Case Study: Silicon Graphics-based workstations

By moving their design process to a set of 3D visualisation workstations from Sillicon Graphics, Rover improved car safety and at...

By moving their design process to a set of 3D visualisation workstations from Sillicon Graphics, Rover improved car safety and at the same time decreased testing times

A key part of the design process at Rover involves the use of digital prototypes to reduce the cost of building physical prototypes for crash tests. The production of a physical prototype can cost as much as £250,000 and can take over 4 months ( for a crash test lasting 0.1 seconds. The use of digital prototypes reduced the cost and time of the total design process by reducing the number of physical prototypes that need to be created. It also allowed Rover to test the safety of vehicles and make changes at an early stage in the design process.

John Hemmings, senior manager of crash simulation at Rover said: "Getting design right at an early stage is essential to manufacturers. Every design change has a cost associated with it. At an early stage in the design process, the cost of change is low. Once you commit to physical testing, the cost of change rises considerably. It involves the cost of building new prototypes ( including the cost of the test-engineer's time ( as well as the cost of reproducing the original design."

All of this means that Rover relies heavily on simulated crash tests and these need to be as realistic possible. Analysts are required to conduct increasingly sophisticated crash analysis so that, year on year, they can have greater confidence in their predictions. Whereas in 1990, analysts tested the impact of a car colliding directly with a wall at 48kph, by 1996 they had begun to test the impact of a deformable offset barrier (with only 40 per cent of the car hitting a wall) at 64kph.

Prior to 1996, Rover's crash analysis team conducted its research using a Cray YMP supercomputer at British Aerospace in Bracknell. This resource was shared with other analysis groups at Rover including the body structures group (stress analysis, analysis of loads, noise vibration and harshness analysis), the chassis group (local stress analysis and fatigue analysis) and the power train group (engine component analysis and combustion analysis).

John Hemmings said: "Competition between analysis groups for time on the supercomputer was great and the crash analysis department was often unable to access the system because of other computing requirements. Crash analysis was very much limited by large analysis run times. At times of heavy Rover workload, much of our analysis had to be conducted at night and weekends."

The matter was made worse by the complexity of using the system. A designer's CAD data was sent to a pre-processing package to create the mesh of the model. The mesh in DYNA 3D NCODE was then manually converted into a form that could be sent to the Cray. After processing, the information was sent back through a data convertor before post-process analysis could begin. The system was running over an FTP network so it could take several hours to get data back from the Cray processor.

Rover looked at a number of alternatives. A Silicon Graphics system was already being used to perform complex crash analysis tests at BMW's Munich design site and came with a strong recommendation from the parent company. It had proved to be reliable and had specific technical advantages over other hardware platforms.

Rover updated its operation to include two O2 workstations with 256Mb of memory, 9Gb of storage and MIPS R5000 processors; six INDIGO workstations with 512Mb of memory, 13Gb of storage and MIPS R10000 processors; and 17 OCTANE workstations with 512Mb of memory, 13Gb of storage and R10000 MIPS processor.

Rover also implemented one dedicated POWER CHALLENGE XL server supercomputer at Cowley with 2Gb of memory and 24 MIPS R 10000 processors, plus one Onyx graphics supercomputer. The files for this system were served by a CHALLENGE DM with 60Gb of storage.

With the new hardware in place, Rover changed software from DYNDA 3D code to PAM-CRASH code. BMW prompted the change to PAM-CRASH because it was already using it with its existing Silicon Graphics hardware and was confident that it would run efficiently.

The Silicon Graphics-based system has already helped Rover to reduce the length and cost of the design process. The upgrade of hardware, software and modelling facilities has allowed analysts to predict safety performance with greater accuracy. It has allowed Rover to remove one phase of physical prototype crash simulation. This can account for up to 10 cars which, at £250,000 per prototype, represents a considerable £2.5 million saving.

Silicon Graphics hardware has also enabled Rover to move from 70,000 element models to 250,000 element models which shows greater detail on the components they expected to fail in the crash. This allows the crash analysis team to model the entire structure of the vehicle and its passengers whereas they could previously only simulate individual parts.

John Hemmings said: "3D visualisation is critically important to us. It is likely to become more important as we change the way we transmit information to the user. We plan to develop a server-based web environment that will allow anybody to pull up the crash results for a particular vehicle and to visualise those results. Test engineers may want to reference past analysis and compare it with their tests, and visualisation is critical for the communication of design change between analysts and designers."

Compiled by Ajith Ram

(c) 1998 SGL

Read more on Data centre hardware