Master Thesis, Internet of Things, Machine Learning, Sonification
Acoustics Simulation, Human-computer Interaction, Urban Ecology
Abstract
In an era of rapid technological and economic development, urbanization has exacerbated noise pollution, posing collective human comfort and well-being challenges. Current urban sound barriers are often static and limited in their effectiveness across different urban settings and noise levels. There is a lack of integration between sound management and adaptive technologies that can respond in real-time to changing urban soundscapes. This thesis delves into developing an adaptive framework for urban sound barriers to reduce noise pollution. By doing so, it seeks to mitigate the physical and mental health effects of excessive noise while fostering positive social impacts by establishing sound visualization “boundaries” within urban spatial environments.
The experiment outlined in this thesis adopts a workflow that incorporates live sound collection, data processing, machine learning, real-time sound visualization, acoustics simulation with analysis, and prototyping. The primary objective is to develop a framework for adaptive sound barrier systems tailored to different decibels and sound frequency levels. These systems are assessed for their effectiveness in noise reduction through simulated evaluation. Key methodologies include field acoustics simulation, as well as the integration of sound visualization, sonification, the Internet of Things (IoT), tiny machine learning, and design systemization.
Preliminary acoustics simulations indicated that adaptive sound barriers can effectively reduce urban noise pollution. The systems demonstrate variability in response to different noise levels and frequencies, showcasing the potential for broader application in diverse urban settings. The findings suggest that adaptive sound barriers could revolutionize urban planning and design, offering a dynamic solution to the persistent problem of noise pollution. Further research will be needed to optimize system designs and evaluate long-term impacts on urban environments. Future studies will explore materiality, scalability, and integration with urban infrastructure.
Hypothesis
Drawing inspiration from studies about bio-digital architecture,
people envision a future where sustainable architecture evolves as a living and adaptive system
-- an adaptive sound barrier system may significantly reduce noise pollution
and mitigate physical health, mental health and sociological impact
by creating a sense of “boundary” with sonification
in urban settings.
and mitigate physical health, mental health and sociological impact
by creating a sense of “boundary” with sonification
in urban settings.
Literature Diagram
Methodology Diagram
Process
Data Collection + Processing + Machine Learning Model
Data collection was facilitated using an Arduino Nano BLE 33 module within the selected building. Four sets of data corresponding to the four predefined sound intensity levels were collected using the Edge Impulse platform, labeled as Level 1, Level 2, Level 3, and Level 4. Each set comprised twenty samples, capturing approximately 10 seconds of sound data. This systematic approach ensured a comprehensive and representative dataset for subsequent analysis and processing.
The performance of the machine learning model on the validation set is quantitatively summarized in the confusion matrix, reflecting an overall accuracy of 92.3% with a loss of 0.36. This high level of accuracy demonstrates the model's efficacy in classifying urban noise into the four predefined categories based on sound intensity levels.
Realtime Sonification
The Arduino module, equipped with the machine learning model, continuously monitored environmental sounds. Utilizing its onboard sensors and processing capabilities, the module classified the noise levels in real time, identifying them as one of the four predefined categories based on their intensity.
Sonification Results
Concurrently, the classified data were transmitted via serial communication to the TouchDesigner program. This program was designed to analyze the acoustic properties of the intensity and frequency levels while generating real-time sonification that responds to the classification data.
Image Processing
Upon successfully implementing and validating the real-time sonification, the next stage in the research involved advanced image processing to further generate design.
Two ditinct image processing options were explored:
Design Generation (2D to 3D)
After the initial image processing steps were completed, the next progression in the experimental methodology was to convert the two sets of processed images into 3D renderings using Rhino and Keyshot software. In both design options, one group of standard 3D displacements and a group of color-inverted displacements are generated.
Prototyping
With the transformation of 2D processed images into 3D renderings, this step is to make physical prototypes of the 3D renderings. The objective is to materialize the virtual 3D models, providing a tangible means to assess and interact with the data-driven designs. To achieve this, two distinct 3D printing techniques were employed to prototype the eight patterns generated from the first option of image processing:
01 Binder Jetting(Sand Printing)
This technique involves layering sand and selectively depositing a liquid binding agent in the cross-sections of the pattern. The choice of sand printing was driven by its ability to accurately render intricate details and subtle topographical variations of the displacement maps.
02 Powder Printing
Acoustic Simulation + Analysis
The final phase of the experiment involved conducting acoustic simulations to assess the noise reduction capabilities of the generated patterns. Utilizing Rhino software integrated with Pachyderm Acoustics and Grasshopper, the simulation process began with creating a detailed environmental model of the selected site—a residential building on 5th Avenue in Pittsburgh.
Simulation Workflow
As shown in the workflow diagram, a sound source and a receiver were placed within the scene to facilitate acoustic analysis. To maintain consistency across the simulations, plaster was chosen as the material for all building elements. This decision ensured that any differences in acoustic performance could be attributed to the facade designs rather than material variations.
The simulation process involved three critical steps:
1. Direct Sound Data Collection:
Initial data on sound propagation directly from the source to the receiver without interacting with surfaces.
2. Image Source Data Gathering:
Data collected from sound reflections that interact with the virtual environment provides insights into the secondary paths
of sound travel.
3. Ray Tracing Data Compilation:
This offered an analysis of the sound paths through the environment, including multiple reflections.
Combining these data points, an Energy-Time Curve was generated, which served as the basis for computing the reverberation time, which is a key metric in assessing the acoustic performance of a space. The reverberation time, measured in seconds, indicates how long it takes for sound to decay to inaudibility after the source has stopped emitting. The calculation result is eight reverberation times of the frequency scope. In this case, the third and fourth levels are focused since common traffic noise is within the range of 500Hz -1000Hz.
Simulation Results
The results of the acoustic simulations are presented in a table(Table 7.2) detailing the reverberation times (in seconds) for the baseline model and the eight facade designs at various frequencies. Given the focus on common traffic noise, which predominantly ranges between 500 Hz and 1000 Hz, this analysis primarily considers the reverberation times at these frequencies.
To better visualize the data, the reverberation times for the eight generated designs were compared with the baseline and converted into percentages. As shown in the table, increases are highlighted in red, while decreases are marked in green.
The comprehensive analysis of the reverberation times across various design patterns reveals insightful trends and outcomes. The experimental results highlight that while some designs increased the reverberation times, others showed significant reductions, enhancing the acoustic properties of the environment. Notably, the design S3_Invert outperformed all others, demonstrating the greatest decrease in reverberation time, which indicates its potential for effective noise reduction in real-world applications. This variability in performance underscores the importance of adaptive design in architectural acoustics to achieve desired outcomes.
Conclusions
The experiment conducted as part of this thesis confirms the effectiveness of the proposed adaptive sound barrier system in mitigating urban noise pollution. The results validate the initial hypothesis and highlight the practical applicability of the designs in real-world environments.
This research aligns with my long-standing conviction that modern sustainability efforts should focus on using technology to improve the symbiotic relationship between humans and living environments. The adaptive design showcased in this thesis exemplifies this principle by demonstrating how even incremental technological enhancements can have significant long- term impacts.
Looking forward, the next step is to enhance the system's adaptive capabilities, enabling it to respond more dynamically to immediate environmental changes. This advancement will pave the way for the development of smarter, more responsive urban infrastructure, furthering the integration of sustainability in everyday living spaces.
DOI
(In Progress...)