Back to Image Processing

Autonomous UV-C Sterilization Machine

Senior Design Capstone · Applied UVc Sponsorship · Ray-Tracing Simulation + Empirical Validation

Abstract

A team Senior Design capstone (UCSD MAE 156B, sponsored by Jeffrey Levitt of Applied UVc) building an autonomous UV-C sterilization fixture for shopping-cart-like contact surfaces. The team's design target was a log-3 reduction (≥ 99.9 % kill) of RNA-virus surrogates. My contributions were the ray-tracing simulation that predicted UV-C dose distribution across the cart geometry and the image-processing pipeline used to quantify dose from indicator-strip photographs after each empirical test run. Together they closed the loop between "what the fixture should deliver" and "what we measured."

UV-C Dose & the Log-Reduction Target

RNA-virus inactivation by UV-C follows a single-hit kinetics model:

N(D) / N_0       =  exp(−k · D)
log10[ N(D)/N_0 ] =  −D · k / ln(10)  =  −D / D_10

Where D = ∫ I(t) dt [mJ/cm²] is the integrated 254-nm fluence and D_10 is the dose for a 1-log reduction (literature values for SARS-CoV-2 and surrogates: 1.7–7.0 mJ/cm² depending on the source). A log-3 reduction therefore requires D ≥ 3 · D_10, with safety margin set by the worst-case material and worst-case geometry. The hard part is not delivering that dose to a flat panel facing the lamp, it is delivering it to handles, undersides, and seam shadows.

Ray-Tracing Simulation

Forward Monte-Carlo ray tracing on a CAD-imported cart geometry. Each lamp face emits ≈ 10⁶ rays per simulation run with cosine-weighted angular distribution; rays are propagated through the chamber and either absorbed at a surface, scattered diffusely (Lambertian for sheet metal, low-albedo for textile handle wraps), or escaped past the chamber boundary. Per-surface incident energy is binned into a UV-C fluence map.

I_surface(x)     ≈  Σ_r  E_r · cos θ_r
D_surface(x)     =  I_surface(x) · t_exposure
log10 reduction  =  D_surface(x) / D_10

The first line bins per-pixel incident fluence over all rays r hitting surface point x; the second integrates over exposure time; the third converts to log-reduction via D_10.

Sweeps over fixture geometry (lamp count, lamp angle, baffle layout) and cycle time identify the configurations that achieve worst-case D ≥ 3 · D_10 on every contact surface inside the manufacturable design envelope.

Inner-surface incident-fluence maps

Per-surface incident UV-C fluence after a full cycle. Hotter colours = higher dose. Handles and overhangs are the worst-case regions and drive the cycle-time choice.

UV-C incident fluence map on the inner bottom surface of the cart sterilization chamber

Inner bottom

UV-C incident fluence map on the inner side wall of the cart sterilization chamber

Inner side

UV-C incident fluence map on the inner front surface

Inner front

UV-C incident fluence map on the inner rear surface

Inner rear

UV-C incident fluence map on the cart handle, the worst-case region for dose coverage

Handle (worst-case region)

Per-region absorbed-ray totals

Volumetric absorbed-ray summaries used as a sanity check on the per-surface fluence maps. The numbers in the labels are absorbed-ray totals, not dose.

Volumetric ray-absorption summary at 4.7 million rays

4.7 M rays absorbed

Volumetric ray-absorption summary at 2.1 million rays from front view

Front view, 2.1 M rays

Volumetric ray-absorption summary at 1.0 million rays from back small view

Back (small), 1.0 M rays

Volumetric ray-absorption summary at 4.3 million rays from back large view

Back (large), 4.3 M rays

Image-Processing Validation

The empirical validation step uses commercial UV-C indicator strips on the contact surfaces. After a sterilisation cycle, each strip is photographed under controlled lighting; the image processing pipeline converts the colour change into a quantitative dose estimate that can be compared against the simulation prediction at the same surface point.

  1. Colour correction. Greycard reference in every photo to normalise white balance and exposure.
  2. Strip localisation. Threshold + contour analysis to isolate each indicator strip from the cart background.
  3. Per-strip dose estimation. Average colour inside the strip mask is mapped to UV-C dose via a calibration curve fit on a known-dose reference panel.
  4. Surface registration. Each strip's known location on the cart maps it back to a pixel in the simulation's surface fluence map for direct comparison.
End-to-end image-processing pipeline for UV-C indicator-strip dose estimation, showing colour correction, strip localisation, and dose calibration

End-to-end image-processing pipeline

Empirical Test Strips · Front and Rear Faces

Per-trial output for two of the cart faces: the raw photograph of the indicator strips, and the corresponding dose-coded overlay produced by the pipeline. Trials were repeated across 100+ exposures to characterise variance.

Photograph of UV-C indicator strips on the inner front face of the cart

Front face inner · raw

Pipeline overlay of dose estimates on the inner front face

Front face inner · dose overlay

Photograph of UV-C indicator strips on the front face of the cart

Front face · raw

Pipeline overlay of dose estimates on the front face

Front face · dose overlay

Photograph of UV-C indicator strips on the rear face of the cart

Rear face · raw

Pipeline overlay of dose estimates on the rear face

Rear face · dose overlay

Sponsored by Jeffrey Levitt and Applied UVc. Built with team-mates Nethmi Arachchi, Liam Carmody, Adam Douglas, and Aryian Schreiner. Project page: UCSD MAE 156B Team 25.