Simulation
Speed up processes with 'RISTRA' technology
When it comes to simulating designs, users have to be prepared for time-consuming processes. The 'RISTRA' technology, which accesses the GPU instead of the CPU for the simulation, is intended to provide a remedy - and thus accelerate the process enormously.
Nowadays, designs are created using powerful 3D CAD programs. The result is geometry data that is used as input for CNC production, for example, via commonly used data formats such as STEP or IGES. However, before a design is transferred to production, it must be ensured that it meets the requirements from the specifications. One way of predicting the behavior of a component is numerical simulation. Starting in high-tech sectors such as aerospace and energy technology, simulation has long since reached SMEs. Numerical simulation is more economical and faster than practical testing, is safe for people and the environment and delivers reproducible, reliable results. One of the aims is to obtain a component that meets all strength requirements with minimum material input by gradually changing parameters such as material thickness.
In order to be able to simulate, a calculation model must be created from the CAD data. A special program, the so-called preprocessor, breaks down the component into small parts, usually tetrahedrons, so-called 'finite elements'. The calculation engineer parameterizes the calculation model in the preprocessor by specifying the material and the boundary conditions of the loads to which the component is to be subjected. Increasingly, however, the designers are also taking on this task.
The model can now be transferred to the solver (structural mechanics solver). This is the name of the actual simulation program that calculates the mechanical behavior. This imports the calculation model and generates a linear system of equations from it, taking into account the partial differential equations of structural mechanics. Several million equations are no exception, depending on the complexity of the design. In the case of mechanical simulation, the solver calculates how the specified load cases affect the structure of the component. The displacements of the individual nodes are calculated. In subsequent steps, distortions, stresses and nodal forces are then calculated. The results are then displayed graphically. False color models are commonly used, which provide information about the stresses in the material and the deformations at a glance.
Labor-intensive and time-consuming
Until now, the people involved in the design process have needed different computer-aided tools, usually on different computers, for this iterative process of geometric modelling, initialization, simulation and analysis of the results. It is not uncommon for the data models to have to be converted between these tools. Up to now, the simulation itself has taken up most of the time. Depending on the computer system used, this can take many hours or even days. To save time, many companies transfer the structural data to high-performance computers that can be used as part of a cloud solution.
Once the simulation results are (finally) available, those responsible can use them to make changes to the design and load parameters. They are often guided by their experience and 'try out' how a change in material thickness affects the stability of the component, for example. And the cycle starts all over again: a new mesh structure has to be calculated as input for the simulator so that the next simulation run can begin. And that can take time... Many designers do not come completely close to their goal of achieving a design that is truly optimized in all respects. They simply run out of time or are slowed down by budget limits. This is unfortunate, as it is of great interest to minimize the use of materials and thus also production time, especially for designs that are manufactured using additive processes, for example.
Technological leap for the construction process
The Fraunhofer Institute for Computer Graphics Research IGD in Darmstadt has been working on improving this situation for years. At the beginning of 2018, the institute presented 'RISTRA' (Rapid, Inter-active Structural Analysis), a new type of technology for structural mechanics solvers. The special feature here is that the solver does not run on the CPU, but uses the massively parallel calculation potential of commercially available, low-cost graphics cards (GPU - graphics processing unit). As the computing time is orders of magnitude shorter than with standard simulations, the designer sees the simulation results on his screen virtually in real time and can draw his conclusions immediately. RISTRA' was developed for commercially available graphics cards from Nvidia. By shifting the simulation calculation from the CPU to the graphics processor, Fraunhofer IGD utilizes the enormous resources of this massively parallel computer architecture. While a CPU today has four to eight cores, the graphics card currently has up to 5000 cores available. The technology currently supports the following structural-mechanical concepts: geometrically linear elasticity, linear isotropic and anisotropic materials as well as linear, quadratic and cubic attachment functions on tetrahedrons.
Figure 1: The 'small iteration loop': Design iterations within 'RISTRA' enable a very fast working method. Any changes made to the geometry must then be reproduced manually in the CAD program.
© Fraunhofer IGDEven the preliminary version presented at the beginning of 2018 delivered impressive results. A comparison of Fraunhofer IGD's interactive simulation solution with fast commercial software revealed that standard software used for the comparison required 150 seconds for a model with more than 1.3 million finite elements. RISTRA' delivered a result after 0.875 seconds of pure computing time. The calculation engineer or designer immediately recognizes potential for improvement and can then make immediate changes to the design. If changes are made to the load situations or the geometry within the solver as a 'small iteration loop', the simulation can be repeated immediately(Fig. 1). This results in unprecedentedly short times - an 'interactive simulation'. The design changes resulting from such an interactive optimization process must then be incorporated into the CAD model in order to obtain consistent, valid production data.
Figure 2: Design process starting with the CAD program, but already with accelerated simulation via GPU. Efficiency depends largely on the speed of the (sub-)process chain CAD program/preprocessor and data exchange.
© Fraunhofer IGDIf more complex geometry modifications are required, the 'large iteration loop'(Fig. 2) must be used and the changes are made directly in the CAD system. The path to the subsequent simulation is via renewed pre-processing with a pre-processor, which generates the calculation model. In the 'large iteration loop', 'RISTRA' must import the new model and analyze the tetrahedral mesh in order to generate the linear system of equations. In the preliminary version of 'RISTRA', these steps still run on the CPU and take around 20 seconds in the above example. Although the simulation solution is already significantly faster than the comparison software, the time saved in the actual simulation was counteracted by the still unacceptably long initialization time.
The path to the optimum
Figure 3: In the new version of 'RISTRA', the pre-processing steps for generating the equation systems also run on the GPU, which shortens the cycle times of the interactive simulation enormously.
© Fraunhofer IGDHowever, the developers at Fraunhofer IGD have now also found a solution for this and presented it in the new version 'RISTRA 2019'. The operations for generating the system of linear equations have been specially optimized for processing with the GPU. As a result, all calculation operations can now be performed on the GPU after importing the calculation model(Fig. 3). The pre-processing steps of the 'large iteration loop', which took 20 seconds in the above example, are drastically reduced to less than one second. For the above-mentioned model with more than 1.3 million finite elements, 'RISTRA' can deliver a result after a total of 1.83 seconds - that is more than 80 times faster than the comparison software. In the large iteration loop, efficiency depends on the sub-process chain of the CAD system and the pre-processor and the data exchange. However, solutions are already available here too, but they need to be worked on together with the software providers in the sub-process chain.
Intuitive and interactive
Of course, changes to the geometry, provided they are based on a given mesh structure, or other parameters such as load cases or material selection, can still be carried out as a small iteration loop directly in 'RISTRA'. The user is offered a self-contained high-speed loop. The pleasant thing about this is that changes to the parameters, their subsequent simulation and the evaluation take place within one and the same system environment. The iteration loops are so short that it is worth going through many steps towards the optimum.
Figure 4: Left: Initial simulation of an idealized motor mount with excessive stresses in the center of the component. Right: After a geometric modification, the stress peaks are lower.
© Fraunhofer IGDDepending on the complexity of the design, the simulation results are available almost in 'real time'. Intuitively, almost playfully, the user can change material parameters, load cases and geometry changes to the given mesh structure in 'RISTRA' and see 'what happens'(Fig. 4). These findings form the basis for the necessary changes to the design or geometry in the CAD program. The researchers at IGD therefore predict a fundamental change in design processes, a development towards a direct, intuitive working style. This will naturally lead to better results, not only in terms of the development time required, but also in terms of the quality of the design.
At the current stage of development, however, only simple, prototypical geometry changes can be made in 'RISTRA', in which the simulation mesh is adapted directly. If the simulation has shown that certain geometry changes come closer to the optimization target, these must be reproduced manually in the CAD system.
The institute is currently in negotiations with software providers who will integrate the new algorithms into their software environment as licensees. An initial partnership already exists with Meshparts, who are integrating 'RISTRA' into their simulation program.
Author:
Dr. Daniel Weber is deputy head of the Interactive Engineering Technologies department at Fraunhofer IGD.
RISTRA' at a glance
Hardware requirements:
Graphics cards with CUDA architecture, at least with Compute-Capability-3.0. Interfaces to many common formats, such as Creo Simulate, Salome or Abaqus-Export can be developed.
'RISTRA' supports the following structural mechanics concepts:
- time-dependent and time-independent deformation,
- geometrically linear elasticity with small strains and small rotations
- geometrically non-linear elasticity with small strains and finite rotations
- linear isotropic and anisotropic materials, as well as linear, quadratic and cubic attachment functions on tetrahedra (finite elements TET4, TET10, TET20).
The timetable
Prof. Dr.-Ing. André Stork, Head of the 'Interactive Engineering Technologies' department at Fraunhofer IGD.
© Fraunhofer IGDThe development of 'RISTRA' is already well advanced. Prof. Dr.-Ing. André Stork, Head of the 'Interactive Engineering Technologies' department at Fraunhofer IGD, explains what's next in an interview.
Prof. Stork, how long have you been working on this project?
Prof. André Stork: The technology has been continuously developed in numerous publicly funded projects. We have now been working on this topic for almost ten years. Intensive cooperation with an industrial partner and European and national research projects have contributed to the current status of this technology.
You will continue to work on this topic. What extensions to 'RISTRA' do you have on the agenda?
Prof. André Stork: Our research group is not only extending the solution to geometrically non-linear elasticity and non-linear materials. We also intend to simulate contact mechanical issues using GPU acceleration - a functionality that is in high demand.
Additive manufacturing processes are currently the hot topic. Will the IGD also have an answer for this area?
Prof. André Stork: Yes. With our current projects 'CAxMan', 'M3D' and 'Future AM', we are developing innovative technologies that support the additive manufacturing process in the areas of modeling and simulation. In these projects, we are researching innovative data structures and algorithms that cover the requirements of the entire process chain, from modeling and simulation to 3D printing - especially for objects with locally varying properties.
You might also be interested in
AI-based component control for faster changeover times
Fraunhofer IGD's Marquis software makes it possible to inspect components on the move and in real time.
read more...Wago takes over 'Zaphire' building management software
The Wago Group is further expanding its building automation business and has acquired the 'Zaphire' building management system (BMS) from the Scandinavian group Granitor.
read more...Platform for AI security in companies
Check Point Software Technologies has introduced 'Check Point AI Defense Plane'. This unified AI security control plane helps organizations manage the networking, deployment and operation of AI across the enterprise.
read more...New EU Toolkit for sustainable Electronic Products
The European research project 'Increace' is setting a milestone for the circular economy: a network of universities, research institutes and industry partners has presented a comprehensive set of basic principles and a new evaluation tool that...
read more...IFS relies on Asset-based Pricing Models
IFS is introducing a new pricing model designed to simplify access to 'Industrial AI'. Instead of paying per user, companies will in future pay per asset operated, enabling company-wide use without cost increases.
read more...Partnership for edge AI in industry
Nota AI and SiMa.ai have agreed on a strategic collaboration to develop AI solutions for industrial applications. The aim is to optimize AI models for use directly on devices at the network edge, where high computing power and energy efficiency are...
read more...Secure data rooms in industrial value creation networks
The Industrial Digital Twin Association (IDTA) and the International Data Spaces Association (IDSA) have agreed to cooperate in order to link digital twins and data spaces more closely in future. The aim is to strengthen interoperability, secure...
read more...Capgemini takes over Piterion
Capgemini has signed an agreement to acquire Piterion. The company specializes in Product Lifecycle Management (PLM) and Manufacturing Operations Management (MOM) and is successful in networking critical manufacturing systems and operating complex...
read more...Digital Intelligence meets Production
What does modern production look like when sensor technology, the cloud and IIoT work together seamlessly? A look behind the scenes at ifm shows how the Group not only develops its own IIoT platform 'moneo', but also consistently uses it in its own...
read more...