A new intelligent design optimization tool for Vector Fields' electromagnetic modeling and simulation software will automatically find the best solution to a design problem, even if a user specifies several or even competing objectives.
Auto-optimization tools are not new, but they usually require manual intervention if the globally optimal solution is to be found, and the simulation times involved often make such a process impractical. Vector Fields’ new intelligent Optimizer automatically selects and manages multiple goal-seeking algorithms to eliminate the need for manual intervention, and to make optimization feasible in virtually all design cases.
Finding the Optimal Solution
"It's always possible to make improvements, even for engineers who routinely create variations of a product and know their subject inside out," says John Simkin, CTO of Vector Fields. "Whether the goal is solving some problem at the frontiers of science, or improving a common product -- such as squeezing a fraction of a percent more efficiency out of an electric motor -- this software can almost always find the optimal solution for any design idea within a practical timescale."
Optimizer works in conjunction with the Opera electromagnetic design package, which is available in general-purpose form, or with solvers dedicated to specific electromagnetic applications including rotating machinery, superconducting magnets, particle beams, dielectric insulation, and the magnetization/demagnetization process.
In partnership with Southampton University, UK, Vector Fields developed the Optimizer tool from scratch because of the drawbacks associated with conventional optimization techniques. With “descent” algorithms, for instance, a tool might only find a local optimization point, and not the optimum for the global design space.
Optimizer breaks new ground because it incorporates numerous algorithmic techniques including stochastic, descent, Pareto, and Kriging, and because it intelligently manages their application. The tool automatically applies and dynamically manages the algorithms to eliminate the usual requirement to perform lots of time-consuming simulations. The range of algorithmic techniques at the disposal of the tool's intelligent core means that optimal solutions can be found for multiple objectives, and objectives that compete with each other, and for problems that would otherwise be too computationally intensive to tackle.
Automatic Operation
Operation is completely automatic. Design optimization problems can be set within a few minutes using a tabular template, and setting lower and upper limits for the design variable(s) in question, and targets for the design objectives, including the ideal value(s) if known. The tool then automatically runs the model a few times, before intelligently analyzing the results offline to determine how to refine the search. It then re-processes the model to find the ideal solution, choosing the most appropriate algorithm(s). In this way, the number of simulations required is dramatically reduced. Simple design models can typically be optimized in less than an hour. Even complex models can typically be handled within a day or two. The tool's ability to choose between one- and two-stage algorithms greatly aids the speed of result.
"There's a great deal of pressure in industry to shorten design cycles and reduce bill-of-materials costs. We believe Optimizer represents a genuine breakthrough for most electromagnetic equipment designers," adds Simkin. "In almost every case, the tool will deliver the best solution to a problem within a day or two at most. And companies can apply it in novel ways, by using the tool to provide the best solution for a range of objectives, such as a combination of performance, cost, and ease of manufacture."
Vector Fields believes the new Optimizer tool will be applicable for well over 90% of current electromagnetic equipment design activity. There are only a small number of cases where the tool will likely not be useful: applications where the design models are extremely complex and represent a processing challenge for today's computers, and when dealing with problems where model behavior is very erratic. Even in this latter case, the tool's intelligence can automatically recognize the need for more data and simply resimulate the model an appropriate number of times.
