Servo Control Gain Tuning Using Nyquist-Based Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning devices for servo motor controllers adjust velocity gains and filters based on phase and gain margins separately, making them susceptible to measurement fluctuations and failing to consider both stability and responsiveness simultaneously.
Innovation Solution
A machine learning device that optimizes filter coefficients and feedback gains by acquiring state information including output/input gain and phase delay, determining rewards based on whether the Nyquist path passes through a predetermined gain and phase margin on a complex plane, and updating value functions accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If phase margin and gain margin are evaluated separately as point evaluations, then the evaluation process is simple, but the system is easily affected by measurement fluctuations and cannot ensure stable optimization
Solution Approach 1:
The patent combines phase margin evaluation and gain margin evaluation into a unified Nyquist curve-based evaluation system. Instead of treating them as separate point evaluations, the system evaluates the entire Nyquist curve against a predetermined closed curve that encloses the critical point (-1, 0), thereby integrating both stability margins into a single comprehensive assessment that reduces sensitivity to measurement fluctuations.
Solution Approach 2:
The patent transitions from evaluating phase margin and gain margin as separate scalar values (0D or 1D) to evaluating the Nyquist curve as a continuous path in the complex plane (2D). This dimensional transformation allows the system to consider the relationship between phase and gain across all frequencies simultaneously, providing a more robust evaluation that is less susceptible to measurement noise at individual frequency points.
2Ease of operation
If velocity gain and filter are adjusted based on separate point evaluations, then adjustment is straightforward, but both phase margin and gain margin cannot be considered simultaneously
Solution Approach 1:
The patent merges the evaluation of phase margin and gain margin into a single Nyquist curve evaluation framework. The predetermined closed curve in the complex plane simultaneously constrains both stability margins, allowing the machine learning system to optimize velocity gain and filter parameters while considering both margins together through a unified reward function.
Solution Approach 2:
The Nyquist curve evaluation serves multiple functions simultaneously: it evaluates both phase margin and gain margin, provides a comprehensive stability assessment across all frequencies, and generates a single reward signal that guides the optimization of multiple parameters (velocity gain and filter coefficients). This universal evaluation method replaces multiple separate evaluation processes.
3Productivity
If machine learning optimizes filter coefficients and feedback gain, then system performance can be improved, but measurement fluctuations can easily affect the evaluation function
Solution Approach 1:
The patent moves the evaluation from individual frequency point measurements (0D/1D) to a comprehensive Nyquist curve analysis in the complex plane (2D). This dimensional elevation integrates information across all frequencies, making the evaluation inherently more robust to measurement fluctuations at any single frequency point while still providing comprehensive system performance optimization.
Data Source
AI summary
A machine learning device which performs machine learning to optimize at least one of a coefficient of a filter and a feedback gain, the machine learning device comprising: a state information acquiring unit which acquires state information including the at least one of the coefficient and the feedback gain and including input/output gain and input/output phase delay of a servo control device; an action information output unit which outputs action information including adjustment information for the at least one of the coefficient and the feedback gain; a reward output unit which obtains and outputs a reward on the basis of whether a Nyquist plot calculated from the input/output gain and the input/output phase delay passes through the inside of a closed curve; and a value function updating unit which updates a value function on the basis of the value of the reward, the state information, and the action information.