Critical analysis of historical British cement chemistry data

Historical Context

Back in the good old days, when I was responsible, among other things, for teaching cement chemistry to cement industry people, I used to start (and finish) with the simple statement:

"It's all chemistry!"

To ameliorate the effect of this statement, which strikes terror into the hearts of all non-chemists, I would add:

"If you need any chemistry doing - find a chemist! (Note 1)"

The presence or absence of a chemist when one was needed had a considerable effect upon the fitful progress of the cement industry. In particular, the early adoption of chemical control principles in France, Germany, the USA, and, in fact, almost everywhere except Britain, was largely responsible for the rapid eclipsing of the previously-dominant British cement industry towards the end of the nineteenth century.

To the historian, trying to track the early development of the industry, the almost total lack of reliable contemporary chemical data is a major obstacle.

Historical cement chemistry data are important because they explain much of the evolution of cement physical properties over the last two centuries, and also because they are needed to trace the environmental effects of the industry. However, the interpretation of the historical data is extremely difficult. This is because, although the components of a chemical analysis appear to be very straightforward and unchanged since chemistry developed from Dalton's atoms (1803), the methods of getting at them and understanding them have changed enormously.

Perhaps unfairly, it is worthwhile remembering the first professional chemical analysis performed on a sample of Portland cement. I. C. Johnson, tasked in 1843 with emulating William Aspdin's cement, realised (as any modern scientific researcher would) that the cement's chemical composition was the key information needed, so he sent a sample to famous consulting chemist Dr Andrew Ure (1778-1857), who reported that the main constituent (45%) was phosphate of lime, which for a while led Johnson down a blind alley, assuming the product was made by burning bones. At the time, the mistake was easy enough to make.

Johnson's further experiments - ultimately successful - used no chemistry at all, and the subsequent growth of the Portland cement industry in Britain involved only rule-of-thumb mixing and burning of materials. Because the "classical" raw materials - soft chalk and river mud - were universally and freely available, new entrants had no need to find out more than the rules of thumb. It is for this reason that serious chemical investigation of cement began outside Britain. In Germany, in particular, raw materials similar to those used in England were hard to find, and it was necessary to determine chemical composition in order to derive new rules for employment of locally available materials. By the time that the need for chemical knowledge had become apparent in Britain (Note 2), German chemists already had decades of experience in cement analysis. British knowledge continued to lag behind thereafter.

Before considering the techniques of analysis, it is worthwhile to establish what we mean by an accurate chemical analysis. Every mineral consists of atoms, each of which is identifiable (by its atomic number) as one of a fairly limited set of chemical elements. The ideal analysis consists of enumerating the atoms of each element in the analytical sample. One can conceive of a technique in which atoms are removed from the sample, one by one, identified, and the count of each type reported as the final result (Note 3). This technique could be described as an "absolute" analysis, and insofar that other techniques produce a different result, that difference can be termed their bias. This definition of bias is different from that currently used in the cement industry; bias is used to signify the systematic offset between the result obtained by a given method and the certified value of a "consensus" standard reference material. The users of a consensus standard tacitly agree to disregard the deficiencies of the standard and its certified values. The certified values of the reference material are of course also subject to bias in that they were analysed by "consensus" methods rather than "absolute" methods.

Scope of cement analysis

Another fundamental point to understand is the number of elements that might appear in a cement analysis. Portland cement is made from natural raw materials that are routinely available, and differs in composition from the average composition of continental rock only in that the calcium content is artificially enhanced. To see what elements are to be expected in the composition of cement, it is worth while to see what we would get if we started with the composition of "Average Continental Rock", as determined in many published studies. First express it as oxides. Then remove volatiles - H₂O, CO₂ and inert gases. Then reduce partially volatile elements - alkali metals and halogens - by - say - 62%. Then add CaO to the mixture until an LSF (Note 4) of 95% is obtained. The result is a clinker that has oxides (%) in descending order thus:

This is a perfectly convincing clinker composition with a silica ratio 2.60 and alumina ratio 2.07. All the oxides are of the normal order of magnitude (except for the sulfur, which is enhanced by absorption of fuel sulfur). So the subsequent sequence (oxides occurring at 10 ppm and over) shows the likely order of minor elements (in ppm):

This order is also borne out in practice, except that SrO is usually 2-10 times higher, because it accompanies calcium, and there is much more ZnO if zinc-filled polymers (notably tyres) are being used as fuel.

Historical methods

The classical technique for determining the chemical composition of inorganic materials was gravimetric analysis. For mineral analysis, this involved systematically separating each element as its oxide, one by one. Gravimetric analysis was the only technique used initially, for a number of reasons:

• The technique was intuitive, the separated components being visible physical substances in the bottom of a crucible.
• In addition to simple laboratory glassware and platinum crucibles, the only equipment required was a balance and a Bunsen burner.
• Only simple chemical reagents were required, at a time when pure forms of more complex reagents were not yet available.

A key point about historical data is that there was no single "standard method" for chemical analysis in Britain until the British Standards Institution (BSI) published one in 1970. Now of course, every cement company (and, if the truth were known, every analyst) would indignantly claim that they had their own "standard method", but without a central authority, the situation was anarchic, right up to that date. And what's more, the further back before that date that one goes, the more anarchic it gets. Reputable consulting analysts such as Butler give example analyses in their books, each presented in slightly different format, indicating the ad hoc way in which methods were applied to each sample.

Actually, the first cement standard to give (and, in fact, mandate) a standard analysis method was that of ASTM, which in 1912 added an appendix to the Portland cement specification, saying:

. . . this method, including all precautions as stated in footnotes and italicised text, must always be followed when the results are to be used as the basis for rejection, or when an accurate knowledge of composition is desired.

That the American standard was the first to do this says a lot about the leap to technical dominance of the US industry in the first decade of the 20th century. That the US prescribed methods changed little in the ensuing century also speaks volumes.

A significant contribution to the exacting method given came from W. F. Hillebrand, who had in 1910 published USGS Bulletin 422, The Analysis of Silicate and Carbonate Rocks, which discussed the matter in enormous detail, and included some remarks that are extraordinarily pertinent in the context of the current discussion. For instance:

How little understood may be the principles underlying the treatment of bodies so complex and the accurate separation and determination of their constituents, even when these are comparatively few in number, has been strikingly shown during recent years in the work of several committees of chemists charged with the investigation of the methods employed in various lines of technical chemistry involving the analysis or assay of zinc ores, slags from the smelting of copper ores, argillaceous limestones, and cements. In all cases a most woeful inability to obtain agreeing results is apparent, not only among those less experienced, but also among those supposed to be most expert in each of the particular fields as well. (My italics: see Note 5)

Such remarks must have caused a frisson of embarrassment in the cement industry, and local methods were certainly sharpened up during the 1920s, but the inherent problems in the analysis technique remained, the seven days or more required for a good analysis being more than most cement industry people could stomach.

The immediate problem with gravimetric analyses is that each separation process, intended to isolate a single element, is inefficient and subject to interference from other elements present. So, for instance, an element in solution is caused to precipitate out: the precipitate is filtered out, washed and weighed. But not all the target element may be precipitated, and some may be washed back into the solution. Other elements may be "carried down" with the precipitate. Thus a biased result is obtained - although the direction of the bias can be indeterminate.

A case in point is the determination of the most important element in Portland cement - calcium (in the form of the oxide). Cement dissolved in acid is first freed of silica, then of elements precipitated by ammonia ("R₂O₃": alumina, iron oxide, etc). The resulting aqueous solution should now contain only calcium, magnesium, alkali metals, sulfates and a few other minor elements. Calcium is now precipitated as the oxalate. The oxalate may be dried and weighed as such, or (more usually) calcined and weighed as the oxide. The principle of the separation is that there are no other elements with insoluble oxalates present, and the other cations present (including magnesium and alkalis) all have soluble oxalates. But the silica and R₂O₃ are in practice never completely removed in the previous stages, and appear in all the subsequent precipitates. Although magnesium oxalate is soluble, nevertheless magnesium is always present in solid solution in the calcium oxalate phase, and no amount of washing will remove it. Alkalis also are incorporated to a limited extent. All the strontium and barium are included. Washing the oxalate on the filter removes much of the foreign material, but also re-dissolves some of the calcium - in fact, quite a lot. Instructions for "high precision" analysis suggest that the solutes should then be concentrated, and the remaining calcium (in fact, most of it) precipitated again as oxalate. How often this was done in historic analyses is open to question - were they intended to be "high precision"?

This is presented as exemplifying the problems involved. In fact, historic calcium determinations (on cement) are among the more reliable, if only because positive and negative biases more or less cancelled out. (However, the calcium data for raw materials were often inexpertly "calculated" from calcimeter data.) Many of the other "separations" in the gravimetric analysis scheme were subject to much greater errors, sufficient to make reported results look odd to the modern cement chemist. The most glaring errors were in iron oxide and in alkalis. These were the result of particularly poor separation processes.

Alumina and iron oxide were frequently reported together as R₂O₃, either because the distinction was not thought important, or because the chemist recognised the poor reliability of the result. Iron was obtained from the R₂O₃ by selective dissolution of alumina using strong alkali, but this process was never complete and high iron oxide values were normal. A better method for iron was to reduce Fe(III) to Fe(II) and titrate with an oxidising agent - permanganate or dichromate. This became available from the 1870s, but such was the antipathy to volumetric methods that the selective dissolution method was still commonly in use in the 1920s.

As for alumina, it was as late as the 1950s before a specific separation was available - and that was poor - and only with the advent of XRF were reliable alumina values obtained. Before around 1969, published alumina values were usually obtained by difference, and usually contained also TiO₂, P₂O₅ and Mn₂O₃, which were rarely separately analysed. In the current (2021) ASTM standard method, C114-2017, alumina is still determined by difference.

Alkalis in old texts are very often valueless, and a knowledge of modern values of the same materials is the best guide. Alkalis were analysed in the classical scheme by eliminating (it was hoped) all other solutes, so that only alkali chlorides remained. In practice, the remaining solutes contained all the accumulated cations that had leaked through the previous precipitation processes. These dried chlorides were weighed as such. Potassium was then derived by precipitating potassium hexachloroplatinate, this being less soluble than the sodium salt, and the sodium obtained by difference. The obvious problems are that the separation is incomplete, and because in any case the mixed chlorides contain other cations, the sodium is massively over-estimated.

Alkali determination remained very unsatisfactory until the 1960s. Flame emission photometry became available in the 1950s, but was not taken up for ten years. This gave good potassium data, and sodium was much better, though still subject to interferences. While XRF has revolutionised accuracy for all elements from magnesium upwards, sodium remains an exception, the only moderately reliable technique being atomic absorption spectrometry, which (to my knowledge) is nowhere used, so if 0.01% precision is expected, then all sodium data, without exception, is unreliable.

Systematic classical analysis

A flavour of the complexity and labour involved in accurate classical methods can be obtained from the schema in the current (2021) ASTM standard method for analysis of cement (C114-2017), summarised as follows:

Note: acid is hydrochloric unless otherwise stated. Ignitions are performed at around 1000°C in platinum crucibles. Filtrations usually require individual specified washing regimes.

1. Determine silicon dioxide by dissolving the cement in acid, and digest at 100°C to precipitate silica. Filter off the gel, ignite, weigh, then volatilise the silica with hydrofluoric acid, ignite and re-weigh. The difference in weights gives the silicon dioxide content. Fuse any residue not volatilised with potassium pyrosulfate, dissolve and add to the filtrate.
2. Determine the "ammonium hydroxide group" (R₂O₃) by neutralising the filtrate from (1) with ammonium hydroxide. Filter off the precipitate, then re-dissolve it in acid and precipitate again with ammonium hydroxide. Filter off the precipitate and ignite: the weight is the R₂O₃ content.
3. Determine iron(III) oxide by dissolving the cement in acid, reduce it with tin(II) chloride, then titrate with standard potassium dichromate(VI) solution using barium diphenylamine-4-sulfonate as indicator.
4. Determine phosphorus(V) oxide by dissolving the cement in acid, filter and make to volume in a volumetric flask. Treat an aliquot with standard ammonium molybdate solution; boil, cool and make to volume. Read the absorbance at 725 nm with a spectrophotometer calibrated with potassium dihydrogen phosphate standards.
5. Determine titanium(IV) oxide by adding tiron (disodium-1,2-dihydroxybenzene-3,5 disulfonate) to an aliquot of the cement solution from (4) with a buffer and making to volume. Read the absorbance at 410 nm with a spectrophotometer calibrated with TiO₂ solutions.
6. Determine aluminium oxide by subtracting the Fe₂O₃, the P₂O₅ and the TiO₂ values from the R₂O₃ value (Note 6).
7. Determine strontium oxide. This is essential in order to determine calcium oxide, but no method is given (Note 6).
8. Determine calcium oxide by removing manganese from the combined filtrates from (2). Make the solution slightly acidic and add ammonium oxalate solution. Heat to 75°C and neutralise with ammonium hydroxide, allowing calcium oxalate to precipitate as the solution cools. Filter, washing to remove excess oxalate, then dissolve the precipitate with sulfuric acid and titrate with standard potassium permanganate solution to obtain a "CaO+SrO as CaO" value. Deduct the molecular equivalent of the known strontium oxide from this to obtain the CaO value.
9. Determine magnesium oxide content by concentrating the filtrate from (8), and adding ammonium dihydrogen phosphate and ammonium hydroxide to precipitate magnesium dihydrogen phosphate. Filter and ignite the precipitate to magnesium pyrophosphate. MgO is 0.362 times the weight of this.
10. Determine sulfur(VI) oxide (equivalent to sulfate ions present) by dissolving the cement in acid, filtering off insoluble matter, and adding barium chloride solution to the boiling solution. Digest this for 12-24 hours, then filter off the precipitate of barium sulfate, ignite and weigh. SO₃ is 0.343 times the weight of this.
11. Determine sulfide by acidifying the cement and absorbing the hydrogen sulfide produced in ammoniacal cadmium chloride solution. The cadmium sulfide precipitated is titrated with standard potassium iodate solution.
12. Determine loss-on-ignition by igniting the cement at 950°C to constant weight, and calculating the loss in mass.
13. Determine sodium and potassium oxides by dissolving the cement in acid and filtering off insoluble matter. Make up the solution to standard volume and aspirate through an atomic absorption or atomic emission spectrometer, at the standard (unspecified) wavelengths for each element. The instrument is calibrated using standard solutions made up from pure chemicals.
14. Determine manganese(III) oxide by dissolving the cement in nitric acid, then boiling with added sodium nitrite solution and filtering off insoluble matter. Add sodium bismuthate(V) followed by dilute nitric acid and filter off the manganese(IV) oxide. Titrate with standard sodium arsenate(III) solution.
15. Determine chloride by dissolving the cement in nitric acid. Filter and titrate with standard silver nitrate solution, monitoring the end-point by chloride-selective electrode.

This daunting process, in addition to a large number of off-the-shelf reagents, also requires some 63 custom-strength reagent solutions, including 7 different dilutions of HCl. The instructions (for an already experienced analyst) occupy 20 A4 pages. It is inconceivable that the entire process can be completed to the required standard by one analyst in less than a week.

Subsequent methods

The speed (if not the precision) of analyses was improved from the 1960s onward by the use of colourimetric methods and EDTA titration of Ca and Mg, but historically this had little effect because its introduction overlapped with the introduction of the XRF technique, reducing the time for a high-precision analysis to about 15 minutes.

With the notable exception of sodium, there is no reason (in principle) why highly accurate and precise data should not be produced by the average cement plant armed with an XRF spectrometer.

I did say "in principle". XRF spectrometry, with the results processed by "fundamental parameter" modelling, can be as nearly definitive as any relatively cheap and rapid technique can be. However, in the absence of complex data processing, the x-ray intensities measured by XRF are subject to complicated inter-element interferences. The necessary corrections for these effects were already well-known by the end of the 1960s, but were not necessarily adopted. My experience of the techniques actually employed are limited to those of pre-2002 Blue Circle (60% of the British industry), and there, to the best of my knowledge, corrections were made only at the Greenhithe Research Department. Just why this was the case, is in my expert opinion because of a lack of mathematical proficiency among those who made policy in this area. This went hand-in-hand with a distrust of the computers which were necessary for doing the calculations.

The use of uncorrected XRF data naturally resulted in a distrust of XRF when it was first introduced. I do recall, in the mid-1970s, politely questioning the accuracy of an analysis produced by the Northern Area Technical Services (NATS - a "reference" laboratory). "Oh no," said the head of the laboratory section, "It's definitely right - it's colourimetric, not XRF!" The XRF technique used on all plants was based on the philosophy that, for each material analysed (e.g. limestone, shale, rawmix, clinker, cement), provided that the same materials were in use every day, a simple straight-line calibration, using standard materials gathered on the plant and analysed by "definitive" methods, was all that was required for "routine" analysis. Of course, this meant that a service lab such as NATS had to maintain hundreds of different calibrations to cover all the plants and materials they encountered. They had no computer! Suggestions that they might get one were met with a snarl of contempt.

On an individual plant, the situation was not much better. At Humber, of which I had a mercifully brief 25 month experience, an XRF (a Telsec - later Oxford Analytical Instruments - TXRF 8-channel simultaneous dispersive instrument) had been obtained in 1973 in order to facilitate SRPC manufacture. It was provided with a DEC PDP-8 mini-computer, and inter-element correction software provided by John Lucas-Tooth, owner of Telsec. Under the direction of NATS (there being no expertise on the plant), the software was discarded, and replaced with a "cement works friendly" locally-written package. "Straight-line" calibrations (in quotes because, due to uncorrected absorption effects, none of the lines actually was straight) were produced using locally-gathered calibration materials analysed "definitively" by NATS. There were, for instance, four separate clinker calibrations: Kilns 1&2 OPC, Kilns 1&2 SRPC, Kiln 3 OPC, and Kiln 3 SRPC. This was necessary because the full range of clinkers would not fit onto a common calibration line. (It should be pointed out that all Portland cement clinkers, prepared for analysis in the same way, will fit very tightly onto a common calibration line if the results are properly processed). It was not uncommon to find clinker samples which, when analysed on an OPC calibration, were found to be SRPC, and when analysed on an SRPC calibration, were apparently OPC. It is clear from this that XRF data from the period was of low quality. This was not the fault of the XRF technique - it was caused by the incompetence of the operators, and the policy-setting hierarchy.

If I seem somewhat over-exercised by all this, it's only because, from 1973 onwards, I was trained and locked into the "Blue Circle Way" of doing things, and I had to wait until 1994 to get my first experience of proper XRF software, with no-one breathing down my neck. On applying Fundamental Parameter corrections, I saw my scattered calibration lines collapse effortlessly into a tight fit of state-of-the-art precision. Lines with virtually no scatter, passing through the origin. I resent the long delay and the waste of so much time and effort on the production of mediocre results for 21 years, while being hounded to achieve a degree of chemical control that was never possible with poor analytical data. Maybe others didn't care so much.

Anyway, since 2002, the situation no longer arises. Everybody, using the standard technique, gets definitive results by XRF.

Processing historical data

Underlying this website is a database quantifying the output of clinker throughout the industry, for each identified kiln. It was desirable to extend this to include the likely chemistry of the product for each record. The emphasis has therefore been on tracking clinker analysis. Large numbers of analyses are scattered around the historic literature, and the objective is to interpret these in the light of modern knowledge, and adjust them to give the results as if analysed by a modern proficient system.

Reviews of historic chemistry have been attempted before. Earlier writers on Portland cement, while tracking development of physical properties, took little interest in trends in chemistry - in fact it would not have occurred to them that there was any trend; the "formula" was set in stone. After WWII, it became apparent that there was scope for change in chemistry, and that change was under way. Initially, in looking back at old data, they were simply taken at face value. In 1962, Alec Skempton drew together numbers of published analyses from the mid-19th century. His earliest data-set was three analyses (believed to be of White's cement) from the period 1849-1852. Inevitably, they were all produced by German analysts, as no analytical chemistry was done in Britain at that time, while German chemists were busy trying to find the "secret" of Portland cement. Skempton summarised these as:

The use of averaging was judicious; the individual analyses were actually as follows:

Here we confront a number of basic problems of interpretation of such data.

• The three analyses present three different sets of analytes
• None of them add up very close to 100%; how much of the deficit is other analytes, and how much is error?
• The main source of iron in the rawmix was the clay. Medway clay had variable iron due to its variable content of pyrite - but not that variable!
• The first two don't give MgO: was it included in the CaO?
• The first and third give sulfur as CaSO₄; the second does not present a value - is the lime content of the CaSO₄ included in the CaO or not?
• The second gives K₂O but not Na₂O - is the sodium included in the K₂O?

To a modern cement chemist, the first looks as though it is not Portland cement at all. And yet it's not inconceivable that these three may all be the same cement.

Clearly, these early analyses were subject to the defects of technique mentioned above. In addition to the errors that arise from imperfect analysis - which may at least be fairly consistent - there are errors due to unspecified methods of calculation - which are anybody's guess. As an example, here's an analysis of Warwickshire Lias clay quoted in Butler's 1899 edition (p 326):

How were the two carbonate values arrived at? Perhaps the clay was fused, and CaO and MgO determined gravimetrically, then re-calculated as carbonate. However, for raw materials, the calcimeter, being quick and relatively accurate, was the method of first resort. The calcimeter actually gives a value of CO₂ content. This can then be converted by calculation (divide by 0.4397, or whatever antediluvian factor the analyst favoured) to a CaCO₃ value. But of course it doesn't distinguish between different carbonates - you just get a total CO₂ value. So maybe the MgO was determined gravimetrically, the associated CO₂ calculated and deducted from the total, and the remaining CO₂ calculated as CaCO₃. The analysis was almost certainly done this way. A word to the wise - lias materials contain only trace amounts of MgO in the carbonate phase. The MgO exists almost entirely as aluminosilicates. So the CaCO₃ value given as 30.2% is actually 38.3%. Even so fundamental a value as the calcium carbonate content is badly wrong, purely due to the ignorance of the analyst. Of course, Butler's chemically-naive clients would not have dreamt of questioning this result.

The early cements (such as those in the German analyses) were made from chalk and Medway mud, and the main source of variation of the product was in the proportion of each in the mix. Proportioning, while said to be of critical importance, was mostly volumetric, occasionally by weight. The chalk varied mainly in its water content - perhaps 20% in the spring, and 15% in late summer. The clay was much more variable - also mainly in water content - varying from 30 to 60% water, 40% being typical. Quality control consisted of burning some of the blended mix in a "trial kiln", and judging the resulting clinker purely in terms of its physical properties.

It can be seen that, with a knowledge of the chemistry of Thames-side chalk and Medway mud as analysed by reliable modern methods, a fair knowledge of the likely range of analyses can be obtained, and any deviation of historical data from this range is likely to be due to analytical error. Hopefully, the average typical biases of a nineteenth century analyst can be deduced in this way.

This article is still in development, and will be much longer.

NOTES

Note 1. These remarks were made in the context of a cement industry that, over the last 50 years, has systematically replaced chemists with engineers, on the grounds that anyone who is "a bit technical" can operate an XRF and generate analytical data. Whether they can tell a good analysis from a bad one is another matter. Well, let's be clear - the empirical fact is, they can't!.

Note 2. Turley in his account of the Casebourne plants of Hartlepool and Billingham (subsequerntly run by ICI!), notes with pride that they employed their first chemist in 1897!

Note 3. Until recently such a technique was "pie in the sky", but now laser ablation inductively couple plasma mass spectrometry (LA-ICP-MS) approximates to it quite closely

Note 4. Lea and Parker Lime Saturation Factor: LSF = (100 CaO - 70 SO₃) / (2.8 SiO₂ + 1.2 Al₂O₃ + 0.65 Fe₂O₃). Because it is defined in BS12, the coefficients of this equation are not up for discussion.

Note 5. There can be little doubt that Hillebrand intended the internationally-known experts David Butler and Bertram Blount to identify themselves in this remark.

Note 6. Strontium is usually present at less than 0.1%, but can occasionally rise to 1% or even 2%. It is usually associated with the use of aragonite (or converted aragonite) limestones. In cement, it acts as a lime replacement, raising the "lime saturation" pro rata. In analytical techniques, it behaves as if it were calcium. Historically, its presence was ignored, since it was a moot point whether a Group II atom was calcium or strontium. However, with element-specific instrumental techniques such as XRF, undetermined strontium may have a significant effect upon the cement. If classically-determined values are to be used to validate element-specific instruments, the two elements must be distinguished. Strontium can be readily determined with high accuracy by Rh-target XRF, using the tube Compton line as an internal standard.