Wire Straightener Feedforward Control With Adaptive Curvature Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing wire straightening device lacks accuracy in feedforward control due to non-uniform movement of the straightening roller and failure to consider variations in factors affecting the straightening curvature beyond the bobbin winding position.
Innovation Solution
A control device that determines a feedforward compensation value based on feedback operation amounts, using a predictive model and machine learning to optimize the feedforward control, considering various factors affecting the straightening curvature, such as roller positions, wire speed, and curvature.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If feedforward compensation is based on a linear expression of bobbin winding position, then the control system is simple, but the straightening accuracy deteriorates because the movement amount of the straightening roller is not uniform with respect to the bobbin winding position
Solution Approach 1:
The patent changes the functional form of the feedforward compensation from a linear expression to a polynomial expression (e.g., quadratic or cubic) of the bobbin winding position. This allows the compensation value to vary non-linearly, matching the actual non-uniform relationship between winding position and required roller movement, thereby improving straightening accuracy while maintaining computational simplicity.
Solution Approach 2:
The patent combines feedforward compensation with feedback control, where the feedback operation amount based on measured straightening curvature is added to the feedforward compensation value. This closed-loop approach corrects residual errors and adapts to variations in factors affecting straightening curvature, significantly improving overall straightening accuracy.
2Device complexity
If only bobbin winding position is considered for feedforward compensation, then the control factors are few, but the straightening accuracy deteriorates because other factors affecting straightening curvature are not considered
Solution Approach 1:
The patent extends the feedforward compensation to consider multiple factors simultaneously, including bobbin winding position, straightening roller position, wire speed, and initial curvature. By making the compensation system multi-functional and comprehensive, it accounts for the combined effects of these factors on straightening curvature, improving accuracy without significantly increasing system complexity.
Solution Approach 2:
The patent transitions from a one-dimensional compensation model (only winding position) to a multi-dimensional model that incorporates additional parameters such as roller position, wire speed, and curvature. This dimensional expansion allows the system to capture the complex, multi-factor nature of the straightening process and achieve higher precision.
3Reliability
If machine learning is performed continuously with all data, then the model adapts well, but the processing time and computational load increase
Solution Approach 1:
The patent performs machine learning selectively rather than continuously on all data. It uses partial action by triggering learning only when teaching data is acquired under specific conditions (e.g., when certain thresholds are met or at scheduled intervals), balancing model adaptation with processing efficiency. This avoids unnecessary computational overhead while maintaining model reliability.
Data Source
AI summary
Accuracy of feedforward control for a wire straightener is improved. A feedback operation amount to a straightener is determined by a feedback control means based on an error between a target curvature and a straightening curvature. A feedforward compensation means determines a feedforward compensation value from a measurement value of the straightener using a prediction model. A learning means performs machine learning on the prediction model using teaching data. The learning means adds at least one combination including the measurement value and a manipulated variable when an absolute value of an error is smaller than a reference value to the teaching data.