For years analysis and simulation has been a ‘follower’ in the design process, something done after the concepts had been developed, the final scheme selected and the initial detailed design had been started.
A much more sensible approach is to have simulation lead the concept development and selection process so the resulting design fulfils all the aesthetic design criteria as well as those of strength and durability.
Recent dramatic increases in computer power, plus the development of more efficient algorithms, have enabled us to model systems with increasing levels of complexity and accuracy. The problem with these models is that they are perfect for verification work and can run overnight on new hardware but are limited when it comes to using them as part of a design process.
Design involving iteration, change and optimisation and single formulation overnight runs — detailed as they are — do not react fast enough to be meaningful in a design context.
For example, in a front-end car crash restraint system there are about 28 variables. If you look at just five values for each variable as a design parameter range, when European and FMVSS 208 crash test requirements are included there are about 12 design points to optimise and balance these 28 variables with in each of the five design parameters.
‘There are something like 1.3 million permutations, which means these sorts of runs are not practicable in a design environment,’ said John Cooper, managing director of TNO Automotive Safety Solutions (TASS) in the Netherlands. ‘Then if you have five design loops typically in a process, you have to do all that work five times.’
Safety legislation is complicated, reducing the available design time and cutting the cash available for development, so the industry is caught in a vicious circle. To overcome this, designers have to use an efficient process for design. ‘This is where we come in,’ said Cooper. ‘Our MADYMO design and crash simulation software, in conjunction with statistical techniques, means designers can solve real safety-related design problems within a realistic timeframe.
‘We take a modular or a sub-system approach to isolate the problem, which allows the suppliers to develop a solution independently of the car manufacturers.’
In terms of the physics involved there are a number of competitive solutions available but TASS claimed MADYMO is easier to use, and design optimisation is far more accessible. The company has modularised its software to allow designers to use the right tool for the job. If the tools are used efficiently in the right areas, they complement each other.
A similar approach is taken by Altair Engineering. Its system, HyperWorks, is an integrated CAE framework containing best-in-class solutions for the complete virtual product development process.
HyperWorks contains finite-element and multi-body simulation solutions and an open, programmable platform that is easily integrated into existing processes. The system includes modelling and assembly, visualisation and reporting, virtual manufacturing and robustness assessment. It interfaces to all major CAD and CAE packages and facilities in a simulation-driven design approach.
The increased integration of simulation studies into product design has accelerated the introduction of CAE technologies such as optimisation and process automation. Optimisation technologies allow the engineer to define a range of acceptable design parameters, performance criteria and physical targets such as mass and analytically iterate to identify the optimum combination of the variables to meet the objectives. Often, tasks and processes of a repeatable, quantifiable nature can be automated using software, thereby improving the productivity and quality of the simulations.
Altair’s OptiStruct/Analysis is a fast, accurate and robust finite element solver for linear solution sequences and contact problems. Up to four times faster than conventional solver technologies, it delivers flexibility and advanced functionality, with no restriction in model size.
HyperWorks includes new technology for conceptual design, model morphing and structural optimisation that strengthens simulation as a design driver. It does this by addressing product performance and manufacturability earlier in the design cycle.
A new approach to mesh and geometry morphing simplifies mesh-based design changes for shell and solid models, as well as automating the creation of design variables for size and shape optimisation. Altair includes stress-constrained topology and topography optimisation in OptiStruct to generate the most efficient conceptual designs for stress-critical structures like aircraft.
OptiStruct also includes ‘free-shape’ optimisation for automated boundary shape variation with minimal user input and ‘free-size’ optimisation to develop robust variable thickness design solutions, composite laminates and ‘shear-web’-type structures.
Comsol, Blue Ridge Numerics, CD-adapco and other multi-physics vendors are turning to statistical approximation and fast solvers to help deal with the complexities of modern simulation in a timeframe that supports the design process.
Instead of solving all the equations over each region, new methods first numerically approximate the unknowns of interest (for example, temperature, frequency, and velocity) using predetermined mathematical functions based on the shape of the domain.
Complex problems need iterative solvers, which start from a suitable initial guess then calculate successive updates that ultimately converge to a solution. These solvers are more difficult to use because they require high-quality preconditioners, basically numerical methods that simplify the analysis. However, a well-chosen preconditioner reduces the number of iterations required and, therefore, saves computing time.
Among the many available preconditioners, the geometric multigrid (GMG) technique is widely used because it handles a large class of problems efficiently, particularly those in structural mechanics, fluid flow, and electromagnetics. Analysis software such as Comsol Multiphysics supports the GMG method.
The GMG algorithm jumps between different linear systems representing coarser and finer meshes of the underlying partial differential equations (PDAs). The idea is to perform a fraction of the computations on a fine mesh while solving the full system under study on a coarser one. This provides an iterative algorithm that is fast and memory-efficient, allowing designers to use complex simulation as part of the design process rather than as post-design verification.