Researchers at Chiba University have developed a machine learning-based method for designing concrete power transfer systems that stabilize the output regardless of load changes. This property is also known as a load-independent operation.
Wireless power transfer systems exist in smartphones, electric toothbrushes and IoT sensors. They use electromagnetic fields to transmit electrical energy wirelessly without using physical connectors. They have also been around since the time of Nikola Tesla.
The importance of load independence
Traditional WPT systems require inductors and capacitors to have accurate component values for stable operation. These values are typically driven from complex analytical equations based on idealized conditions.
Factors such as parasitic capacitance, manufacturing resistance, and environmental conditions can adversely affect these calculations in real-world scenarios. This leads to unstable output voltage and losses of zero voltage switching (ZVS), which is considered an important efficiency factor.
Load-independent operation allows for stable ZVS and output voltage even when the load changes.
New solutions using machine learning
Professor Hirao Sekiya, who leads a research team at Chiba University, proposed a machine learning-based design method for designing WPT systems with load-independent (LI) operations.
This approach describes a WPT circuit using differential equations that capture how voltage and current evolve within a system. Consider real-world component properties for this purpose.
These equations are resolved numerically in stages until the system reaches steady state. The evaluation function scores the system's performance on key metrics: output voltage stability, efficiency, and total harmonic distortion.
The genetic algorithm then adjusts the circuit parameters to improve the score. This algorithm is a type of machine learning inspired by natural selection. The optimization cycle is repeated until the system meets the LI operation requirements.
Place the method to test
The researchers applied a design approach to a Class-EF WPT system that combines a Class-EF inverter and a Class-D rectifier. In traditional setups, ClassF inverters can only maintain ZV at their rated operating point. Changing the load usually causes ZV to fail and reduces efficiency.
Machine learning design LI system limited voltage variation to less than 5% over different load variations. This figure is considerably lower compared to the 18% fluctuation achieved with traditional systems.
It also managed ZV and high efficiency well under different load conditions. It delivers 23 watts of power at 6.78 MHz with 86.7% efficiency. The performance of the system has also been improved even with optical loads thanks to precise modeling of the diode parasitic capacitance.
Detailed power loss analysis revealed that transmission coil losses remained roughly constant at various loads. This indicates that the system remains current and is a key factor in efficiency.
Larger scope
In the future, researchers affirm that the meaning of their work can be easily expanded beyond WPT.
“We believe that the results of this study are an important step towards a fully wireless society,” says Professor Sekiya, who explains his broad vision of WPT.
Furthermore, the operation of LI makes WPT systems easy to build, thereby reducing costs and scale. Our goal is to make WPT general within the next five to ten years,” he continued.
