Public Lab Research note


The user 'dhaffnersr' has been placed in moderation and will not be able to respond to comments.

Final data analysis for 10w40 motor oil using 400nm UV LED mar12

by dhaffnersr | March 12, 2016 18:56 12 Mar 18:56 | #12838 | #12838

Below is my final data analysis for the 10w40 sample I have been working on, I have been working hard on perfecting my sample preparation techniques and data analysis using spekwin32, So I really DO NOT mind a very critical analysis of my data and sample processing technique, its the only way I'm going to learn and know if I am progressing or what I need to do in order to do better quality work.

I have found something very interesting though, but I will have to do a more thorough comparison to find out why I got better results with the 400nm LED than with the 405nm UV laser, now I had similar peak values as compared to the 400nm scans but the concentration levels were much lower, maybe it's because of the higher energy levels, I'm not sure which readings are more accurate, but I know there are those here with much more experience with this type of work than I.

Well anyway here it is:

final_data_analysis_10w40_mar12_400nm_UV_LED_USE_THIS_ONE.png


27 Comments

Again, I DID NOT award myself this barnstar!

Honest!

Reply to this comment...


I'm not understanding your results as shown in the graph of processed data so here's a few thoughts and questions:

  • It can be helpful, when showing processed data plots, to include the sequence of data (eg. reference, difference, averaging, processing, etc.) plots which clearly show how the final plot was produced.

  • The average intensity scale indicates a peak of '~25' but w/o the RGB channel data or reference plots, it is hard to guess at what to understand about the spectral intensity relative to noise. It looks like the noise is high but perhaps the signal energy is low.

  • When attempting to observe fluorescence, it is desirable to eliminate the carrier signal (400nm LED spike) and keep the SNR high (low noise). The 405nm laser has the advantage that the small beam spot size (monochromatic laser instead of 30-deg angle LED) allows the UV to enter and exit through parallel sides (at a 90-deg angle to the spectrometer) of the cuvette with minimal scattering (which can enter the spectrometer). This allows for minimal attenuation for the spectrometer signal which maximizes the SNR for fluorescence spectra detection. If your LED is directly opposite (inline with the spectrometer) you might rotate the LED to the side (orthogonal to the axis of the spectrometer) when hunting for fluorescence spectra.

  • The base-line noise appears to be significant (on the order of '3' pk on the scale (~=12% of pk which is, at best, only 18dB SNR for the 400nm peak) and some peaks are assigned molecular equivalents. Generally, identifying signal peaks within noise, it requires at least a 6dB SNR to declare that a 'peak' is really a signal. However, since this plot appears to be of processed data, and that data process is not described, it is not possible to know if the "signal" is a real molecular-related signal or not.

  • The large peak (360-440nm) would appear to be just the '400'nm LED as it has the same general shape and width as other typical LEDs. From looking at the plot, I do not see evidence of oil fluorescence -- as typically detectable by the PLab type devices. All oil fluorescence successfully detected so far (via PLab devices) have generally been only able to detect broad signatures across the 400-650nm band. However, if you have been able to somehow extract molecular signatures of specific components which appear even within the large 400nm LED carrier signal, that would be interesting to understand. However, the measurement and data processing trail need to be explained. When signal are present within the bandwidth of the carrier signal (e,g. like with the LED source) it usually requires a very strong signal to be detectable -- and even then, that signal will be heavily corrupted by the carrier so the weaker molecular signal's characteristics will be very difficult to detect.

  • Yes, the plot shows a "lumpy" spectra of the 400nm LED carrier signature but I see no proof within this single plot that those additional contributions to the "normally smooth" broad LED spectrum are, in fact, signatures of fluorescence of oil molecules. I've not noticed such peaks within the fluorescence spectra I've observed. Again, perhaps something else in the data processing which produced this plot could explain.

  • The broader question is: If the identified specific peaks, within the carrier and within the noise, are real fluorescence signals, then where is the rest of the broadband fluorescence spectra from 400-650nm which seems to be typical of crude and refined oils? The absence of that spectra suggest the light intensity of the UV LEDs is simply insufficient to excite sufficient fluorescence in the sample such that all you are viewing is the LED + noise.

Perhaps it would be good to show the data collection path (eg. plots, from reference through sample data through post-processing) and describe the processing steps which support your final plot, analysis and conclusions. That sequence makes it easier for others to both understand your work and to leverage your observations.

Cheers, Dave

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Hey Dave, thank you so very much for your analysis and input, I do have a few questions for you though first, I know that you are using a Sanm CM-020 camera module in your prototype and upon researching that module I know the only specs on it are pretty much just dimensions, there are no given values for the camera's lux sensitivity or any other info.

So my question is, even though your prototype has greater stability and grating design, how can it be anymore sensitive than the Plab spectrometer?

I ran a series of methylene blue tests yesterday from my Gram staining kit, (which the bottle of methylene blue is 1 percent aqueous solution) using my UV laser 405nm, I got the 2 peaks at 603nm and 663nm which the data on record shows is already predicted at shown at 609nm and 668nm. Now my procedure for this was pretty straight forward, I made 7 dilutions so I could compare each for absorption and concentration and my 1st and 2nd dilutions provided results that were, in my opinion, very close to prediction.

I also know my snr with the 0.18mm slit that I am using

signal shot noise flicker noise total noise signal/noise signal/background 180.0 13.4 18.0 25.43 7.08 5.56

Background 32.4 5.7 3.2

Now as far as processing the oil, I can do 1 of 2 things, 1 just scan it the way it is and just get a lump on the graph, which is fine if that's the goal (which is probably fine if your sampling oil in water) or 2, breaking down the aromatic hydrocarbon bonds into their respective components, which is cool also, and maybe more useful, maybe just a perspective thing. So my goal is to test the limits and capabilities of this spectrometer the way it is, with the hardware that it comes with.

I did specify in my post how I separated the hydrocarbons in the oil sample, that's why I used the ethyl acetate and ethanol mix, they can dissolve both polar and non-polar substances, which of course oil is non-polar, each solvent has it's own unique properties for separating conjugation.

I have another question, I have to use a CFL calibration to export my spectral data, is that a problem, and if so how do I solve that?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Hey dave, for some reason the snr data is all messed up and out of line so I'm just going to write it out so it can be clear:

signal-180 shot noise-13.4 flicker noise-18 total noise-25.3 signal noise-7.08 signal background-5.56 background 32.4 5.7 3.2

Reply to this comment...


dhafnersr,

I was particularly interested in your Molar Absorption Coefficients vs Wavelength.

The values seem to be a liitle peculiar to me, but I am not familiar with the calculations that you must have used to get the values.

If you can let me know the calculations that you used, I will then be able to compare my results with yours.

gwilljones

Reply to this comment...


Hi Dave, I'm also curious about the details of your setup. I don't see any fluorescence in you scans, which is confusing... Can you post some photos of how you're set up and how the light is passing through the sample, not only for this but for tests like in your earlier note at https://publiclab.org/notes/dhaffnersr/03-11-2016/tested-10-380nm-uv-leds-with-10w40mtr-oil-thru-quartz-cuvette ?

Thanks. Some more context and process documentation would be very helpful, as Dave Stoft points out.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Hey Guillaume851. here is the equation I use: A=log10(IoI)=ϵlc

Reply to this comment...


Hey Jeff, I know, there is NO fluorescence, I tend to agree with Dave, I think the LEDs just don't have the intensity to cause the type of fluorescence that i am trying to get.

Yes I will post a complete note with pic's showing my set up.

Thanks Jeff

Reply to this comment...


Ok Jeff, here are the pictures of my set up as it stands now:

PIC_0053.JPG

PIC_0055.JPG

PIC_0056.JPG

Reply to this comment...


This is Part two

PIC_0057.JPG

PIC_0058.JPG

PIC_0059.JPG

PIC_0060.JPG

PIC_0062.JPG

PIC_0063.JPG

Reply to this comment...


dhafnersr,

That hasn't helped me one bit.

Does lol mean "lots of love" in email language.

Goodness knows what ϵlc means.

The only bit that is meaningful is A=log(10)

An equation without a description of what is in the equation is meaningless, unless of course it's an equation that everyone knows like E=mc2. Sorry! there's no way I can put the 2 as a superscript in this reply system.

Guillaume 850

Reply to this comment...


Hey Guillaume851, that's good, I like a sense of humor, so try and keep up here, number 1, I use a data program called spekwin32 to process the spectral data that I export, the program does it for me if I want, or you can do it by hand also, which ever is the most "loving" way you like. Anyway below might be a better description for you:

Deriving the Beer-Lambert Law Assumption one relates the absorbance to concentration and can be expressed as A∝c(1)(1)A∝c The absorbance (AA) is defined via the incident intensity IoIo and transmitted intensity II by A=log10(IoI)(2)(2)A=log10⁡(IoI) Assumption two can be expressed as A∝l(3)(3)A∝l Combining Equations 1 and 3: A∝cl(4)(4)A∝cl This proportionality can be converted into an equality by including a proportionality constant. A=ϵcl(5)(5)A=ϵcl This formula is the common form of the Beer-Lambert Law, although it can be also written in terms of intensities: A=log10(IoI)=ϵlc

Reply to this comment...


Also as a quick aside, from all the spectra that I have taken (and they are many,) it is starting to seem that perhaps this spectrometer kit as it is intitally designed, can only really measure absorption more so than fluorescence, which is still a very useful data tool.

Reply to this comment...


Hey everyone, this is part3 of what I am using for data processing this is spekwin32:

PIC_0064.JPG

PIC_0065.JPG

PIC_0066.JPG

PIC_0067.JPG

PIC_0068.JPG

PIC_0069.JPG

Reply to this comment...


Dave, I'll comment in reverse order:

At a more abstract level, I'm suggesting you use a step-by-step process in documentation within your research note (especially when the process is potentially this complex) such that readers of many backgrounds can follow your reasoning and observations. Hope this helps.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Thanks for replying Dave, yes I should have stated more clearly that the LEDs, if they were causing fluorescence then as you stated, it's buried in the noise. The reason I have the laser directly inline with the sample and slit are for that very reason you stated, to maximize the laser signature. I have tried to aim the laser at 90deg to the sample, even with the cuvette right in front of the slit (I used every slit width I had,) and all my intensities were all below 25 percent on workbench capture.

The cuvettes that I have are frosted on 2 sides, it's suppose to maximize light intensity within the cuvette and also it's handy for not putting finger print all over it. The 180 on the SNR is the analyte,

The methylene blue is for 2 purposes, one is, since it is such a common stain used in biology and has a very distinct singature I want to see if I can use that as a calibration curve.

No problem, I can certainly present a separate piece of my exact step by step process and the chemistry behind why I am doing that, I guess I just assume that most everyone here just knows that. Some of my reasoning behind such complex sample preparation, is that refined oil has additives in it that can skew data in directions that might not be where you want to go.

Also I paid attention to one of your notes and you were talking about when taking the spectral image to just have the rgb channel just barely clipping the top.I found your line of thought on more data collection on that level interesting.

You are right though, there is a stability problem with this design but I am trying to tweak it.

Thanks again Dave for your input have a good one!

Reply to this comment...


Ok, your comments help.

Reply to this comment...


Hey Dave, I got those cuvettes from researching the different properties and their function, but I understand what your saying about the narrow gap on the bottom of the acrylic one's. Does Plab sell them with the oil testing kits, because I didn't see them just by themselves in the plab store?

With the cuvettes I have now, there is no choice but straight ahead. I will try your suggestions, thanks.

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Here's the link: http://store.publiclab.org/collections/spectrometry/products/cuvettes-qty-10-or-100

Reply to this comment...


stoft,

You wrote: While I do not have/use that program

Just as a point of interest; if you do ever have a desire to get Spekwin32, it is supplied as freeware to the likes of you and I. I have already downloaded it!

Gwill Jones

Reply to this comment...


PIC_0074.JPG

Hey Dave, here are some more screen shots from spekwin32 and the methylen blue work I have been doing, just thought you would be interested.

Have a good one

Dave H

PIC_0073.JPG

PIC_0074.JPG

PIC_0075.JPG

PIC_0076.JPG

Reply to this comment...


Gwill, thanks. I'll take a look.

Dave, It appears you are looking to observe the effect of methylene blue on the 405nm carrier. If so, I'd suggest that you need to try to observe and separate the noise from the signal. Comparing (like your difference plots) single measurement plots can too easily display noise as if it were a meaningful signal.

The noise, in this case, is essentially the repeatability error which is represented by a band, or range of values, the spectrum can display for the exact same measurement. To observe this, setup the system, drop in the sample, grab the data, remove the sample, replace the sample (same sample), grab new data ... repeat 3 to 5 times and then plot the data. I don't see a function in Spekwin32 to automatically show the repeatability band (which is not just the average) but when the plots overlap, the visually 'fuzzy' band which is formed gives a rough visual idea of the repeatability. (When plotting, only display 400 +/-25nm so the noise becomes clearly visible. Remember, the camera data is roughly 1nm/pixel so this is only 50 data points.)

If you did this for a) the reference solution and then b) after methylene blue is added (to that same solution), and again 3-5 measurement cycle data is collected and plotted (overlap), then visually comparing the two plots sets will start to give an indication of any measurable difference. I can imagine how to compare these in matlab, but I'm not familiar enough with spekwin32 to know how to perform this task with that utility.

Reply to this comment...


Hey Dave, just wanted to show ya I got a fit of 4 today on my CFL calibration

https://spectralworkbench.org/spectrums/74468# CFL calibration mar15 (fit of 4)

CFL_cal_fit_of_4_mar15.png

Reply to this comment...


I wanted to say thanks to @stoft for so thoughtfully responding to each of these questions. I've found it hard to keep up with the rate of publication here, but I agree that it'd be helpful to see the LED setup so we can try to determine why you're not seeing fluorescence (or perhaps why you're not seeing enough fluorescence).

IMG_20160309_170741.jpg

Reply to this comment...


I added a picture of my own setup, which shows very bright fluorescence. This piece slides into the end of the paper spectrometer. It's just a prototype, but it illustrates how I'm generating fluorescence, and provides photographic context of the setup. Can you show us something like that?

Is this a question? Click here to post it to the Questions page.

Reply to this comment...


Hey Jeff, I figured out what was wrong, I redesigned the cuvette holder and shot the laser through at a perpendicular angle here is a set I did with a green laser and here is a plot using a 405nm UV laser and clearly demonstrates fluorescence, I haven't processed it yet because I am still working out some kinks.

https://spectralworkbench.org/spectrums/75046

405nm_UV_eosin_in_iso_mar21_mar21.png

Reply to this comment...


Hey Jeff, here are a few pics of what I got going now;

PIC_0093.JPG

PIC_0094.JPG

PIC_0095.JPG

I think that's pretty good fluorescence

Reply to this comment...


Login to comment.