Controller with Dual Correction Units for Abnormal Value Prevention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-order mathematical models obtained through machine learning for feed-forward control in servomotor and spindle motor controllers may produce abnormal correction amounts, which can lead to unsatisfactory control outcomes.
Innovation Solution
A controller is designed with dual correction amount computation units, one based on a high-order mathematical model determined by machine learning and another based on a lower-order model with parameters set by a different method, along with an abnormality detection unit to select appropriate correction amounts and prevent abnormal control.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-order mathematical models with machine learning are used for feed-forward control, then positional deviation and speed deviation decrease, but abnormal correction amounts may occur leading to unreliable control
Solution Approach 1:
The control system is segmented into multiple correction amount computation units, each handling different aspects of feed-forward control. One unit processes high-order models while another handles lower-order models, allowing the system to benefit from high precision while maintaining reliability through modular verification.
Solution Approach 2:
An abnormality detection unit acts as an intermediary between the correction amount computation units and the control execution. This mediator verifies the correctness of correction amounts before they are applied, preventing abnormal values from compromising system reliability while allowing high-order models to improve precision.
2Speed
If high-order mathematical models are used for feed-forward control, then trackability with respect to position command increases, but determination of model parameters becomes difficult
Solution Approach 1:
The feed-forward control is divided into multiple computation units with different model orders. This segmentation allows the system to use high-order models for trackability improvement while using lower-order models for simpler parameter determination and verification.
Solution Approach 2:
Multiple correction amount computation units essentially create parallel copies of the feed-forward control function with different model complexities. This allows the system to leverage both high-order and lower-order model characteristics without requiring a single complex model to handle all aspects.
3Ease of operation
If machine learning is used to determine control parameters, then adjustment complexity is reduced, but abnormal correction amounts may be generated
Solution Approach 1:
The system employs self-service through automatic abnormality detection and verification mechanisms. The abnormality detection unit automatically verifies correction amounts generated by machine learning-based computation units, reducing the need for manual parameter adjustment while ensuring reliability through automated validation.
Solution Approach 2:
A feedback loop is established where the abnormality detection unit continuously monitors correction amounts from machine learning-based computation units. When abnormalities are detected, the system can adjust or reject the machine learning outputs, creating a feedback mechanism that maintains reliability while preserving the ease of parameter adjustment provided by machine learning.
Data Source
AI summary
A controller capable of preventing a control target from being controlled by an abnormal value output from a mathematical model is provided. A controller includes a control unit, the control unit including: a first correction amount computation unit that computes a first correction amount for correction from a command value to a second command value; a second correction amount computation unit that computes a second correction amount for correction from the command value to the second command value; and a correction amount selecting unit that selects either one of the first correction amount and the second correction amount. The first correction amount computation unit computes the first correction amount using a first mathematical model configured by machine learning, and the second correction amount computation unit computes the second correction amount using a second mathematical model configured by a method different from that of the first correction amount computation unit.