AI-Assisted Design for Magnetics and More

­

Magnetics are required for almost all power conversion designs. Until fairly recently, specifying and purchasing those magnets was fairly straightforward. However, as higher frequencies become more common in power conversion, magnetics can be responsible for a greater proportion of losses, making it increasingly difficult to determine the best type of magnetic design for each circuit. Often, there is no ideal solution available off-the-shelf, so custom designs are becoming popular as engineers try to make their designs as efficient as possible.

Dr Chema Molina, founder and CEO of Frenetic talks about the challenges facing designers by saying“The design of the magnetics in a power conversion circuit requires very complex calculations and predictions. Often designers don't have the knowledge of software required to design the magnetics properly, even if they have studied the basics in further education. The main reason for that is that magnetics at today’s high frequencies have interactions between the magnetic and electric fields that act as a feedback loop. Calculations may show that everything is working well, but in reality these high frequency effects can increase the losses by a lot. You can spend a lot of time doing the design, then go to a lab, and the design doesn’t work at all. Finite theorem analysis and simulations can be undertaken to calculate the size of these effects through software like Maxwell from Ansys. However, there are two problems with this process. Firstly, the software is extremely complicated to use and can take years to master. The second drawback is that the software is purely for simulation, so it cannot provide design support. That means spending a lot of time building models.”

In 2020, Frenetic started creating its original software solution to help design magnetics solutions using only simple electrical parameter inputs, such as inductance, precision, frequency, and input voltage. The software builds a model, runs simulations and then designs a transformer that will meet the electrical requirements in a very short period of time. The design includes core materials, winding materials and configuration, and lead wires. It can then be customized even further, including for volume and size, and analyzed under different voltages, currents, frequencies and temperatures until the design is finalized. The software can predict the new magnetic component’s losses at over 85% accuracy.

Since last year, Frenetic has been working to add an artificial intelligence assistant to the software, which will expand its abilities to accomplish a wider range of features and tasks, including full converter designs.  Molina expands, “by specifying the input voltage, the output voltage and the desired power, the system will understand what you require and generate a complete design for a power supply, a converter or an inventor, generating the circuit, the simulation files and the magnetics design. Frenetic AI is intended to free up engineers from a lot of the manual work in converter design. It automatically makes decisions when information is missing, but then allows the engineer to modify the output, by asking it to use different types of control, or technology. The SPICE or simulation files can then be downloaded to run simulations.”

If a more detailed analysis is required on the magnetics of the circuit, a page is easily accessible to provide more details of the component, including the winding, the lead wires, etc., as well as the results of the simulation in terms of proximity losses, DC resistance and total losses of the component, and even the maximum temperature that the circuit will attain.

Currently, the software is in beta, with around 600 testers. In the future, Molina sees further advances, including a search facility to see if any current off-the-shelf power supply already meets the requirements. It will also include software to run all the necessary sweep analysis and simulations, so that there is no need to use external software.

https://www.frenetic.ai/