Theme 5/6 High Performance Computing (HPC) and Simulation Knowledge Mining and Abstraction (SKMA)

It costs roughly £0.5 million to build a prototype of a car. Many prototypes need to be constructed during the process of designing a new vehicle and also when developing a new version of an existing vehicle. Clearly, anything that can reduce the number of prototypes needed will have a significant impact on the cost of automobile design. Simulation is already used extensively in the automotive industry. For instance, assessing car crash worthiness during the design process is now entirely performed computationally. The aim of the Programme for Simulation Innovation is to increase the use of simulation in car design. Indeed it can be seen as a further significant step towards the goal of 'virtual vehicle design' where the entire design process is carried out on the computer. However, increasing the amount of simulation used in the design process will significantly increase the demands on the computational ecosystem and, perhaps more seriously, introduce many layers of complexity which could lead to inefficient use of the computational ecosystem and poor decision making. Our work is targeted at addressing these potential problems by increasing the efficiency of individual computational tasks, automatically documenting simulation data to avoid misintepretation, allow data re-use and enable automated data archiving protocols and, finally, modelling the complexity of the virtual vehicle design task and, based on this, implementing rules to resolve conflicts in the computational and data flows and in the decision stages of the design process.

Modelling and simulation are very widely used in science and engineering. Our work under Theme 5/6 of the Programme for Simulation Innovation is an enabling role to allow: more complex simulations to be peformed; computational performance across the compute and data ecosystem to be improved; complex workflow tasks to be efficiently implemented; and data to be automatically annotated with meaningful metadata, accessed and archived according to rule sets and also mined for additional information. As such our work will have en extremely wide impact across academia and industry.

There has been much discussion of multiscale modelling methodologies across many branches of sience and engineering but this is still usually achieved by passing parameters between models. Similarly, despite the importance of multiple elements of physics and engineering that affect many processes, there are no general purpose codes that allow this different approaches to be used together. The proof of principles applications of the multi-scale/phyiscs/engineering code carried under HPC WP1,2 will thus have a wide impact on simulations in both academia and industry by demonstrating the flexibility and power of this new methodology.

The impact of 'commodity' low cost hardware and software on high performance computing during the last 10 years has been enormous. It has led to an order of magnitude decrease in computational costs. The proposed work under HPC WP3-6 continues progress in this area by bringing further increases of speed on single CPUs, assessing to effectiveness of various accelerators such as GPUs and MICs for computer aided engineering tasks, and further extending the commodity compute model to data curation and to high quality and remote visualisation. This research will have significant impact on both the academic community and industry by continuing to enhance the performance yet drive down the cost of the computational ecosystem.

It is clear that one major challenge for the next generation of simulation is to carry out much more complex computational tasks involving many compute tasks, frequent use of data and also possibly visualisation. Developing methodologies to handle the complex workflows generated by these conflicting computational demands will be crucial for moving along the path to virtual design in a whole host of fields, not just automotive. The research we will carry out on this under SKMA WP3 and as part of HPC WP5 will have the largest impact on industry where there is the strongest driver to moving to these much more complex compute problems.

Data and compute have to a large extent developed as completely separate problems in the computational ecosystem. This can be attributed to cost and technical issues which no longer apply. Our research under SKMA WP1,2,4 will demonstrate the added value that can be achieved once data is fully annotated and used within the computational ecosystem and will provide a paradigm shift for most existing users who presently only use either high performance computing or large data.