Smart Wind Tunnel For Bio-Inspired Flying Robot Analysis: Advanced Studies in Hardware and Software
Korai Baloch, Fahad Ahmed (2025)
2025
Master's Programme in Biomedical Sciences and Engineering
Lääketieteen ja terveysteknologian tiedekunta - Faculty of Medicine and Health Technology
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2025-11-28
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2025112610960
https://urn.fi/URN:NBN:fi:tuni-2025112610960
Tiivistelmä
The field of bioinspired soft robotics has seen rapid growth in the development of lightweight, passive micro-fliers, such as those mimicking dandelion seeds. However, experimental validation of these sensitive, sub-gram structures remains a significant challenge, as conventional wind tunnels are designed for high-speed aerodynamics and are ill-suited for low wind velocity flight regimes. This thesis addresses this critical gap by detailing the design, fabrication, and validation of a programmable, wind-tunnel platform specifically tailored for the aerodynamic testing of bioinspired micro-fliers.
The platform's mechanical design features a 36-fan (6×6) array, with each fan individually regulated by Pulse-Width Modulation (PWM). To ensure high-quality, stable airflow, the system integrates custom 3D-printed ducts, dual stainless-steel mesh screens, and a 60 mm thick aluminium honeycomb flow straightener (L/D ≈ 9.4). The electronic subsystem is implemented using two custom-designed, four-layer ESP32-S3-based control PCBs. Each board integrates isolated 12 V and 3.3 V power domains, hardware-timed LEDC/MCPWM channels for 18 independent PWM fan outputs, USB-routed differential pairs, and dual environmental sensors (BME680/SHTC3). Operating in a Master–Slave configuration, the boards collectively provide synchronized actuation of all 36 fans and real-time environmental monitoring via a Wi-Fi–enabled dashboard.
Experimental validation, conducted with a hot-wire anemometer, confirmed the platform's good performance. The system generates a highly stable flow field with a turbulence intensity (TI) below 0.4% and an excellent spatial velocity non-uniformity of approximately 3.5% across the test section. Step-response tests also verified that the control system can achieve new velocity setpoints rapidly and without overshoot.
Furthermore, a proof-of-concept machine-learning pipeline, incorporating a deep-learning–based YOLOv8 detector and a Random Forest classifier, was implemented to explore automated analysis of flier behavior. The model achieved 90% overall accuracy in classifying flight states, demonstrating its viability as a tool for high-throughput data analysis.
This thesis delivers an integrated, validated, and accessible experimental platform. It successfully bridges the gap between materials-driven flier design and rigorous aerodynamic characterization, providing a standardized, reproducible environment to advance the study of bioinspired soft aerial robots.
The platform's mechanical design features a 36-fan (6×6) array, with each fan individually regulated by Pulse-Width Modulation (PWM). To ensure high-quality, stable airflow, the system integrates custom 3D-printed ducts, dual stainless-steel mesh screens, and a 60 mm thick aluminium honeycomb flow straightener (L/D ≈ 9.4). The electronic subsystem is implemented using two custom-designed, four-layer ESP32-S3-based control PCBs. Each board integrates isolated 12 V and 3.3 V power domains, hardware-timed LEDC/MCPWM channels for 18 independent PWM fan outputs, USB-routed differential pairs, and dual environmental sensors (BME680/SHTC3). Operating in a Master–Slave configuration, the boards collectively provide synchronized actuation of all 36 fans and real-time environmental monitoring via a Wi-Fi–enabled dashboard.
Experimental validation, conducted with a hot-wire anemometer, confirmed the platform's good performance. The system generates a highly stable flow field with a turbulence intensity (TI) below 0.4% and an excellent spatial velocity non-uniformity of approximately 3.5% across the test section. Step-response tests also verified that the control system can achieve new velocity setpoints rapidly and without overshoot.
Furthermore, a proof-of-concept machine-learning pipeline, incorporating a deep-learning–based YOLOv8 detector and a Random Forest classifier, was implemented to explore automated analysis of flier behavior. The model achieved 90% overall accuracy in classifying flight states, demonstrating its viability as a tool for high-throughput data analysis.
This thesis delivers an integrated, validated, and accessible experimental platform. It successfully bridges the gap between materials-driven flier design and rigorous aerodynamic characterization, providing a standardized, reproducible environment to advance the study of bioinspired soft aerial robots.