ML Phase Current Balancer for Multiphase Power Converters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Multiphase power converters face challenges in maintaining phase current balance, especially under varying load conditions and thermal disturbances, which can lead to efficiency reduction and potential damage due to circulating currents and inductor saturation.
Innovation Solution
A machine learning-based phase current balancer using an artificial neural network is implemented to correct phase current imbalances by analyzing individual phase currents, temperature, and output voltage information, providing corrective feedback to modulate phase currents and maintain balance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional phase current balancing methods using exact same components and high quality components with very low tolerance are used, then phase current balance is improved, but device complexity and cost increase
Solution Approach 1:
The patent changes the control parameters by using an artificial neural network to dynamically adjust switching signals based on real-time phase current measurements. Instead of relying on precise component parameters, the system uses software-based parameter adjustment to achieve current balance, thereby reducing the need for component matching while maintaining balance performance.
Solution Approach 2:
The patent replaces the mechanical/component-based balancing approach with a digital/neural network-based control system. The artificial neural network processes current measurements and generates corrective switching signals, substituting the need for precise physical component matching with an intelligent control algorithm.
2Reliability
If component matching and extra margin are used to ensure phase current balance, then reliability is improved, but the solution becomes larger which is problematic for mobile devices
Solution Approach 1:
The system uses dynamic parameter adjustment through the neural network to achieve current balance without requiring oversized components. The neural network optimizes switching parameters in real-time, allowing the use of smaller, more compact components while maintaining balance performance.
3Reliability
If additional regulator circuitry is provided for balancing phase currents, then phase current balance is improved, but device complexity increases
Solution Approach 1:
The patent replaces additional analog regulator circuitry with a digital neural network implementation. The neural network, which can be implemented in software or firmware, processes current measurements and generates balancing control signals, substituting complex analog circuitry with a more compact and flexible digital solution.
4Power
If more phases are used to meet increasing power demand, then power delivery capability is improved, but phase current balancing becomes more difficult
Solution Approach 1:
The artificial neural network serves multiple functions simultaneously: it monitors all phase currents, processes the measurements, determines imbalance conditions, and generates corrective control signals for multiple phases. This universal control approach scales efficiently to accommodate increasing numbers of phases without proportionally increasing control complexity.
Data Source
AI summary
A machine learning (ML)-based phase current balancer for a multiphase power converter includes one or more inputs, one or more outputs, and an artificial neural network. The artificial neural network includes a plurality of artificial neurons and is trained to provide corrective phase current imbalance information at the one or more outputs for correcting phase current imbalance within the multiphase power converter, based on information available at the one or more inputs and indicative of individual phase currents of the multiphase power converter.