Ontario Tech University

An 8-month project across two semesters (Fall 2021 to Winter 2022), driven by the COVID-19 pandemic. The challenge: build a robot that could autonomously navigate indoor spaces, disinfect the air, detect human occupancy to avoid exposing people to UV-C radiation, and operate continuously without human intervention.

Six members: three mechanical engineering students and three mechatronics engineering students. The mechanical side (my focus) handled disinfection system design, CFD simulation, stress analysis, material testing, and manufacturing. The mechatronics side handled autonomous navigation, motor control, and face detection.

A fully autonomous robot combining dual-stage air disinfection (HEPA filtration + UV-C germicidal irradiation in a spiral maze chamber), V-SLAM autonomous navigation using LiDAR, and a face detection algorithm achieving 90% accuracy to detect human presence. UV-C was fully enclosed with emergency stops for safety.

The design went through 4+ iterations, each driven by CFD simulation results and stakeholder feedback.

We started with a rectangular maze design. CFD simulation revealed dead zones in the corners where no airflow occurred. We moved to a two-chamber design with dual HEPA filters, but simulation showed severe flow impedance where the second filter was counterproductive.

After stakeholder feedback pushing for softer aesthetics, reduced weight, and empirical fan testing, we arrived at the final design: a cylindrical spiral. The spiral baffles eliminated dead zones, maximized UV-C exposure time, and allowed 360° air intake through a cylindrical HEPA cartridge. CFD validated the final architecture completely.

The system used a three-chamber architecture. The first chamber contained a cylindrical H13 True HEPA filter (99.97% efficiency for particles ≥0.3 µm) with internal UV-C lights for surface disinfection of trapped particles. The second chamber was the spiral maze with quartz glass baffles (selected for high UV-C transmissibility) and 7 UV-C germicidal LED strips at 265nm wavelength, providing ~0.25 seconds of exposure to break viral RNA. The third chamber handled exhaust.

Total fan configuration: 8 fans (3 intake + 3 transition + 2 exhaust), each 120mm 12V DC axial rated at 250 CFM individually.

The structural frame used aluminum extrusion (T-slotted profiles) for high strength-to-weight ratio, rigidity, and modularity. The enclosure was heat-formed acrylic sheets, laser-cut from SolidWorks 2D drawings. We 3D-printed 16 custom curved brackets (7 kg of PLA filament) to connect the aluminum extrusion to the cylindrical geometry, designed in modular pieces to accommodate printer size limits. UV reflective anodized aluminum sheets lined the interior to maximize UV-C intensity.

FEA in SolidWorks validated bracket and housing structural integrity under load. The bottom acrylic panel was a critical test: FEA predicted failure at motor mount holes due to stress concentration under 50 lbs of load. We ran a physical three-point bending test (modified ASTM-D5023), and the acrylic fractured at exactly the predicted location. The simulation accurately predicted the real-world failure mode.

We replaced the panel with MDF (medium density fiberboard), which showed no deformation under the same 50 lb load. Data-driven material substitution, validated by physical testing.

Air quality testing: all particulate concentrations reduced to 0 µg/m³ within 30 seconds. EPA AQI achieved: 0 (requirement was ≤30). Measured airflow: 72 CFM, providing 2 air changes per hour for a 250 ft² room. Face detection: 90% accuracy with zero false positives.

While the mechatronics students led the software and controls work, I gained exposure to and understanding of these systems through 8 months of collaboration: V-SLAM navigation using Intel RealSense L515 LiDAR, RTABMAP algorithm, dead reckoning with encoder-based odometry, ROS Melodic running on an NVIDIA Jetson Nano, PID motor control with Arduino Due, and Haar cascade face detection with OpenCV. I found this work genuinely interesting, and it contributed to my eventual pivot toward software.