The research team at Chiba University and their collaborators have developed a completely numerical machine learning-based design method that greatly improves the stability and efficiency of wireless power transfer (WPT) systems.
This new approach overcomes several long-standing challenges in maintaining a constant output voltage across a variety of loads. This is an important step towards a practical and cost-effective wireless power solution.
Issues in load fluctuations in WPT systems
Wireless power transfer systems transmit power without a physical connector and rely on the electromagnetic field between the transmitter and receiver coils. Although already applied to everyday devices such as smartphones, electric toothbrushes, and sensors, WPT systems often suffer from voltage instability when electrical loads change. This variation can cause loss of efficiency and reduce system performance, especially in more demanding or variable use cases.
Typically, circuit components such as capacitors and inductors need to be accurately adjusted to achieve load-independent (LI) operations. However, these values are usually calculated using idealized analytical equations that cannot explain the behavior of actual circuits that contain parasitic elements and nonlinear properties. As a result, the actual implementation of Li operations remains a critical engineering challenge.
A fully numerical AI-driven approach
To address these limitations, researchers led by Professor Hirao Sekiya of the Faculty of Information Studies at Chiba University have developed a new machine learning-based method that uses numerical optimization instead of analytical formulas. System performance can be evaluated and staged optimized by modeling WPT circuits with differential equations that capture real-world behavior over time.
The team employed genetic algorithms to adjust circuit parameters based on evaluation functions. This function takes into account output voltage stability, power delivery efficiency, and harmonic distortion. Iteratively refine the design in this way allows the system to reach optimal steady-state behavior without relying on idealized assumptions.
This fully numerical method illustrates a change in WPT design, highlighting how machine learning and artificial intelligence can quickly and quickly accelerate the development of power electronics, leading to more robust and adaptable systems.
Experimental success
To verify their methods, researchers applied it to the Class-EF WPT system, a design that combines Class-F inverters and Class-D rectifiers.
Traditional class-EF systems with no load-independent optimization can only maintain zero voltage switching (ZVS) under certain load conditions. Deviating load will result in loss of ZVS, reducing overall efficiency and stability.
The numerically optimized Li-Class-EF system maintained both ZV and steady-state output voltage over a wide range of load conditions. The researchers reported output voltage fluctuations of less than 5%. This is a significant improvement compared to the 18% fluctuation observed in traditional systems.
The system achieved high power delivery efficiency of 86.7% at 6.78 MHz and delivered power of over 23 W. This design also demonstrated consistent performance at optical loads with effective management of parasitic diode capacitance and stable power dissipation of the transmitting coil.
A more practical wireless future
The implications of this study go beyond performance improvements alone. Machine learning-driven design processes simplify circuit development, reduce cost and component sensitivity, and bring WPT technology closer to widespread commercial adoption.
The team wants applications in home appliances, medical devices and industrial systems where cable-free power delivery can provide convenience, safety and flexibility.
As the world moves towards more innovative and connected systems, this advances show a future in which wireless power is practical and ubiquitous. With continued research and improvements, a fully wireless infrastructure could become a standard feature of everyday life within the next decade.
