Dear Paulina

Generally the ATR will be better than FTIR in surface analyzing. It is very simple and also it doesn't need any spot. On the other hand, I agree with Jan. Using the average value of several replication could be roughly incorrect. It is better you double check the homogeneity of the samples.

Also, sometimes you need to inverse the peaks for some investigations, so the exported data is needed.

Hai,

Unless u r not bothered about intensity and relative measurement ( mean comparison of this spectra with other spectra interns of intensity ) averaging is not required.

and if you are getting all the peaks in all the runs at the same wave number then also u no need :) if u are missing any peaks :) then u need to average :)

Each sample is likely to present different superficial composition in different spots. usually for FT-IR measurement the sample should be homogeneous :) if they are all the same sample then why need to put in different spots

Hope it helps you

Classical measurement of transmittance in IR region cannot be generally used to ”determine the composition of obtained materials' surface” as You wrote, in my opinion. The result will be related to whole irradiated volume.

And answer to your question – it depends on homogeneity of your samples, reasons of observed differences and finally measurement purposes.

I don’t know any details about your samples however, sometimes this averaging could be roughly incorrect and confusing. (E.g. if the differences are significant, You may finally not measure some uniform material which You want to characterize but some kind of mixture. Also, if You are dealing with monocrystalline material and finally declare something like averaged ”isotropic value”. Because in that case You will lose information related to anisotropy of monocrystall and averaged values necessary needn’t be useful for pollycrystalline material due to scattering and other effects on grain boundaries. There are plenty of possible reasons caused observed differences. Thus, be careful of real reasons behind differences in your results.

Otherwise, I don’t see any problem, if You describe your procedure in experimental part. However, the number of spots should be high enough to obtain some meaningful average value. But generally – measurement with bigger spot (if possible) will be better.

Dear Paulina

Generally the ATR will be better than FTIR in surface analyzing. It is very simple and also it doesn't need any spot. On the other hand, I agree with Jan. Using the average value of several replication could be roughly incorrect. It is better you double check the homogeneity of the samples.

Also, sometimes you need to inverse the peaks for some investigations, so the exported data is needed.

Dear Paulina

If the problem is surface analysis, use ATR-IR.

Hi Paulina,

If you need to get transmission spectrum of the whole sample just move it away from the focus until it is all covered by the IR beam. In this way you will get the averaging automatically. Still it is worrisome that you have a large inhomogeneity in you sample composition. I would recommend to mix components more thoroughly. If you have an access to the freezer mill (http://www.spexsampleprep.com/products_by_category.aspx?cat=2) - it would be the best mixing technique.

Since "Each sample is likely to present different superficial composition in different spots", I guess you have to do average - either average the compositions (a mean) of obtained from multiple surface spots or average the spectra first (no matter what IR you use) than assess the composition (a mean) of something you are interested using the averaged spectrum. Good luck.

Dear Paulina,

I'm assuming you are using ATR or HATR with your samples. My comment would be to present an example spectrum so long as all peaks are presented in all of your spectra and then state that you observe variations in peak heights which may be induced by inhomogeneity of the sample composition or by changes in surface texture.

Bear in mind that the evanescent wave penetrates a few microns (1-5 depending on wavenumber) into the sample so you are in fact sampling both the surface and a portion of the bulk.

If your spectra evidence different peaks than you may probably want to show a composite Spectrum rather than trying to average as the later might yield strange peak shapes.

Paulina,

First of all when you say " I often apply FTIR (Fourier Transform Infrared Spectroscopy) to determine the composition of obtained materials' surface", you didn't say enough ! what method of IR spectroscopy you use ? is it reflectance, is it transmittance, is it ATR ...? What kind of samples you use ? pure or KBr (CsI) dilute one ? If you want a good answer ask for good questions ... waiting for more détails ...

if you know the film tickness , you can use the Brodsky formula of bonds density

Dear All,

Thank you very much for your interest and all the answers. As some of you suggested, here I provide more details.

These are stainless steel samples. At first, I electropolish them to remove oxide layer. Then the samples undergo a specific procedure of polymer grafting. Among other methods of analysis, I use FTIR-ATR to determine the composition of surface after electropolishing.

I try to obtain FTIR-ATR spectra for electropolished steel. I do this because I suppose that after electropolishing there are functional groups present on steel surface (e.g. phosphates). The problem is that the spectra that I get are sometimes different on different spots on surface of the sample. I mean that on one spot on the sample surface I get peaks corresponding to phosphates and on the other spot on the same piece of steel I don't get these peaks. This is the reason of my question about averaging the spectra got on several spots on the same sample.

I agree that FTIR-ATR may be not be the best method of surface analysis in this case. I intend to use more accurate methods of chemical characterization of surface of samples, e. g. EDS.

I would like to add some information to my post. FTIR-ATR is in this case a qualitative technique for me. Thus, I will not take any information form the peaks intensitivity.

Dear Jan Gutwirth, thank you for your answer. My software allows to export obtained spectra as a table - as I wrote in my first post.

Again, I want to thank you all for your kind answers.

Paulina,

If you can be sure that the ATR material is always at the same distance from your metal sample then you could average results to quantify amount of residuals after polishing. But I would suggest using thermal emission measurement where you can sample all your surface at once. With a cooled MCT detector you can set the temperature of your sample at slightly higher than 50 C - it will be enough to measure the thermal emission signal. Molecular residues will produce emission peaks on a broad black body background emission from the metal (its emissivity is low).

After looking to all comments here, I would say that different spectra appeared due to, inhomogeneity of the surface, sometimes some parts are not polished properly then it will have two different properties of deposited polish and steel it self. What I would suggest, you should do AFM analysis of the surface to check the homogeneity of the surface and then do ATR, for the similar surface you should get the same spectra with little variation. I think you have problem of homogenous electro-polishing, if surface is not covered completely, you will definitely get different peaks. I hope, it will give an idea, where is the problem?

Now, after some additional information, I’m thinking about two possible reasons which overlap each other. The first is ratio of area covered by your ATR crystal and grain size of your stainless steel, i.e. especially if the measured area is not large enough to cover sufficient number of grains, your measurements may become not representative. The second is, due to the complex character of surface, phosphates may be bonded preferentially due to: i) surface (i.e. grain and especially grain boundaries) curvature (this will play role especially if You consider more physical-based bonding (sorption) rather than regular chemical bonding); ii) surface composition (this will be more important for chemical bonding) and may be also due to crystallographic orientation of grains. However, without any deeper knowledge of topic, this is quite speculative.

You can use Energy Dispersive Analysis (EDX), however, I see some weakness of this technique. Firstly, it is not necessary enough sensitive for light elements like Phosphorus. Also precious resolution between Cr/Mn and Mn/Fe is not so easy due to overlapping of K-alpha and K-beta lines. Concentration of Mn is low in common stainless steels, however I am not sure about its distribution. Probably the best solution may give X-Ray Photoelectron spectroscopy (XPS) which also provides information about bonding relations.

Dear All,

I am using Nicolet 6700 FT-IR Spectrometer with Smart Orbit accessory. Smart Orbit provides ATR analysis. It is designed especially for hard samples. With Smart Orbit active sample area is 1,5mm. It is said that it's a large area for FTIR-ATR analysis. However, my samples are much larger (discs with 15mm of diameter).

At the moment, I cannot tell what detector is provided with my spectrometer. I will need to check it after the Christmas break.

Dear Jan, it is very interesting what you said abour EDX technique. (I believe that it is the same as EDS). Since I have little experience with it, I didn't know that it may be impossible to detect phosphates with the use of this method. However, I have been advised against performing XPS. I was told that it might be too sensitive method. I could get varied results not only because of my surface modification experiments, but also because of the composition of steel itself. I mean that the steel processing, e.g. rolling, cause varied composition of different pieces of steel.

Dear Ritesh Tailor, I will definitely perform AFM as you suggested. I already planned to do this, but I intended to use FTIR-ATR for preliminary analysis.

Dear Nick Agladze, thank you for your answers. However, I am not sure if I understood it properly. Do you mean that with thermal emission measurment I could measure the whole sample at once?

For any spectroscopy, never co-add spectra from different regions to create averages. Average the composition results of each spectrum.

Use XPS to determine surface chemistry, because EDS is far less surface sensitive and does poorly for low-Z elements.

Depending on sample roughness, use surface profilometry or interferometric microscopy rather than AFM to determine surface topography.

Dear Paulina,

Yes, thermal emission can be measured from the whole sample. I do not know whether your FTIR is equipped with the emission port and what optics does it have there. But assuming that the emission is collected by the parabolic reflector with ~ 150 mm focal distance a typical spot size for a moderate resolution ~ 4 wavenumbers will be 10 mm diameter at 2400 wavenumbers. Of course thermal emission is not one of the standard techniques and you might need to make some hardware yourself. If fact you can set it up right in the sample compartment using a beamsplitter according to the attached sketch.

Yes, EDX and EDS is acronym for the same technique.

Depends on configuration of particular EDX analyzer, sometimes it is not possible to obtain peaks related to elements lighter than Na. And if these peaks are possible to be obtained, quantitative analysis is not recommended (generally or without making of calibration set). However, newer equipment is announced for better detection of light elements, so You should try here. And may be some semi-quantitative analysis (comparison of content without any specification of content) will be OK for You.

Generally, for tracing of light elements is better to use EDX coupled with Scanning Electron Microscope (SEM-EDX) than X-Ray Fluorescence (XRF) and Silicon Drift Detector (SDD) vs Si(Li) diode or PIN diode. Also Wavelength Dispersive X-Ray Analysis (WDX) is more suitable for light elements. But practically it will depend on particular spectrometer configuration. E.g. Total Reflection XRF (TRXRF) in high vacuum with suitable detector finally may give You best results because of low background and limited penetration depth.

Alina Sims, Ministry of the Environment, Toronto

If you sometimes see phosphate peaks and on other sampling spots not it there could be 2 reasons. First that surface is inhomogeneous i.e. phosphate layer doesn't have uniform thickness. The second reason is that surface is not flat enough to achieve sufficient contact with ATR despite small domensions of ATR (ca. 1mm). Perhaps try to look at the sample under stereomicroscope and check surface topography and homogeneity of covering layer. If substrate (steel) is polished why don't you try reflectance. You could also send your spectra. Anyway I wouldn't avarage spectra, it doesn't solve problem.

pozdrawiam

I may also suggest Raman microscopy to identify phosphates beside the microanalysis using SEM-EDX

Dear Paulina,

What kind of measurement are doing? ATR or reflection (specular or difuse). Are you quantifying the different compositions? If yes, what I suggest is you determine the compositions for each spectrum (each point) and, after average the quantities!

Dear Paulina, have you tried to analyse your FTIR spectra dataset by Principal Component Analysis (on mean centerd spectra - in case you are interested and you have doubts on PCA, please ask me for details) to compare the between/within spectra variability ? A simple look to the PCA scores plot could be of great help to decide whether it makes sense to average the spectra of a single sample - if the spectra of the different spots of the same sample are clustered together - or if the variability of the spectra acquired on a given sample is comparble to that of specrta from different samples - in that case it would not make sense to average spectra.

Hi Paulina,

I think the decision to average the spectra depends on what information you want from the sample. If you want to determine the average composition of the surface, then averaging the five spectra is the correct procedure, In OMNIC you can do this by loading up the five spectra into the window, then add the first two (by selecting them both, then choose Process, Add) then select this one and the third spectrum, add the addition result to the third spectrum, and follow this procedure for the fourth and fifth spectra until you've added them all together, Then select the final addition result, choose Process, Multiply, and enter 0,2 to effectively divide by 5; this will give you the average spectrum. If on the other hand you're looking for variation in composition over the surface, you'll want to overlay the spectra and look for differences between the bands. You can also apply chemometrics or 2D correlation spectroscopy to look for which wavenumbers show any differences .

Hi Paulina, I forgot to add that you most probably have a TEC (thermo electrically cooled) DTGS detector. I use a Nicolet 8700 FTIR spectrometer which I think has the same one for the mid-IR region :)

@A Ulrici: " A simple look to the PCA scores plot could be of great help to decide whether it makes sense to average the spectra of a single sample …"

@M Nieuwoudt: "If you want to determine the average composition of the surface, then averaging the five spectra is the correct procedure …"

It is NEVER the correct procedure to co-add spectra from different sampling regions on a sample. NEVER! Nothing lends any degree of statistical validity to this method of "averaging". It is nothing more but a sloppy, lazy, or blissfully ignorant approximation to the rigorous analysis that is to be applied.

When you want to determine an average numerical value from N measured spectra, average the set of N numerical results from each spectrum. This way, you obtain both an average and a standard uncertainty value.

When you want to show a representative spectrum, show one of the N set. Every scientist understands the word "representative" means one that is chosen to display the basic features of the set of N.

If description of the average composition of your sample is the goal, then you may like to average the spectra. If the uniformity of the composition is the point, then you don\t need averaging.

Taking all your remarks into account, I will just take advantage of the fact that FTIR indicated the inhomogeneity of my samples. This is an important information for me and will help me to plan next experiments.

Of course, I definitely won't co-add the spectra taken from different spots of the sample.

I am glad that I received so many answers. I think this is a valuable discussion. Thank you!

@ Jeffrey Weimer - "It is NEVER the correct procedure to co-add spectra from different sampling regions on a sample. NEVER! Nothing lends any degree of statistical validity to this method of "averaging". It is nothing more but a sloppy, lazy, or blissfully ignorant approximation to the rigorous analysis that is to be applied"

What I meant is that making a thorough evaluation of the spectra variability, i.e., performing a statistically sound comparison between repeated spectra acquired at single points and spectra measured at different locations of the same sample, together with comparing this spectral variability with the variability between different samples can be of great help to compare measurement reproducibility, spatial variability and variability between samples. Avoiding any a priori assumption ("It is NEVER the correct procedure to co-add spectra from different sampling regions on a sample. NEVER! "), deciding instead what to do after having performed a proper statistical data analysis is not at all a "sloppy, lazy, or blissfully ignorant approximation " (what a bad language for a scientific discussion!!). This helps to define whether the spectra vary with varying the measuring locations, i.e., if it is reasonable or not to define a given spectrum (or the average of spectra whose difference is exclusively due to instrumental noise) as being REALLY representative or not. The term "representative" is too often misused. Using a lot of spectra measured at different locations is surely preferable to better account for all the possible sources of variability, but in some cases it is necessary to reduce the amount of data, and when averaging just means lowering the random error, this is not a sin.

@A Ulrici: I can agree with one thing that you may be saying. As I understand, you propose that someone might report a co-added spectrum as a "representative" result of a set of spectra because any one spectrum from the set by itself is too deviant to report as representative. OK, BUT!!! ... This better not be the ONLY report that is made about the entire data set. Indeed, I had better be able to find quantitative statistics in the report about the levels of uncertainty that arose across the entire set of spectra before they were co-added.

I can amend what I said as follows:

A report that shows just one co-added spectrum by itself (as a "representative" spectrum) and otherwise has absolutely no information about the level of uncertainty found over the entire set of spectra that were used to make the co-added spectrum is truly from someone who is lazy, sloppy, or blissfully ignorant. This approach is NEVER correct under any circumstance. NEVER!

Paulina,

I think you said that you have been informed that the composition of the surface of the stainless steel sample (substrate) will vary from location to location. Normally the steel strip surface (substrate) will be of a fairly consistent composition over the short distances it appears you are working with. If this is commercial material it will most likely have been continuously cast, there can be some elemental fade from the front to the back of the cast, (which are typically of size 100 + tonnes), there will be some more local macro segregation, and micro segregation of elements, the former is found over 'large' distances, can be cm to metres (macro flow effects), the latter occurs over original individual grains during their dendritic solidification, driven by partitioning between solid and liquid phases (thermal gradients, chemical potential gradients are involved). However, you will not encounter these on a typical strip surface. If it is an austenitic stainless steel, (300 series) it will be more or less a single austenitic (FCC) matrix phase (if it has been made/treated correctly !). There can be other small micro-nano sized phases present, but these should be quite infrequent in the these days more standard 300 'L' grades, (low carbon).  You can ask the spectroscopists critical questions regarding their different techniques for quantification of chemistry (& surface chemistry), i.e. repeatability, reproducibility, calibration, standards, cross verification between techniques etc. etc. Influence of the characteristics (physical/chemical etc.) of the sample presented to the machine on the results obtained and so on. You mentioned EDA/EDX, unfortunately  this method does not give the accuracy of the old wet chemical (bulk) analysis methods, or the newer techniques for bulk analysis of metals, e.g., ICP OES ICP MS, spark OES, combustion method ('Leco') for C, N, S. It might have been updated with advances in detectors, but the old NAMAS accreditation of that method (EDX) used to specify that we quote errors of  +/- 1% absolute or +/- 10% relative which ever was greater. That I used to think was a conservative figure, and you can do better and use that data, if not supplying to an external third party. Alternatively, if you use EPMA (WDS) and sit on a spot to get enough X-Ray counts from the characteristic lines of the elements of interest you can do a lot better (i.e. much improved accuracy) than EDX.

Sorry if this post of mine is a bit off topic, I found your thread when I was searching for info on the practical application of FTIR ATR to determine the composition of oxides on steel surfaces (limitations accuracy etc. etc.). Found the discussion on averaging interesting. If I read it right, the nub was that you shouldn't overlay the spectra and average them.

If the technique does not affect the surface and you can repeat your analysis at exactly the same locations. I'd do that over and over again, and repeat the process on several samples, then analyse the variance in the results formally, i.e. after applying whatever is the accredited method to convert those spectra to composition.

I might include different operators, and the effect of doing the same samples on different days on the same machine. NIST in the US and the equivalent body in the UK, UKAS give general recommendations on the estimation of uncertainty of measurements, but it's not often that those methods are adopted for PhD work/papers.