Rotor Position Estimation via Learned Current Function
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor-based systems for estimating rotor position in permanent magnet motor drives are costly, complex, and sensitive to physical non-idealities, making them unreliable for sensorless control.
Innovation Solution
A method for designing an estimation module that learns to estimate the angular position of a rotor from phase currents, eliminating the need for modeling the electrical machine by using machine learning techniques to associate phase currents with angular positions, and implementing this learned function for rotor position estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a mechanical sensor (encoder or resolver) is integrated with the electric motor to measure rotor position, then measurement precision of rotor position is improved, but device complexity and cost increase
Solution Approach 1:
The patent extracts the position measurement function from physical sensors (encoders or resolvers) and implements it through a software estimator that calculates rotor position based on electrical measurements. This removes the mechanical sensing components while maintaining the essential measurement capability through computational methods.
Solution Approach 2:
The patent replaces the mechanical sensor system with an electrical/software-based estimation system. Instead of using physical encoders or resolvers to directly measure position, the system uses electrical measurements (currents, voltages) combined with machine models to computationally estimate rotor position, substituting mechanical measurement with electrical measurement and software processing.
2Device complexity
If model-based sensorless control is used to eliminate physical sensors, then device complexity is reduced, but reliability deteriorates due to sensitivity to non-ideal physical attributes
Solution Approach 1:
The patent employs feedback mechanisms where the estimated rotor position and speed are continuously used to update control decisions and refine estimation accuracy. The system monitors electrical measurements and adjusts estimation parameters based on actual system behavior, creating a closed-loop estimation process that improves robustness against model inaccuracies and non-ideal conditions.
Solution Approach 2:
The patent adapts estimation parameters and model parameters based on operating conditions such as speed, load, and temperature. By dynamically adjusting parameters rather than using fixed values, the system maintains accuracy across varying conditions and compensates for non-ideal physical attributes that change with operating state.
3Ease of manufacture
If physical sensors are eliminated for sensorless control, then cost is reduced, but measurement precision deteriorates due to sensitivity to non-linearities and signal delays
Solution Approach 1:
The patent performs preliminary characterization of the electrical machine during manufacturing or initial operation to establish accurate machine parameters and models. This pre-calibration process captures the specific non-ideal attributes of each machine, allowing the sensorless estimator to compensate for these characteristics during normal operation and maintain high measurement precision without physical sensors.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A method for designing an estimation module (118) for estimating an angular position of a rotor (104) of an electrical machine (102) relative to a stator (106) of the electrical machine (102), on the basis of phase currents ([i]) intended, respectively, to pass through stator phases (A, B, C) of the electrical machine (102), the design of the estimation module (118) comprising: - measuring the phase currents ([i]) and the angular position of the rotor (104), - learning (306) a generic function from learning measurements in order to obtain an instructed function, the learning process being carried out by using the phase currents ([i]) from the learning measurements as input data for the generic function and the angular position from the learning measurements as output data for the generic function, and - implementing the instructed function in an estimation module (118).