Yearly Archive: 2019

Issue No. 2 of the magazine “Benchmark”, the international magazine for engineering designers and analysts from NAFEMS, in 2019 focuses on using the cloud for simulations. NAFEMS is the international association for the engineering modelling, analysis and simulation community, a not-for-profit organization established in 1983. Dr.-Ing. Volker Gravemeier, chief executive officer of AdCo Engineering^GW, is an active member of the NAFEMS Multiphysics Working Group (NAFEMS MWG).

In the introductory article to this main topic of the issue, “The costs and benefits of using the cloud for simulations”, commissioning editor Althea de Souza invited Lee Margetts, Andrew Jones, Wolfgang Gentzsch, Rodney March and John Baxter to discuss the issues around the total cost of running simulations in the cloud or in an on-site facility. The bottom line of this discussion is that we still remain to be in the early days of moving to the cloud in terms of engineering simulations. Moreover, while “there is certainly a place for the cloud, it is not the answer for everything everywhere”. As pointed out in the article, there are definitely situations to benefit from the cloud: for instance, “for an organization new to HPC, or even considering occasional large simulations, it makes sense to start off using the cloud, or some external model. You start using the cloud to find out if HPC can help, then, maybe find out that it will eventually be cost-effective to move to in-house facilities, but the cloud has allowed to ‘try then buy’”. In another article, “Head in the clouds, feet on the ground: my experience using Saas”, Althea de Souza describes her experience when using SaaS for simulation and states her personal pros and cons in this context.

NAFEMS Technical Officer Ian Symington reached out to members of the NAFEMS Vendor Network to get their thoughts on the topic of cloud computing and SaaS for simulation. Among others, it is discussed how major code vendors adopt their licensing models to accommodate cloud computing platforms. In this context, Wolfgang Gentzsch, President of UberCloud, mentioned that “larger independent software vendors (ISVs), like ANSYS, Dassault Systèmes, and Siemens, tend to have their own home-made cloud solution which is usually quite proprietary and restrictive. Customers describe it as a vendor lock-in in different forms; … This has a good reason and stems from the early days of cloud computing when ISVs believed … that their own customers would tend to prefer affordable, short-term, on-demand licenses over their usually quite expensive traditional annual/perpetual licenses.” However, there is another, probably even more severe reason preventing a full adoption, as noted by Ian Campbell, CEO of OnScale: “traditional engineering simulation software is also not architected for the highly parallel nature of the cloud. ‘Simple’ things like running a model across several hundred nodes of commodity cloud instances is impossible with legacy codebases because these code bases evolved through acquisition (e.g. buying a mechanical solver and then buying a separate electrical solver and stitching them together under the hood in a very inefficient way), as opposed to organic development with a goal of highly scalable delivery”.

AdCo Engineering^GW will soon launch its new cloud-computing platform “AdCo On Demand” and offer its multiphysics software AMSE as well as its UQ software QUEENS as a service. Evidently, there will not be any lock-in effects whatsoever, and all software codes were organically developed with the goal of high scalability. Thus, you will take full advantage of the highly parallel nature of the cloud. Do not hesitate, schedule your individual test drive with “AdCo On Demand” – and enjoy the ride!

The demand on the predictive capability of modern Computer Aided Engineering (CAE) toolchains constantly increases. This implies that simulation models are not only required to correctly represent the underlying physical system, but also that real-world variability and uncertainty must be taken into account for enabling a shifting of the product development cycle more and more from physical experiments to the computer. Uncertainty quantification (UQ) is key in improving the predictive capabilities of state-of-the-art Computer Aided Engineering (CAE) workflows. Furthermore, it speeds up the product design and development processes, particularly in the early phases of those processes, which not only reduces the time required to develop a product, but also the number of potentially expensive physical experiments. For all our clients as well as all visitors of our website who are interested in UQ and its enormous potential, we would like to provide a white paper as an introduction to the why and how of UQ.

AdCo Engineering^GW bundles know-how from statistics, machine learning and engineering to provide innovative algorithms for the quantification of uncertainties in our software QUEENS. Thereby, we enable our clients to harness the full potential that digital product development tools and processes can offer. Please visit our page on UQ.

The third issue of the magazine “Benchmark”, the international magazine for engineering designers and analysts from NAFEMS, in 2018 is entitled “Artificial Intelligence & Machine Learning”. NAFEMS is the international association for the engineering modelling, analysis and simulation community, a not-for-profit organization established in 1983. Dr.-Ing. Volker Gravemeier and Dr.-Ing. Jonas Biehler of AdCo Engineering^GW are active members of the NAFEMS Uncertainty Quantification Working Group, which topics are tightly related to artificial intelligence and machine learning.

In his article on “The Applicability of Artificial Intelligence in Design and Manufacturing”, Phil Cartwright argues that the ability to harness data generated along the product lifecycle and feed this information back into the design and development process will lead to significant benefits. In the remainder of the article, an example for machine learning techniques in combination with state-of-the-art simulation approaches is provided, to reduce rework, scrap, and repair in a liquid composite moulding process. In this context, the machine learning approach results in improved tuning of the manufacturing process parameters and ultimately reduced manufacturing cost for the production company.

Combining techniques from the machine learning community with state-of-the-art simulation approaches is without a doubt a very promising way to go for improving efficiency and speed, which will eventually result in reduced costs. In our view, this does not only hold true for manufacturing processes, but also for the complete product development cycle in general.   This is why AdCo Engineering^GW develops novel approaches to fuse information from simulation models with experimental data through so-called multi-fidelity approaches. These approaches are powerful tools to identify, e.g., those production parameters with the highest impact on the reliability of a product or a component thereof, respectively, via sensitivity analysis.  Further product or system attributes can be assessed as well, and both production and design parameters may be investigated to substantially improve the results in the end.