<<BACK TO HOME

CAE has improved component design, but new techniques give it much more scope. By Simon Bickerstaffe

Computer-aided engineering (CAE) is the key enabler for better designs to reach the market in more quickly. Tools and their applications are evolving, workflows are changing and greater computing power is being exploited for new analyses.

The trend is toward more thorough analysis. While most CAE analyses converge to a single solution, the use of optimisation software with standard finite-element analysis (FEA) and computational fluid dynamics (CFD) solvers can evaluate a series of potential designs and offer a range of solutions. Optimisation can be automated, reducing run times.

“The point of this technique is to convince people that their traditional trial and error approach can be substituted for something more logical that gives their design process an edge,” says Francisco Campos, senior consulting engineer with CAE consultancy Icon. “You get a shape that a person would probably never have thought of. The advantage comes from introducing statistical tools into the design process.”

Solutions are only based on nominal input values. These can be made more robust by redefining the CAD variables as normal stochastic distributions with a standard deviation, to account for variables such as component tolerances and manufacturing variance.

“But now, on top of probing the mean, you have to probe the whole function,” says Campos. “You might have to run a further 20 designs, multiplying the number of experiments you have to perform. It’s expensive, but it’s the logical thing to do: it’s going to tell you how good your design is.”

Campos says optimisation is limited by hardware, but clustering processors and the use of open source software can reduce costs. “We use the Open Foam CFD solver because it provides a solution without additional cost,” he says. “There’s no limit to the number of licences you can run . You’re only limited to whatever hardware capabilities you have.”

In one example, Icon split the analysis of a 200,000-cell mesh over a cluster of 16 processors. “OEMs have access to much larger clusters, but have to pay for each licence. If you want your runs to go in parallel, you have to pay,” says Campos.

Adding robustness to optimisation has been investigated at Jaguar. Like Icon, it has used stochastics to produce a robust design after optimisation. But the company’s research shows that robustness applied after optimisation under ideal conditions yields large numbers of solutions that are, in fact, failing performance criteria and are therefore not robust. In one case, the analysis of a knee bolster, 60 per cent of the solutions didn’t meet the requirements.

Jaguar’s answer has been to add the “noise” factors – component, material and manufacturing tolerances – during optimisation. “Traditionally, you add your allowances, but then when you perform real tests, components can still fail,” says Tayeb Zeguer, senior technical specialist in safety and CAE at Jaguar. “When you do a correlation, you may find a material degradation or a geometry change. We are trying to use the simulation to capture all the issues, reduce test workload and get designs right first time.”

Jaguar uses Altair’s Hyperstudy software for the optimisation process. New algorithms were added to introduce the noise into the system.

The design for a Six Sigma process was applied to a knee bolster. “Some points don’t meet the requirements, but these might only be 0.1 per cent of the population,” says Zeguer.

“Adding noise factors during optimisation is the best way to generate robust design. This is a new way of working. Get it right first time, reduce cost and develop products in a shorter time.” Jaguar is still evaluating the process.

Zeguer says there are two main reasons why this approach hasn’t been taken before: “It’s complicated to do, and it takes a lot of processors.”

Optimising product design at the concept phase is another approach – get the basic shape correct from the outset. “The way to have optimisation make a significant impact in quality, robustness and shortening of the design cycle is to move it up front,” says Richard Austreng, business development director of modelling and visualisation at Altair. “We’ve been pushing that for 10 years. A large part of the problem is the standard design process: create, analyse, optimise. By the time optimisation happens, the results are too late to benefit the design.”

Altair’s Optistruct software can, based on parametric inputs, develop the ideal shape of a structure and optimise it for the given constraints.

Others agree that this will be a growth area in CAE. “You take something you already have and improve it. It’s about the part before that. How do you get to that?” says Campos. Icon and others are looking at a process called polyoptimisation – a rewriting of the fluid equations to develop a shape from a design space.

Fluid-structure-interaction problems are widespread in automotive engineering but are rarely simulated because of the complexity. Their effects are often ignored because the amount of effort to get the solutions isn’t worth the investment of time.
FEA and CFD solvers often come from different vendors and transferring information between codes can be very difficult.

Products such as Mode Frontier and Hyperstudy can manage the data flow and get results. Alternatively, the two codes can be coupled with a 1D analysis tool. Jaguar has experimented with this approach to investigate exhaust manifold impingement noise with some degree of success.

CAE software vendor Ansys has invested in this field. Having acquired Fluent last year, it recognises the issues posed by fluid-structure-interaction analysis. “It can work well in aircraft engines, but we haven’t got there yet for automotive – it’s just too difficult,” says its product manager Paul Bemis. “The opportunity to couple those two codes in an easy-to-use way exists. There’s evidence within Ansys of being able to do that with CFX, but its still maturing.”

The meshes need to map to one another and this is complicated. They tend not to align well and this causes problems with convergence.

Cost pressures and a move toward wholly digital design are creating the need for these analyses because there simply isn’t an alternative. “Historically, fluid or structural analysis gets close, then you build prototypes,” says Bemis. “Prototypes are being cut from programmes due to cost, forcing engineers to do the analysis with math-based tools. That is creating demand for more comprehensive analyses. That’s why we’ve brought structures, fluids and acoustics into one environment.”

Looking further ahead, there is a consensus that the future of CAE and optimisation is multiple-objective live trade-offs. “You don’t want a crash engineer optimising for crashes, NVH engineers for NVH, and so on. What you have to do is take the system and put in all the boundaries. NVH, crash, durability are all inputs,” says Zeguer. “You apply all the noises and the outputs are torsional stiffnesses, crash scenarios, injuries. You get out all the cloud data for all the different things. You look for the cloud that meets all the different requirements.”

“Instead of sequential serial activities, the new approach is parallel activities,” says John Sullivan, Ford’s director of process, methods, tools and information. Sullivan is working on the CAE processes for a 12-month car – just one year from design sign-off to the first car you drive.

“We can probably do that within the next five years and that’s by compressing the timing using CAE,” he says.”The challenges are how to make it work quicker, eliminate waste and deliver a real-time analytical trade-off in the compressed timeframes.”
If this is possible – and it seems likely given the pace of developments – the deployment of such advanced CAE techniques could be one of the factors in deciding OEMs’ global competitiveness.