Thecomplex and demanding structural engineering challenge presented by the steel and glass roof over the Great Court produced a breathtaking solution.
But gestation was far from straightforward.
The perimeter of the Great Court measures 92m by 73m, slightly larger than a standard football pitch. Some 5m north of its true centre stands the 43m diameter round reading room, a cast iron framed structure of uncertain stability. Three of the perimeter walls are weather beaten Georgian stone facades.
All four walls are of slightly different heights; all existing structures are Grade 1 listed. And to minimise the visual impact outside the museum planning consents placed strict limits on the height of the roof structure above the existing roofline.
Structural engineer Buro Happold's original concept featured an almost flat structure made up of heavy steel sections spanning between rocker bearings at the perimeter and a ring of 20 new columns surrounding the Reading Room. Glazing was lightweight ETFE cushions. 'No extra loads could be applied to the reading room at all, and no extra lateral loads to the perimeter structures, ' explains Buro Happold group director Steve Brown.
Furthermore: 'We could only increase vertical loads on the perimeter by 10%. These restrictions still applied after the decision was taken to switch to a glass roof, so we knew that a very efficient roof structure would be needed.'
Some form of lattice structure was the obvious answer. 'But arch action could not be developed, ' points out Buro Happold partner Mike Cook. 'We needed more curvature to make it work as a shell spanning in three directions, even though the height restrictions put a tight limit on how much curvature we could use.'
Early concepts featured a steel tube diagrid design. The ring of columns around the reading room was moved in close to its outer wall and hidden behind new stone cladding. Around the perimeter sliding bearings and a concrete ring beam were pencilled in, supported on and largely hidden by the existing stonework. But there were problems, not least the fixing of the glass panes.
Brown explains: 'The diagrid concept just didn't node out on the perimeter and the tube diameters were far too large. And there was no way the glass could cope with the curvature with rectangular panes. Triangular panes were the only way - which meant a triagrid, with six way nodes.'
Made up of radial elements connected by two opposing spirals, the elegant geometry that was developed promised great structural efficiency as well. But even switching to standard rectangular hollow sections for the steelwork - which simplified glass fixing - could not make the numbers work.
'In the end we had to go for more structurally efficient, specially fabricated steel box sections with thick plates forming the top and bottom flanges, ' Cook reports. 'At first we thought this would be prohibitively expensive.
But this approach allowed us to use smoothly tapered members, and produced a roof design which was 40% lighter, which is economic.'
At the centre the roof loads are taken by a steel ring beam supported on a ring of 20, 457mm diameter tubular concrete filled steel columns. Lateral forces are resisted by a reinforced concrete replacement for the Snow Gallery which originally surrounded the dome of the reading room and prevented snow slipping off the roof and crashing through the skylights of the bookstacks below. As a further precaution against out-of-balance forces being transmitted to the reading room's cast iron structure this new concrete ring also sits on sliding bearings.
Around the perimeter the roof is supported by 120m by 120m square steel posts at about 6m centres, mounted on sliding bearings normal to the existing facade. Tension cables behind the facades provide lateral stability while external horizontal steel trusses stiffen each corner of the roof.
Such elegance and structural efficiency usually comes at a price. Here, thanks to the lack of basic symmetry, more than 1800 different node designs were needed to connect the 6,000 individual structural members.
Each node would have to be located to an accuracy of +/3mm.
The architect naturally wanted the slimmest, most unobtrusive node design possible.
'We started with a blob design and moved onto a flower, both rather large and clumsy, ' Brown says. 'Eventually we finished up with a star made up of a central billet with welded on arms.
'But after talking to potential suppliers, we adopted the suggestion of the Austrian fabricator Waagner-Biro, which was a star cut out of a single 200mm thick sheet of steel. The structural members have tapered ends which are welded into the spaces between the star's points.'
To minimise the risk of weld failures Buro Happold specified an offshore grade of steel. A computer program was developed which set out the whole roof and detailed every node. 'Foster also used it to set out the glass, ' Cook adds. 'Otherwise we would have had to erect the steel then measure up for the glass cutting, which would have added months to the project.'
Nodes and members were manufactured in Austria and sent to a Waagner-Biro subsidiary in the Midlands. Here they were made up into 152 prefabricated ladders weighing no more than 5t. The ladders were assembled into radial trusses on the working deck that covered the entire Great Court, set into position then linked insitu into the final lattice form.
Of the thousands of welds needed in the factory and on site, all of which were tested, only two failed.
Setting out of the construction was complicated by the lack of symmetry. However, when the props were removed in March and April 2000, the roof settled 150mm and spread 90mm; within 20mm of the predicted figures.