DIY UV Steriliser


I'm not a doctor nor have I any background in this topic or in any Medical field.

As I learn about this, I'm sharing it as it may be of interest to someone.

Do not blindly trust any information found here.

Use at your own risk!

You may become blind and develop skin issues like cancer when using the material described here.

I have no way of personally testing the efficiency of UV sterilization on anything ,and I only base my theories on documents I link.

We're all aware of the current worldwide shortage of N95 and surgical masks .

Many are trying to build their own, with questionable efficiency, and risky potential outcomes.

And then, I stumble upon the Nukebox project describing how to build a DIY UV-C sterilizer for masks. It seemed too good to be true.

After reading Nebraska Medicine Center's Publication, I got convinced of the plausible technique's efficiency.

But, there is a huge gap between the two projects:

  • 1 The NukeBox is worth less than 100$ while the Nebraska project is using >2000$ equipment from Clordisys. Can it be equivalent ?

  • 2 The Nebraska project use monitoring to verify the dosage received my the mask. Given the nature of the UV-C radiation, you can't look at it naked eye, and especially not for a whole hour! We need a way to testify the mask received a proper dosage. Overkill is the key, but, how can we be sure the tube or ballast is not defective in anyway, that power was "on" all the time, there was not power outage or that the tube is not at the end of it use-full life? This is risky. This is were I found a cheap way to get a (not super precise) way to measure UV-C, that I will share here.


What is UV-C ? And how it kill virus ?

Wikipedia explain it great

Also, have a look to this great document from Clordisys for germicidal usage of UV-C.

What surprised me the most is how little UV-C it takes to kill 99.99% of germs, bacteria, viruses and many others.

By the way, 4-Log reduction = 99.99%.

Based on Nebraska's project, a dose of 60 and 300 mJ/cm2 did a 6-log reduction of bacterial and viral organisms.

This is 99.9999% killing !

Even if only 2-5 mJ/cm2 seems to be generally accepted to kill SARS-CoV-2 virus that causes Covid-19, a 300 mJ/cm2 dosage target looks to me like having a safe margin. Also considering precision of a DIY sensor, and many out of control effects like the unknown "opacity" of different parts of the mask, like were the seams are, we really need a overkill target of many magnitude bigger.

UV-C light is not present in our every day life. UV-C from the sun is mostly all filtered-out by the atmosphere. This may explain why it so easily kill viruses , bacteria, moulds, ext. and why it is harmful to your eyes and skin . These Lamps are DANGEROUS, never look at them with your naked eye, and avoid skin exposure.


First you need a UV-C Germicidal lamp emitting 253.7nm light (sometimes rounded to 254nm in the datasheets). Any other UV flavor is NOT equivalent . Tanning UV lamp or night club UV lamp emit other UV Frequency and are not useful for this. There're some LED available in the UV-C ranges, but the $ per watt is too high to consider them. After some shopping, I've decide to use USHIO tubes. These are high quality, made in japan, that look way more serious that all the no-name Chinese lamps. I mean, the specific glass tube quality composition (that must be quartz), gas filling quality, over-all mechanical quality and longevity of the tube are all details I don't what to mess with. Panasonic seems to build decent tube too. USHIO offer a great variety off tube types and have great documentation.

By the way , the blueish light the bulb emit IS NOT the UV-C we're looking for. What I mean is that if you filter the blue light, the UV-C radiation might still be present, and on the opposite, you may have a transparent surface (like plain glass) in front of the lamp that allow the blue light to pass, but block the UV-C radiation and not have the desired sanitizing effect anymore (more on that later).

These lamps are DANGEROUS, never look at them naked eye and avoid skin exposure.

Fixture and ballast

A bigger challenge is actually finding a fixture and ballast that fit the tube you chose.

These are the four characteristics for the fixture you have or might buy:

    • Connectors on the tube. Something like E17, G5, G13 or G23. Look at the tube's specs for the supported fixture if the fixture doesn't supply the info;
    • Tube power. This will be ruled by the ballast of the fixture you have. You can use a ballast a bit more powerful than the tube, but not the opposite;
    • Tube length. This one is tricky. I found many tube that have the right Power and connectors, but are too short for the fixture/balast available;
    • Tube diameter. This one can sometime forgive a little bit , and will usually follow nicely if the other 3 characteristics match. These are the T5, T8 and T10 numbers and means 5/8, 8/8 and 10/8 inches respectively.

Fixture ballast and bulb match I found

First match I found is an old black light 9W ballast I had with a G23 removable bulb socket and standard north-american 120v E26 lamp input. I've try hard to find another ballast like this one with no success, it is probably more than 20 yrs old, and has a purely magnetic ballast. It still work, and easily power a USHIO 3000304 GPX9 - 9 Watt Germicidal UV-C Lamp.

    • Connectors : G23
    • Tube power:9W Balast power: 5-7 or 9W
    • Tube length : na
    • Tube diameter: na

Current consumption on the 120v side was similar for Standard White 9w Lamp, my old Black-Lite and the UV USHIO lamp. ( 0.150A , 0.152A and 0.146A ). This look very inefficient because 120v x 0.15A = 18W. Amp reading are questionable because there may be many harmonics or inductive power factor, not giving not a true RMS value. But at least the 3 bulb give the similar reading, so it doesn't disapprove the compatibility of the ballast.

3 Lamps comparison

Old Balast Close-up

UV-C USHIO 3000304 Lamp

The second match I found was a "under the kitchen cabinet fixtures" type . This is my preferred one since I can purchase more. I've use a 8W Miyako LS-08-08 with a USHIO 3000016 - G8T5 - 8 Watt - T5 Germicidal Lamp.

    • Connectors : Miniature Bi-Pin (G5)
    • Tube Power:7.2W BalastPower:8W
    • Tube length : 11.3 in
    • Tube diameter: T5

This time I got a 0.112A reading on the 120v Supply with the normal 8w white tube , and a 0.106A reading with the 7.2W USHIO tube

This mean:

13.44w for a 8W white tube --> 0.59 efficiency

12.72w for the 7.2w USHIO tube --> 0.56 efficiency

This electronic ballast is more efficient than my old magnetic one and seems to drive the UV-C tube nicely, but the efficiency difference make me wonder if it is actual driving the 7.2w tube at 7.2w . Would the efficiency be equivalent for both tubes , it may be actually powering the 7.2w tube at 12.72*0.59 = 7.5W . This may explain the small reading error we will see later.

Fixture and Lamp

USHIO 3000016 powered

Also , get rid of any translucent shield the fixture may have, they may block the UV-C


Now the fun part begins.

I've search a bit on major suppliers (digikey, mouser, Newark Av-net, ext) for a UV-C sensor, either photo-diode or photo-transistor, without any luck. I didn't want to pay 100$+ neither was I willing to wait months for not in stock items. When digging deeper and wider I found this guy:

Adafruit 1918 UVA UVB Sensor

I know, I know, it is not a UV-C Sensor per say, but the GUVA-S12SD photo diode it use has just a wide enough detection range that goes a bit into UV-C land. The Low Pressure Mercury Arc tube emit a pretty sharp 253.7nm wavelenght light and this sensor detect from 240 to 370nm . The photo diode is not that sensitive at this wavelength, and we will have to account for it later to get accurate measurement.

Fun fact, I've add the colour spectrum of light on the right graph, so you can see the little spikes in the blue-green and yellow , that give the tube the nice visible blueish colour. Notice how bigger is the invisible UV-C spikes, did I mention it was dangerous?

Lamp Emitted Spectrum

Photo-diode Responsivity

Now, can we get quantifiable reading from this little guy? Adafruit did a pretty good job of packaging a nice op-amp close to the diode that allow you to simply do a voltage reading to get a measurement.

The first test I did was with the sensor as close as I could to the tube. It took about a minute to get the Reading to max out, thus indicating the tube need a certain time to reach full power.

I got a 1.559V reading. Considering the Adafruit circuit is power on a stable 5Vdc source, it should be able to give me up to a 4v reading if radiation was maxing out the sensor. So 1.559v is not in saturation.

Per the Adafruit documentation

Vo = 4.3 * Diode-Current-in-uA. so

Vo/4.3 = Diode-Current-in-uA. so

1.559V/4.3V = 0.36uA = GUVA-S12SD Diode-Current

Per the GUVA-S12SD Datasheet

Photocurent/113 = UV Power in mW/cm2

So 0.36uA/113 = 3.2mW/cm2


This would be true only if the source was at 352 nm , and it is not

Remember the responsibility curve we've look at before, it now comes in handy.

The preceding 3.2mW/cm2 reading would have been good at 352nm , but the wavelenght peek we are mesuring is at 253.7nm . So we have to boost our reading by 0.14/0.023 =6.09

so the 3.2mW/cm2 , should be around 19.5 mW/cm2

Last step is dangerous , because it will falsely boost the reading of ANY light in the 250-360 range .

And it is actually a bit the case since our tube emit really small amount of radiation in the 300nm and the 360nm range. This is also probably explain the small discrepancy we will have at the end.

So , 19.5 mW/cm2 ?!?!?

What does it mean ?

The tube has a diameter of 1.6cm , and emit light on only approx 24cm of its length .

So on the distance I took this measurement (as close as I could) the tube as a surface of Diam x Pi x Length = 1.6cm x 3.14 x 24cm = 120.6cm2

So with this experimental measurement , 19.5 mW/cm2 x 120.6cm2 = 2.32 W of UV light exit the tube

Wait ... why only 2.32 W ?

It is a actually pretty close to specs measurement

Considering the official tube datasheet, only 2.2w out of the 7.2w power given to the tube will exit the tube as UV Light

Ok then , why 2.32w and not 2.2w. Why this 5% imprecision ?

  • As describes before, the 8w ballast may be over-driving the 7.2w tube a bit, maybe as high as 7.5w. Thus, the tube life expectancy may be a bit shortened?
  • As describes before, the 6.09 boost factor that boost our 253.7nm reading AND small wavelengths that are normally negligible and now should not be boosted because the sensor is already sensitive to them .
  • The tube is brand new , and datasheet may be conservative a bit to account for many factory imperfections, or a bit of wear overtime?
  • The sensor was already giving a 0.04v reading with the tube off , from ambient light , or electronic noise, this may also contribute to the error
  • All of the above combined together ?
  • At the end, this is only a 6$ sensor, not made for this. I'm super surprise the number match-up that good!

Derating the 6.09 boost factor on a 2.32/2.2 gain to account for all of this would be safer. -->5.78

Distance Test

Lets try further away to test if readings stay consistent,

The flowing test is only a sanity test , because the equivalent reception cylinder length is only a wild cheated guess.

So all red numbers are guessed





Now that we have a UV-C sensor with some precision we can also verify many items for UV-C transmissivity.


Mylar Sheet

Clear Safety glass

Red Safety glass

Dolar Store reading Glass

Myopia Prescription glass

Original fixture plastic cover

Can we reflect UV and thus improve sterilizer box efficiency ?

Aluminium Foil Self Adhesive seems to have a beneficial effect , like a mirror would have

Since plain glass absorb UV-C , i didn't even try a that.