1、Designation: E2310 04 (Reapproved 2015)Standard Guide forUse of Spectral Searching by Curve Matching Algorithmswith Data Recorded Using Mid-Infrared Spectroscopy1This standard is issued under the fixed designation E2310; the number immediately following the designation indicates the year oforiginal
2、adoption or, in the case of revision, the year of last revision. A number in parentheses indicates the year of last reapproval. Asuperscript epsilon () indicates an editorial change since the last revision or reapproval.1. Scope1.1 Spectral searching is the process whereby a spectrum ofan unknown ma
3、terial is evaluated against a library (database)of digitally recorded reference spectra. The purpose of thisevaluation is classification of the unknown and, wherepossible, identification of the unknown. Spectral searching isintended as a screening method to assist the analyst and is notan absolute i
4、dentification technique. Spectral searching is notintended to replace an expert in infrared spectroscopy. Spectralsearching should not be used without suitable training.1.2 The user of this guide should be aware that the results ofa spectral search can be affected by the following factorsdescribed i
5、n Section 5: (1) baselines, (2) sample purity, (3)Absorbance linearity (Beers Law), (4) sample thickness, (5)sample technique and preparation, (6) physical state of thesample, (7) wavenumber range, (8) spectral resolution, and (9)choice of algorithm.1.2.1 Many other factors can affect spectral searc
6、hing re-sults.1.3 The scope of this guide is to provide a guide for the useof search algorithms for mid-infrared spectroscopy. The meth-ods described herein may be applicable to the use of thesealgorithms for other types of spectroscopic data, but each typeof data search should be assessed separatel
7、y.1.4 The Euclidean distance algorithm and the first derivativeEuclidean distance algorithm are described and their usediscussed. The theory and common assumptions made whenusing search algorithms are also discussed, along with guide-lines for the use and interpretation of the search results.1.5 The
8、 values stated in SI units are to be regarded asstandard. No other units of measurement are included in thisstandard.2. Referenced Documents2.1 ASTM Standards:2E131 Terminology Relating to Molecular SpectroscopyE334 Practice for General Techniques of Infrared Micro-analysisE573 Practices for Interna
9、l Reflection SpectroscopyE1252 Practice for General Techniques for Obtaining Infra-red Spectra for Qualitative AnalysisE1642 Practice for General Techniques of Gas Chromatog-raphy Infrared (GC/IR) AnalysisE2105 Practice for General Techniques of Thermogravimet-ric Analysis (TGA) Coupled With Infrare
10、d Analysis(TGA/IR)E2106 Practice for General Techniques of LiquidChromatography-Infrared (LC/IR) and Size ExclusionChromatography-Infrared (SEC/IR) Analyses3. Terminology3.1 DefinitionsFor general definitions of terms andsymbols, refer to Terminology E131.3.1.1 Euclidean distance algorithmthe Euclid
11、ean distancealgorithm measures the Euclidean distance between eachlibrary spectrum and the unknown spectrum by treating thespectra as normalized vectors. The closeness of the match, orhit quality index (HQI), is calculated from the square root ofthe sum of the squares of the difference between the v
12、ectors forthe unknown spectrum and each library spectrum.3.1.2 first derivative Euclidean distance algorithmin thefirst derivative Euclidean distance algorithm the Euclideandistance is also computed, except the derivative of eachspectrum is calculated prior to the Euclidean distance calcula-tion.3.1
13、.3 hit quality index (HQI)a table which ranks thelibrary spectra in the database according to their hit qualityvalues (see 7.5).3.1.4 hit quality valuethe spectral search software com-pares each spectrum in the database to that of the unknown and1This guide is under the jurisdiction of ASTM Committe
14、e E13 on MolecularSpectroscopy and Separation Science and is the direct responsibility of Subcom-mittee E13.03 on Infrared and Near Infrared Spectroscopy.Current edition approved May 1, 2015. Published June 2015. Originallyapproved in 2004. Last previous edition approved in 2009 as E2310 04 (2009).D
15、OI: 10.1520/E2310-04R15.2For referenced ASTM standards, visit the ASTM website, www.astm.org, orcontact ASTM Customer Service at serviceastm.org. For Annual Book of ASTMStandards volume information, refer to the standards Document Summary page onthe ASTM website.Copyright ASTM International, 100 Bar
16、r Harbor Drive, PO Box C700, West Conshohocken, PA 19428-2959. United States1assigns a numeric value for each library entry demonstratinghow similar the two spectra are.3.1.4.1 DiscussionThere are several methods for assign-ing hit quality values and either a high or low value can beassigned as the
17、best match. Refer to the software manufacturersdocumentation.3.1.5 normalizationthe mathematical technique used tocompensate for an intensity difference between two spectra(see 5.1).3.1.6 peak searchingthe process whereby the peak tableof the spectrum of an unknown material is evaluated against alib
18、rary of peak tables. Each reference spectrum in the librarycontains a peak table and the peak table is individuallycompared to the peak table of the unknown, and assigned anumerical value as to the goodness of fit.3.1.7 reference spectruman established spectrum of aknown compound or chemical sample.
19、3.1.7.1 DiscussionThis spectrum is typically stored inretrievable format so that it may be compared against thesample spectrum of an analyte.3.1.7.2 DiscussionThis term has sometimes been used torefer to a background spectrum; such usage is not recom-mended.3.1.8 search algorithmthe mathematical for
20、mula used tomake a point-by-point comparison of two spectra.3.1.9 spectral librarya collection of reference spectrastored in a computer readable form, also called a library,database, or spectral database.3.1.10 spectral searchingthe process whereby a spectrumof an unknown material is evaluated again
21、st a library of digitalreference spectra. Each reference spectrum in the library isindividually compared to the spectrum of the unknown, andassigned a numerical value as to the goodness of fit. To performthis comparison, each data point in the unknown spectrum iscompared to each corresponding point
22、in the reference spec-trum.4. Theory4.1 Beers LawOne of the basic principles that makespectral searching possible is Beers Law (see TerminologyE131), which states that A = abc, where A is the absorbance, ais the absorptivity, b is the sample pathlength, and c is theconcentration of the analyte of in
23、terest. As long as Beers Lawapplies, two spectra of the same material recorded undersimilar conditions can be made to appear the same by normal-ization of the data.NOTE 1In an ideal case, this is true for transmittance spectra, but thereare differences in the spectral peak intensities when reflectan
24、ce spectra arecompared to transmittance spectra.5. Spectral Data Pre-Treatment5.1 Normalization:5.1.1 Normalization of spectra compensates for the differ-ences in sample quantity (concentration or pathlength, or both)used to generate the reference spectra in the library and that ofthe unknown. The s
25、pectra are normalized over the completespectral range of the library. When searching less than the fullspectral range of the library, the spectra must be re-normalizedover the new range before an accurate comparison can bemade. Normalization of a spectrum for library searching is atwo step process.
26、First, the minimum absorbance value in theselected spectral range is subtracted from all the absorbances inthe same range. The resulting values are then scaled bydividing by the maximum result value in the range. The endresult is a spectrum (or a sub-range portion of a spectrum)where the minimum val
27、ue is zero (0) and the maximum is one(1) absorbance. If the range chosen for normalization has onlyone or two strong bands and a few medium intensity bands, therange of the spectrum must be reselected or the spectrum willbe dominated by the strong bands in the spectrum and the HQIwill be insensitive
28、 to weaker fingerprint bands necessary foridentification of a specific compound. Successful compoundidentification may require the spectral match exclude thestrongest bands, then the normalization will be based on amedium intensity band, and weak fingerprint bands will beemphasized in the HQI.5.2 Da
29、ta Point Matching:5.2.1 The algorithms used for searching a spectrum againsta library use a calculation that mathematically compares thedata points of the spectrum being searched to the data points ofthe spectra in the library. This requires that the data points inboth the sample and library spectra
30、 occur at the same fre-quency. If the data points in the sample and library spectra arenot aligned in this manner, then one of the spectra must bemathematically altered (interpolated) to make the data pointsmatch. Typically the unknown spectrum being searched isaltered to match the data point spacin
31、g of the spectra in thelibrary.5.2.2 Data point matching is commonly accomplished usinga linear data point interpolation method. In this method, theslope and offset of a line segment is calculated between theabsorbances of every pair of data points in the spectrum.Anewset of absorbances is calculate
32、d by locating the values thatoccur on the line segments at positions corresponding to thedatapoint frequency of the library spectrum.6. Conditions or Issues Affecting Results6.1 Spectral quality is one of the primary conditions orissues that can affect search results. There is no substitute fora car
33、efully recorded spectrum. There are several conditions orissues that affect spectral quality as pertains to spectral search-ing. These conditions or issues apply to both the spectra usedto create the reference database and to the unknown spectrum.6.2 Baselines:6.2.1 A flat baseline is preferred for
34、the Euclidean distancealgorithm as the Euclidean distance algorithm compares eachdata point in the unknown spectrum to the corresponding datapoint in the reference spectrum. The effect of an offset or slopein the baseline is interpreted as a difference between the twospectra. Therefore, when a spect
35、rum with a sloping baseline oroffset is evaluated using the Euclidean distance algorithm, asimple baseline correction should be used.NOTE 2Negative bands can also produce an offset in the baseline asa result of the data normalization process.6.2.2 The first derivative Euclidean distance algorithmmin
36、imizes the effect of an offset or sloping baseline. In thisE2310 04 (2015)2algorithm, the comparison is made between the difference of apair of adjacent points in the unknown spectrum to thedifference between the corresponding pair of adjacent points inthe reference spectrum. In effect, this causes
37、the first derivativeEuclidean distance algorithm to look only at the differences inthe slope of adjacent data points between the two spectra. Fig.1 shows how the two algorithms view the same two spectra.NOTE 3The first derivative algorithm converts a sloping baseline intoan offset that is then elimi
38、nated by the normalization procedure.6.3 Sample Purity:6.3.1 The physical state of the sample should be as close aspossible to the physical state of the reference materials used toobtain the library. For example, a pure liquid sample wouldideally be searched against a library of spectra of only liqu
39、idreference materials. A sample which is probably a mixture,such as a commercial formulation, should be compared to alibrary of commercial formulations.6.3.2 In some cases the nature of the sample may not be wellunderstood. An unknown sample may be a pure material or amixture. It may have additional
40、 contaminants that will affect itsspectrum by adding spurious bands. In addition there areseveral other sources of spurious spectral features that mayappear as either positive or negative bands. Several of these arelisted below:36.3.2.1 Features due to variations in the carbon dioxide orwater vapor
41、levels in the optical path,6.3.2.2 Bands from a mulling agent,6.3.2.3 Halide salts used as window material and as thediluent for both pellets and diffuse reflection analysis oftencontain contaminants such as adsorbed water, hydrocarbon,and nitrates. Always use dry halide salts and keep unusedhalide
42、salts in a desiccator,6.3.2.4 Water can alter the spectrum of the sample from itsdry state. Spectra of inorganic samples with waters of hydra-tion are particularly sensitive to adsorbed water,6.3.2.5 Solvent bands from samples run in solution, and6.3.2.6 Bands from solvents left over from an extract
43、ion orfrom casting a film from a solution.NOTE 4Retain spectra of any solvents used, so that bands due to thesolvent can be identified in the spectrum of the unknown.NOTE 5If the solvent bands in a region of the spectrum cannot beremoved from the spectrum (by either re-recording the spectrum, using
44、anuncontaminated sample, or by spectral subtraction using the solventreference spectrum), then that region of the spectrum should be excludedduring a search. It is not sufficient to remove the offending bands digitallyby drawing a straight line through the region before the search. The searchalgorit
45、hm will calculate a poor match in this region for any referencespectrum containing features in the region. It should be realized that theremoval of the solvent bands may also remove underlying features in thesample spectrum.6.4 Absorbance Linearity (Beers Law):6.4.1 A spectrum recorded using good pr
46、actices (see Prac-tices E334, E1252, E1642, E2105, and E2106) should followBeers Law, and so maintain the relative absorbance intensitiesof its bands, independently of sample thickness.As long as thisratio between the bands is maintained, the spectra can benormalized and a good comparison between sp
47、ectra can be3Coleman, Patricia B., Practical Sample Techniques for Infrared Analysis, CRCPress, FSBN# 0849342031: 8/26/93.The bottom two spectra demonstrate the results of the 1st derivative of a spectrum with a sloping baseline as compared to a spectrum with a flat baseline.The two spectra in the b
48、ottom trace are almost completely overlapped.FIG. 1E2310 04 (2015)3made. For a spectrum to meet this requirement, each ray oflight of a given frequency must pass through the same amountof sample. There are at least two general cases where this maynot happen.6.4.1.1 One case occurs when there is an u
49、neven thicknessof sample in the beam. For example, if the sample is wedgeshaped in thickness, or irregular in shape, some rays of lightpass through the thin part and some rays pass through thethicker part of the wedge. A similar concern arises whenmaking KBr pellets for analysis. Unless the powder is carefullyspread in the pellet die, the pellet can be pressed with a densitygradient across the diameter. The sample must also be evenlydistributed by thorough mixing of the sample and pellet matrix.This is of particular concern when the beam geometry issmalle
