| Pages:
1
2 |
Jenks
Hazard to Others
Posts: 163
Registered: 1-12-2019
Member Is Offline
|
|
Here is the "finished" detector:
Like many of my perfectionistic projects, it looks good, at least to me, but the problem is that it doesn't work. Worse than that, I'm not entirely
sure why. I've been charging along, and as I approach the end of my five weeks in Ghana and find this main project not working out as I envisioned, I
can only stop worrying about running out of time and return to prayer to ask for a better way.
I'm pretty sure the amplifiers are working correctly, exactly the way the prototype in the solderless breadboard has. But the maximum voltage it puts
out is only 2.2 volts, for a range of only 1.57 - 2.2 volts, not the 1.57 - 4.3 volts put out by the prototype. I don't know how to account for this,
particularly the maximum voltage being the same within 1% on both channels. The only difference between the prototype and the pictured detector is
that the single flow cell available is installed between the UV LEDs and the photodiodes. If I insert the 1mm quartz cuvette between the UV LED and
photodiode/UV filter glass on the prototype, the extra distance brings the maximum voltage down to 4.0 volts, which actually would be ideal because it
has to be limited to that voltage anyway. Maybe the optics of the flow cell are different, as putting two photocells on it forces them near the ends
near the fluid connectors. The photodiodes could be out of alignment, but for them to be out of alignment exactly the same amount is unlikely.
I proceeded to test the detector anyway. The signal is noisy, varying by +/- 50 millivolts when the photodiodes connected straight to the Arduino,
across 100k resistors, had noise around 5 millivolts, which is the limit of the Arduino's A/D converter to measure. Since the plain photodiodes give
out 0 - 0.40 volts without amplification, the amplifier's 0.63 volt range gives less resolution because of the noise.
I planned to start my tests by sending a sample of toluene through the detector, eluting with acetone since it is our cheapest solvent. I quickly discovered, to my surprise, that acetone absorbs 254 nm UV:
Fortunately, ethyl acetate doesn't, so I switched to the 15% ethyl acetate/naphtha mixture I had prepared for actual chromatographic tests. This gave
a peak when the toluene sample was flushed through, but exposed new problems. One is that the UV LEDs overheat in only a few minutes despite the heat
sinks, causing the baseline to rise as the UV output drops. But the most troubling was that the absorbance for the toluene wasn't nearly as high as
expected. I could feed half a milliliter of straight toluene through the system and it only absorbed 30% of the UV on passing through the detector. I
thought maybe the little bit of visible light making it through the UV filter might still be exciting the photodiode, but thanks to the experiment
running straight acetone through the detector, which lowered the voltage almost as low as with the UV LEDs off, that was ruled out. But the fact that
I see any visible light coming out of the UV LEDs bothers me. The specifications volunteered by the vendor of the 0.1 watt, 250 nm UV LED claim that all emmission is between 250 and 260 nm:
If there is visible light, enough to overwhelm the photodiodes, there may be radiation between visible and UV-C also, unless it is due to fluorescense
of some component in the LED. Maybe the UV absorption of toluene, which is sharper than that of acetone, is leaving more UV-B to excite the
photodiodes. I plan to run straight toluene through the detector to see if it is maxed out as with acetone. And it may be that the absorption of
iboga alkaloids is broad enough to be detected owing to it being bicyclic aromatic. But I have another idea I have to try with my remaining
little time.
I brought some cadmium sulfide (CdS) photocells, just because I was trying to bring anything photoelectric I could conceivably use in this system, and I thought
in my fantasy that I might get around to designing drop counters. The peak sensitivity of CdS is 540 nm, green. We know how effective the fluorescent zinc silicate indicator is for visualizing aromatic compounds on
TLC. Taking advantage of this, I plan to try a detector that shines the 254 nm UV through the sample onto a piece of TLC plate and measures its
fluorescence with a CdS cell. Response time of CdS cells is only 100 ms, but since I was planning to collect data every second, that is not a problem. I also hope to use 4mm ID/6mm OD quartz tubing as the flow cell, as I brought two sizes, and I expect it to seal well to polypropylene tubing to connect it to the chromatography
column. Maybe if the UV LEDs can be switched on only 100 milliseconds every second it will prevent overheating, or I can install the UV LEDs on a
fan-cooled CPU heat sink.
Update:
Here is what I have so far:
Of course it will benefit from plenty of screening to block any UV not passing through the sample.
Update:
The detector is coming along, and I'm glad I stopped waiting 24 hours for the silicone to completely dry as it seems firm enough to continue working
gently after an hour or two. Here is the detector with aluminum screens installed, adhered to the quartz tube and heat sink with silicone, of course,
and made by scraping the silica off TLC. There is a clear view of the LED through the "flow cell":
I also dared to try switching the power to the UV LED through a transistor, even though, with a TO-92 case, it isn't meant to dissipate much power. Of
the ten types in my kit I selected the BC337 based on its high 800 mA maximum collector current, high gain of 630 and it being NPN. Its base is simply connected to an
Arduino digital output through a 4.7k resistor with the emitter grounded and the collector controlling power to the UV LED. Neither the resistor nor
the transistor gets hot.
Here is the detector in action, just needing the TLC and CdS cell to be mounted:
The name for this style of detector is the "dog house". The CdS cell selected from my kit was the GL5528, based on this design.
I ran toluene through the original detector. Unlike acetone, it absorbs only about 65% of the 250 nm UV and about 15% of 270 nm UV:
I interpret this as toluene having a narrower, or slightly different, absorption spectrum than the emission spectrum of the UV LEDs.
So far the detector seems to be working, if the room lights are turned off. The voltage in darkness is 0.1 volts and illuminated it is 3 volts, but
I'm sure that can be increased. I was relieved to see the TLC go dark when toluene was passed through the flow cell, meaning this UV isn't just going
around the inside of the tube:
[Edited on 8-8-2024 by Jenks]
|
|
|
Jenks
Hazard to Others
Posts: 163
Registered: 1-12-2019
Member Is Offline
|
|
Today I prepared a case for the "dog house" detector out of a cookie box to protect it from ambient light, connected it to the ~15 mm diameter
chromatography column, flushed it with 15% ethyl acetate/naphtha, purged air from the system and loaded it with 13 cm of 230 mesh silica gel. A
solution of 10 mg of voacangine in 1 mL of the solvent mixture was loaded on the column and eluted. After around ten minutes of elution, the peak was
poorly detected, only 15-20% of maximum absorbance, with noise around 5% of absorbance. Before the peak could pass, the baseline seemed to rise so
that the absorbance stayed 10% above the original baseline. Thinking the drift could be things getting hot inside the box, I measured the temperature
of the new detector case with an IR thermometer, but the temperature was identical to ambient. With the UV LED only turning on for 100 milliseconds
every second I haven't noticed any heating of the heat sink at all. The software is now subtracting the reading with the LED on with the LED off to
correct for changes in ambient lighting, so the noise is despite this.
I reduced the height of the silica to 4 cm and ran a drop of toluene through but it was barely detected. I ran through about 0.1 mL of toluene and
this gave a peak around 30% of maximum absorbance, which is the first peak I have obtained that might give an accurate integration, but that would
make sample requirements too large. I ran straight toluene through and it blocked 75% of the fluorescence. Even studying how the pattern of
fluorescence on the TLC in the detector changed before and after, I don't know if that means UV is getting around the sample or if some is getting
through. The UV LED has its spectrum, the toluene has its different absorption spectrum, and the zinc silicate in the TLC, I learned, can be made to
fluoresce with UV up to around 300 nm. I could stick a dowel into the flow cell to be sure it is blocked, but I am down to my last week in Ghana and
have to move on.
I broke out the VEML6070 UV detector I was able to get through eBay. It was extremely hard to find these for sale. Harrison soldered the header on (his first
soldering experience, well done!) and we plugged it into the Arduino. The readings were integers that went from zero for room lighting to around 30
when the 250 nm LED was shined into it point blank. I couldn't find any way to get a finer-grained value out, and my searching only turned up that the device detects 335-375 nm UV and nothing really below 300 nm. Still, it might have worked if it would output more than two
digits.
|
|
|
Jenks
Hazard to Others
Posts: 163
Registered: 1-12-2019
Member Is Offline
|
|
I don't have much progress to report. I build a one-transistor amplifier for the photodiode in front of the 250 nm LED on the original flow cell. If
gave near zero volts with the LED off and near five volts with it on. But when I ran toluene through the flow cell, the voltage didn't drop.
Apparently, all I made was a UV detecting switch, and making a usefully linear amplifier will take more careful work.
I also tried the LM393 photodiode light sensor made to interface with the Arduino. This circuit is designed to switch a digital output when the light level set by
the trimmer is exceeded. However, it also has an analog output, and this seemed to give a linear output that used the full range of 0-5 volts when the
photodiode was covered or exposed to indoor light sources. Unfortunately, exposing it to the glowing TLC gave a voltage under 0.1 volts. Holding a TLC
closer to the photodiode, which was situated directly above the tubular flow cell in the "dog house" detector, did not improve the signal. Apparently
the fluorescence looks brighter to my eye than it really is compared with ambient lighting.
Resorting to fluorescense as an indirect way to detect UV reminds me of the very promising work that has been done on quantitative TLC. The idea is to apply carefully measured amounts of unknown and known reference samples to a TLC plate and measure the
suppression of fluorescence by the various spots to estimate the amount of material represented by each one. In practice, a photo of the TLC is
analyzed by a computer. According to research applying this to an educational setting, accuracy can get within about 3%. I had difficulty configuring the software to get useful
results when I tested it, but I didn't find any problem with the concept. If the TLC setup and software could be made more convenient and tailored to
the needs of a production laboratory rather than a classroom, this might address the needs sought in this thread.
By now I've tested all the relevant equipment I packed, and it looks unlikely there will be any breakthrough before I leave Ghana next weekend. I am
starting to plan further experiments to continue in California. But given all the responsibilities waiting for me there, and having to rebuild my
laboratory to get started, I don't hope to have further results soon. Thanks to all of you who have patiently followed this thread. I hope some of the
ideas here can inspire your own work.
|
|
|
Jenks
Hazard to Others
Posts: 163
Registered: 1-12-2019
Member Is Offline
|
|
Now I'm not sure if that transistor amplifier was non-linear or if the toluene I tested it with was simply not detected by the detector. After my last
post I got around to making the dual power supply that can be made from a 555 timer IC and a few diodes and capacitors:
This low-power supply is often proposed for use with a dual-rail op-amp, which is the kind (LM1458N) that I have. The advantage is that it removes the
severe constraint from the output range so that it can go all the way, in this case, from ground to five volts. It can go beyond that too, which is why the Arduino input it is connected to should be protected by a 10k resistor.
This power supply worked, but as I was connecting it to the aforementioned op-amp I remembered why I hadn't done this earlier - the ground of this
power supply is not compatible with the Arduino's ground because the 12 voltage power supply, which itself was not properly isolated from the mains,
had been grounded to the Arduino to prevent damage to the Arduino.
But I do actually have an isolated power adapter to use, but it is 24 volts. Looking back on it, my life would have been easier if I had brought the
voltage of this down instead of using the TV board. I scrapped the dual power supply (voltage doubler) circuit and built a power supply splitter for the 24 volt adapter instead.
I used the LM1458N instead of the LM741 in the diagram above. This circuit adds a ground in between power and ground of a single-pole power supply. By
carefully selecting matching 10k resistors, I changed 0-24 volts to +/- 11.6 volts.
Connecting this to an LM1458N, the voltage follower worked, follow by this now simple amplifier:
Since it worked well, except for amplitude, to connect the photodide (in parallel with a 100k resistor) directly to an Arduino analog input, the
photodiode was connected, in photovoltaic mode, to the input of the above amplifier. With the UV LED in the detector on, full range was obtained at
resistance values of 1k for R1 and 14k for R2. Rather than build a 14k resistor, I put a 10k potentiometer in series with a 10k resistor to make a
gain control. When tested with my meter, this circuit worked perfectly and finally gave the desired 0-5 volt output, utilizing the full input range
for the Arduino, maximizing resolution.
This was connected, through a 10k resistor, to the Arduino. The range was still good, but I was seeing a lot of noise in the readings that didn't show
up on the slow meter. Adding the typically recommended capacitors (47 pF in this case) across R2 and the op-amp inputs to prevent oscillation did not
solve the noise. But not wanting to be distracted from a demonstration of feasibility, I ran a few drops of toluene through the detector - and got a
barely discernable peak.
The problem could be that the amplifier is somehow still not linear, and response in the detector is not being amplified. Or maybe the photodiode is
not responding primarily to 250-260 nm UV, despite the UV filter glass intended to block visible light. I had the bright idea to hook the detector
output to the Arduino also, so that the correlation of the detector and amplifier outputs could be observed. This caused the output of the op-amp to
rise to three volts with the photodiode in the dark. That turned out to be because the voltage across the photodiode was now 0.20 volts, which could
not come from the photodiode. The only things across the photodiode were the resistor, the Arduino A1 input and the op-amp non-inverting input, none
of which are supposed to provide power. Disconnecting the photodiode from either the Arduino or the op-amp solved the problem. Thinking I had damaged
A1 I reconnected the photodiode to A3 but got the same result. I think this was noise, as touching the wire gave a similar result.
So I ran three drops of toluene while monitoring the amplifier with the Arduino and the photodiode with my meter. There was no change in the
three-decimal-place meter reading, showing that the problem with this detector design is the UV LED producing UV at wavelengths not depicted on the
datasheet.
I plan to leave my equipment in Ghana, where it is more valuable than in the United States, owing to the difficulty of obtaining it here, hoping it
will be put to use someday by the right person. A new batch of supplies is on its way to my house to continue these experiments. I'm down to two
ideas. Either use a diffraction grating to select the proper range of wavelengths to detect, or find a combination of fluorescent material and
corresponding detector sensitive enough to properly measure the fluorescence. I'm learning toward the latter idea to try to keep the skill needed to
build the detector more easily attainable.
|
|
|
Twospoons
International Hazard
Posts: 1324
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline
Mood: A trace of hope...
|
|
https://www.allaboutcircuits.com/technical-articles/understanding-photovoltaic-and-photoconductive-modes-of-photodiode-operation/
In your case I would use photo conductive mode for the extra linear range. Note that you want your diode reverse biased, as shown in the diagrams.
Helicopter: "helico" -> spiral, "pter" -> with wings
|
|
|
Jenks
Hazard to Others
Posts: 163
Registered: 1-12-2019
Member Is Offline
|
|
Thanks for the link. I think I am getting closer to understand these circuits. I have a question though. The article says,
| Quote: | | Applying a reverse-bias voltage to a pn junction causes the depletion region to become wider. This has two beneficial effects in the context of
photodiode applications. First, a wider depletion region makes the photodiode more sensitive, as explained in the preceding article. Thus,
photoconductive mode is a good choice when you want to produce more output signal relative to illuminance. |
But then it uses this graph as an illustration:
The graph shows the same increase in photocurrent resulting from the same increase in illumination regardless of the bias at low illumination. So how
can increasing the bias be said to increase sensitivity, particularly if it increases the dark current?
[Edited on 15-8-2024 by Jenks]
|
|
|
Twospoons
International Hazard
Posts: 1324
Registered: 26-7-2004
Location: Middle Earth
Member Is Offline
Mood: A trace of hope...
|
|
I think their statement is a bit misleading without some context. Most applications are looking for a rapidly varying signal, and the reduction in
junction capacitance due to reverse bias is useful for increasing the bandwidth of the system.
The other thing that graph shows is an order of magnitude increase in dynamic range - which allows the use of a stronger light source and leads to a
better signal to noise ratio.
In your case that would certainly be true - you could use a lower gain on your amplifier which would reduce the noise floor.
Which brings me to another thought- a couple of posts back you described using the photodiode in photovoltaic mode by putting a 100k resistor across
it and amplifying the resulting voltage.
There are a couple of problems with this approach:
-it has very limited dynamic range as at some point the diode becomes forward biased and any increase in photocurrent is lost;
-it is inherently non-linear as the relationship between current and voltage in a diode is exponential (this is related to the previous issue);
- your 100k resistor is a significant noise source, both on its own and because it will effectively amplify the 1/f noise (shot noise) from the
photodiode.
I would strongly recommend returning to the usual transconductance amplifier setup, with some reverse bias on the diode. The increase in dark current
is easily accounted for in your software.
[Edited on 15-8-2024 by Twospoons]
Helicopter: "helico" -> spiral, "pter" -> with wings
|
|
|
| Pages:
1
2 |
|
![[*]](images/xpblue/default_icon.gif){kind=icon}
