Only time and resource constraints limit the size and complexity of the implicit analyses that LS-DYNA users would like to perform. Rolls-Royce is an example thereof, challenging its suppliers of computers and mechanical computer aided engineering (MCAE) software to run ever larger models, with more physics, in shorter periods of time. This will allow CAE to have a greater impact on the design cycle for new engines, and is a step towards the long-term vision of digital twins. Towards this end, Rolls-Royce created a family of representative engine models, with as many as 66 million finite elements. Figure 1 depicts a cross-section of the representative engine model.

With the rapid development of sensing, communication, and computing technologies and infrastructure, today’s manufacturing industry is marching towards a big data era and a new generation of digitalization and intelligence. The availability of big data provides us with a golden opportunity to promote smart manufacturing. Nevertheless, the deployment and popularization of big data analytics in manufacturing is still at its nascent stage. One critical challenge results from the lack of high-performance computing (HPC) capability, which is crucial for responsive and intelligent decision-making in the modern manufacturing industry. To address this challenge, this paper proposes a framework and some general guidelines for implementing big data analytics in an HPC environment. The details of the whole workflow, from the prototype to the final application, are high-lighted. A case study for intelligent 3D sensing with real-world manufacturing data is presented to demonstrate the effectiveness of the proposed framework.

In this study, the sensitivity of the apparent response of trabecular bone to different constitutive models at the tissue level was investigated using finite element (FE) modeling based on micro‐computed tomography (micro‐CT). Trabecular bone specimens from porcine femurs were loaded under a uniaxial compression experimentally and computationally. The apparent behaviors computed using von Mises, Drucker‐Prager, and Cast Iron plasticity models were compared. Secondly, the effect of bone volume fraction was studied by changing the bone volume fraction of a trabecular bone sample while keeping the same basic architecture. Also, constitutive models' parameters of the tissue were calibrated for porcine bone, and the effects of different parameters on resulting apparent response were investigated through a parametric study. The calibrated effective tissue elastic modulus of porcine trabecular bone was 10±1.2 GPa, which is in the lower range of modulus values reported in the literature for human and bovine trabecular bones (4–23.8 GPa). It was also observed that, unlike elastic modulus, yield properties of tissue could not be uniquely calibrated by fitting an apparent response from simulations to experiments under a uniaxial compression. Our results demonstrated that using these 3 tissue constitutive models had only a slight effect on the apparent response. As expected, there was a significant change in the apparent response with varying bone volume fraction. Also, both apparent modulus and maximum stress had a linear relation with bone volume fraction.

Cray, LSTC, NCSA, and Rolls-Royce formed a partnership to explore the future of implicit computations as the scale of both finite element models and the systems that run them increase. Rolls-Royce created a family of dummy engine models, using solid elements, with as many as 200,000,000 degrees of freedom. NCSA ran these with specialized LS-DYNA® variants, generated by Cray, on their Blue Waters machine, a hybrid Cray® XETM/XKTM system with 360,000 AMD cores. Processing and memory bottlenecks revealed themselves as the number of processors increased by an order of magnitude beyond that familiar to today’s developers and users, and LSTC made improvements to LS-DYNA. This paper will discuss the challenges encountered, enhancements made to LS-DYNA, and the results when extending the limits, both in terms of the scale of the model and the number of processors. This is ongoing work, and we will conclude by discussing the path forward that has been illuminated.

One of the difficulties in Fluid-Structure Interaction(FSI) co-simulations is the convergence within each time step, which include the convergence of each participating solver and the data transfer among them. Especially, when the solid material non-linearity or thin geometries are involved, there are typically large deformations which could eventually result in convergence difficulty and even the failure of numerical algorithms. Adopting a hyper-elastic pipe benchmark case, this paper explores the effects of different scale factors on the overall convergence in each co-simulation time step. Three constituent components are involved and dedicated to structure, fluid and system coupling, respectively. The coupling scheme is in two-way.

Lean premixed (LP) combustion in gas turbines has a great potential to achieve higher operating efficiency and lower emissions and hence draws an immense interest from both the research and the industrial community. Due to the nature of the flame in an LP combustion mode, the turbulent eddies alter the flame structure thereby enhancing the mixing of the reactants and the products. The transport of scalar quantities is augmented due to the turbulence. Lean nature of the fuel-air mixture results in a lower overall temperature thereby reducing the potential for NOx generation. The design constraints involved in achieving these objectives are challenging. LP combustion requires the fuel-air mixture to be as lean as possible to attain lower equivalence ratio, stable combustion dynamics at all operating conditions and sufficient residence time.