Wound-Rotor Synchronous Machine Torque Estimation With One Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the torque of an electric machine with a wound rotor in motor vehicles with electric or hybrid propulsion face challenges due to strong magnetic coupling, leading to inaccuracies in torque estimation, particularly with synchronous machines, and require complex neural network structures that are resource-intensive and prone to errors.
Innovation Solution
A method utilizing a single neural network that processes current values from the stator and rotor phases, transformed into a two-phase reference, to estimate torque by outputting a global magnetic flux component, allowing direct parameter determination from torque measurements on a test bench, reducing computational and memory requirements and estimation errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a two-neural network structure is used to estimate torque components, then the estimation capability is improved, but the device complexity and computational resources increase
Solution Approach 1:
The patent combines two separate neural networks into a single neural network that directly estimates the complete torque vector. This merging reduces device complexity while maintaining estimation accuracy by integrating the functionality of multiple networks into one unified structure that processes the same input data to produce the full torque estimate.
2Ease of manufacture
If a theoretical machine model is used for training neural networks, then the training process is simplified, but the torque estimation accuracy deteriorates due to model imperfections
Solution Approach 1:
The patent implements a feedback mechanism where the single neural network is trained using actual measured torque values from the machine. The network's torque estimates are compared with real measurements, and the training adjusts the network parameters to minimize the difference between estimated and actual torque, thereby improving accuracy while maintaining training simplicity.
3Measurement precision
If multiple neural networks are deployed for torque component estimation, then the estimation coverage is improved, but the memory and computing resources required increase
Solution Approach 1:
The patent merges multiple neural networks into a single network that estimates the complete torque vector in one operation. This reduces computational resource consumption by eliminating redundant processing steps and memory requirements associated with running multiple separate networks, while still providing comprehensive torque component estimation.
Data Source
Figure 1~2
AI summary
The invention relates to a method for estimating the torque of a wound-rotor electric machine, comprising the steps of: receiving values of the currents of the phases (la, Ib, lc) of the stator and of the rotor current (If) of said electric machine, converting said values of the currents of the phases of the stator to a two-phase coordinate system, determining a current vector containing said converted current values (Id, Iq) and said value of the rotor current (If), inputting said current vector into a single neural network (RDN), which delivers as output a quantity (PhiRDN) corresponding to an overall flux derived from the two components in the two-phase coordinate system of the magnetic flux induced by the electric machine and the value of which is determinable, for a given current vector, from a torque measurement taken on a testbed for testing said machine.