Rust and the sound of steel bells

In my article written last year on the acoustics of steel bells, I suggested that although it is often said that rust affects the tone of steel bells, in practice this seems not to be the case. I have observed that Bochumer steel bells of the 1950s and 1960s have true-harmonic partials when measured today, presumably little changed from when they were cast and tuned (though often Bochumer steel bells were painted to protect them). Academic research suggests that once steel has acquired a surface layer of rust, subsequent corrosion takes place very slowly provided the steel is kept dry. This note describes actual measurements on the peal of Naylor-Vickers steel bells previously as Bassaleg which confirms that their distinct tone is not due to rust.

Since the first version of this article, Matthew Higby has provided more information on the removal and subsequent history of the bells, and Dave Kelly has provided an additional set of recordings.

The bells of St Basil, Bassaleg, Newport

The church of St Basil, Bassaleg

Until 1999, the church at Bassaleg had a peal of six Naylor Vickers steel bells cast at various dates between 1862 and 1866. In 1998 they were replaced by a peal of 8 bronze bells.

The bells were removed by Matthew Higby, and subsequently acquired by David Potter. They are now preserved in the Millennium Gallery in Sheffield as an example of that city’s heritage in steel production. The bells had some remnants of original paint when they were removed, and were subsequently shot blast cleaned, but were rusty again a few days later.

The bells are not painted and have the surface layer of rust typical of steel bells.

The bells in the Millennium Gallery, Sheffield

I have three sets of recordings of these bells. The first was made when they were at Bassaleg by Derek J Carr in 1993 for the centenary of the Llandaff and Monmouth Diocesan Association. There are two copies of this recording on youtube. One is the full recording of a touch of Grandsire Doubles, and the other a short extract of the full recording made by Nick Bowden – the Basseleg recording starts 5m34secs into the second video. Dave Kelly provided me with a set of recordings taken in 2001 at the Millienium Gallery. Finally, with the kind permission of the Gallery I was able to take good quality recordings of the individual bells in the museum on 13 March 2017.

Assuming the bells were installed in 1866, the 1993 recording took place 127 years after installation. The 2001 recordings were taken 8 years later, after the bells had been shot-blasted. The elapsed time from the 1993 recording to the ones I took in 2017 were 24 years after the previous recording, and 151 years after installation, a period of time representing 15% of the whole lifetime of the bells. As the 2017 recordings were the best in quality, the tuning figures from the older two sets of recordings are compared with the most recent. Many of the partials have doublets, and the frequencies identified in the 1993 recordings were used to guide which half of the doublet to select in the 2017 and 2001 recordings.

Partial frequencies from the 2017 recordings

The figures are presented both as frequencies and as cents from the nominal in the normal fashion:

BellHumcentsPrimecentsTiercecentsQuintcentsNominalcentsS’quintcentsOct NomcentsI7I8I9I10
1441.0-2045.7743.5-1141.4880.1-849.41253.5-237.11437.5942.12152.5698.92982.01263.33904.54929.06012.0
2357.3-2172.9637.5-1170.6748.7-892.21039.2-324.61253.5705.01907.0726.42666.91307.03517.2
3326.4-2142.6561.5-1203.4674.5-885.9940.6-310.21125.2518.11711.2725.82390.51304.63146.53977.5
4320.0-2120.2536.7-1225.0648.0-898.7915.4-300.61089.0461.41664.4734.42325.11313.13065.4
5276.2-2127.7491.0-1131.7568.6-877.6817.6-248.9944.0214.11423.5711.11985.01286.72616.03306.04053.54838.5
6248.5-2096.6430.3-1146.1499.2-888.9734.4-220.6834.20.01273.1731.91784.51316.52356.42982.03645.24352.5

These figures show the usual Naylor-Vickers deficiencies: nominals across the bells up to half a semitone out, grossly sharp hums and upper partials, and primes that are sometimes rather sharp. The question to be investigated is whether these deficiencies are down to the original design and shape of the bells, or whether they are caused by metal loss through rust over the past 151 years.

Partial frequencies from the 2001 recordings

The recordings provided by Dave Kelly also include a 440Hz tuning fork as a frequency reference and the figures have been adjusted to bring this reference to exactly 440Hz. The recorder was running about 0.95% or 16.3 cents slow on the recording of the fork.

Two sets of figures are presented in the tables below: the actual frequencies measured from the 2001 recordings, and the difference in cents between each frequency in the 2001 recordings and the equivalent frequency measure in the 2017 recordings:

Frequencies:

BellHumPrimeTierceQuintNominalS’quintOct NomI7I8I9
1441.4746.2880.91253.81438.12154.32982.73905.3
2357.5638.8749.71040.81254.41907.82665.1
3326.7561.9673.6941.91126.01712.82392.43152.73980.54858.8
4321.6537.1647.3915.11090.31665.12327.43061.4
5276.2491.0569.2816.7944.71424.91982.92612.1
6248.3430.1498.9733.9833.91273.51783.62354.62973.03639.3

Difference in cents from the 2017 recordings:

BellHumPrimeTierceQuintNominalS’quintOct NomI7I8I9
1-1.4-1.7-1.6-0.4-0.7-1.5-0.4-0.4
2-0.8-3.6-2.2-2.7-1.3-0.71.1
3-1.5-1.2-0.1-2.4-1.3-1.6-1.3-3.4-1.3
4-1.2-1.21.90.5-2.0-0.7-1.72.3
50.0-0.1-1.71.9-1.3-1.81.82.6
61.11.01.01.10.7-0.60.91.35.22.8

Given the fact that the recorder was running 16 cents slow, I think there is good agreement between these measurements and the ones taken in 2017. In particular, there is no unidirectional trend in the differences. If they were due to thinning of the bell through rust, one would expect all the higher partials (Superquint upwards) to be sharper in the 2017 recordings. In fact, some have moved up a little, some down, and I think the differences are due to the difficulty in taking precise measurements from the recordings.

Tuning figures from the 1993 recording

Getting individual partial frequencies from recordings of bells in changes is difficult. Individual bells ring for much longer than the gap between each bell’s strike, and so the partials of all the bells overlap in the recording, making it difficult to tell which frequency belongs to which bell. Bells have very many partials – many more than in the table above. If the short period of sound between each bell strike is analysed, partial frequencies can’t be measured to better than a few cycles per second, often not sufficient to distinguish between frequencies from different bells.

The investigation of this recording took place over a number of months, and included the following steps:

  • An initial measurement of the partials for each bell based on analysing short sections of sound as each bell strikes, having first determined which order the bells were striking in the change
  • Analysing longer sections of sound (to allow accurate partial measurement) at around 20 points in the recordings, spaced a few seconds apart
  • The very large number of frequencies measured from these sections were provisionally allocated to a bell
  • This was done for both of the youtube recordings, and the partials that could definitely be identified were used to calculate the speed compensation for the two recordings (which were at slightly different speeds) against the Millenium Gallery recordings
  • With the speed compensation applied, exact allocation of frequencies to each bell was done (for many more partials tha in the above table). In some cases, the Millenium Gallery recordings were searched for particualr frequencies not found in the initial analysis, including doublet pairs which helped confirm the identification of some partials
  • All the measured frequencies, compensated for speed, and averaged over all the measurement points and across both recordings, were used to create the final figures below

As before, presented in the tables below are the actual frequencies measured from the 1993 recordings, and the difference in cents between each frequency in the 1993 recording and the equivalent frequency measured in the 2017 recordings:

Frequencies:

BellHumPrimeTierceQuintNominalS’quintOct NomI7I8I9
1440.1743.1880.61437.72157.42983.83899.44930.1
2357.3637.4748.51038.91254.21906.52671.5
3326.8560.0674.4940.91124.41712.52390.33152.8
4320.9536.4648.1916.11087.81667.02326.03065.33862.54757.2
5276.3491.6568.9817.9944.31421.61984.52623.73311.84055.3
6248.6431.4499.2734.3834.81273.71784.22353.02986.83638.9

Difference in cents from Millenium Gallery recordings:

BellHumPrimeTierceQuintNominalS’quintOct NomI7I8I9
13.40.8-1.1-0.2-3.9-1.02.3-0.4
2-0.20.30.40.5-1.00.5-3.0
3-2.34.60.4-0.61.2-1.40.2-3.4
4-4.91.0-0.3-1.41.9-2.7-0.70.1
5-0.6-2.2-1.0-0.7-0.62.30.4-5.1-3.0-0.8
6-1.0-4.2-0.10.2-1.2-0.80.32.5-2.83.0

There is no significance in the absolute values of the frequencies, as correction factors have been applied, assuming differences in recorder speed or temperature, to bring the average frequency deviation to zero. As for the 2001 recordings, there is no monotonic trend to the differences – both above and below the nominal, partials in some bells are higher, and some lower. Given the difficulty in measuring partial frequencies from recordings in changes, the minor discrepancies are not a surprise.

The effect of rust on steel bells

The conclusion from this investigation is that rust is not the cause of the tonal deficiencies in the Bassaleg bells. There is no particular pattern to the changes in partial frequencies between 1993, 2001 and 2017 – corresponding frequencies have gone up in some bells, down in others by broadly equal amounts, so the change is as likely to be measurement error as rust or the shot blast cleaning. Taking as an example the hum of the treble, this may (subject to measurement error) have gone up by 3.4 cents in 24 years, which would be a change of 21 cents in 151 years if the change was in proportion – whereas this partial is a full 354 cents sharp of true harmonic. In the same period 1993 to 2017, the hum of the third bell appears to have gone down by 4.9 cents, the opposite direction. In general for all the partials the change from 1993 to 2017 is a small fraction in the discrepancy of the partial from that in a bell with better tonal characteristics. It is clear that the tonal deficiencies in these bells are due to the original design and shape, not rust in the last one and a half centuries.

References and links

More details on the history of the Bassaleg bells from the Landaff and Monmouth website.

The Millennium Gallery website