Movable discs. YouTube/Stanford Energy
A research team from Chiba University has developed a machine learning-based method to design Wireless Power Transfer systems that keep their output stable regardless of load changes. This property is also known as load-independent operation.
Wireless power transfer systems exist in smartphones, electric toothbrushes, and IoT sensors. They use electromagnetic fields to send electrical energy wirelessly, without using physical connectors; they have also been around since the days of Nikola Tesla.
The importance of load independence
Traditional WPT systems need inductors and capacitors to have precise component values in a bid to achieve stable operation. These values are usually driven from complex analytical equations based on idealized conditions.
Factors such as parasitic capacitance, manufacturing tolerances, and environmental conditions can negatively impact these calculations in real-world scenarios. This leads to unstable output voltage and loss of zero voltage switching (ZVS), regarded as a critical efficiency factor.
Load-independent operation can keep the ZVS and output voltage stable even when the load changes.
A novel solution using machine learning
Professor Hiroo Sekiya, leading the research team at Chiba University, has proposed a machine-learning-based design method for designing a WPT system with load-independent (LI) operation.
This approach describes the WPT circuit using differential equations that capture how voltages and currents evolve within the system. It takes into account real-world component characteristics for this purpose.
These equations are solved numerically, step by step, until the system reaches steady-state conditions. An evaluation function then scores the system’s performance on key metrics: output voltage stability, efficiency, and total harmonic distortion.
Then, a genetic algorithm adjusts circuit parameters to improve the score. This algorithm is a type of machine learning inspired by natural selection. The optimization cycle repeats until the system meets the LI operation requirements.
Putting the method to test
The researchers applied their design approach to a class-EF WPT system, which combines a class-EF inverter with a class-D rectifier. In a conventional setup, the class-EF inverter can maintain ZVS only at its rated operating point. Changing the load typically causes ZVS to fail and efficiency to drop.
The machine-learning designed LI system restricted voltage fluctuations to under 5 percent across different load variations. This figure is significantly low compared to the 18 percent fluctuation achieved in traditional systems.
It also managed ZVS and high efficiency successfully under different load conditions. delivering 23 watts of power with 86.7 percent efficiency at 6.78 MHz. The system’s performance improved even at light loads, thanks to accurate modeling of diode parasitic capacitance.
A detailed power-loss analysis revealed that the transmission coil’s losses remained nearly constant across different loads — a sign that the system kept output current steady, a key factor in efficiency.
A larger scope
Looking ahead, the researchers are positive that the implications of their work could easily extend beyond WPT.
“We are confident that the results of this research are a significant step toward a fully wireless society,” said Prof. Sekiya, describing his broad vision regarding WPT.
“Moreover, due to LI operation, the WPT system can be constructed simply, thereby reducing the cost and size. Our goal is to make WPT commonplace within the next 5 to 10 years,” he continued.
The research demonstrates how AI can help automate and optimize hardware design by combining accurate physical modeling with evolutionary algorithms.
The study was published online in the journal IEEE Transactions on Circuits and Systems I: Regular Papers.
MasterCard