Learned Compensation Model for Press Die Height Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In press systems, adjusting the optimal die height requires multiple trial-and-error iterations for each plate thickness, and the control variable is influenced by various object parameters such as hardness and temperature, necessitating extensive testing to confirm the appropriateness of compensation amounts.
Innovation Solution
A learning device that generates a candidate compensation amount using data-driven control methods like VRFT, FRIT, or ERIT, allowing for efficient learning and evaluation of control models without actual operations, thereby eliminating inappropriate models and optimizing control variable quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If trial and error method is used to adjust die height for each plate thickness, then appropriate compensation amount can be obtained, but huge amount of time is required
Solution Approach 1:
The patent performs preliminary learning offline to generate a learned model before actual production. The learned model is created by collecting operation data including target values, command values, and control variables, then training a model to predict appropriate compensation amounts. This preliminary action eliminates the need for time-consuming trial and error during actual manufacturing, as the model can directly predict compensation amounts for new plate thicknesses.
Solution Approach 2:
The patent creates a virtual copy of the press machine's control system in the form of a learned model. This model replicates the relationship between plate thickness and optimal compensation amount by learning from historical operation data. Instead of physically performing trial and error adjustments on the actual machine, the system uses this virtual model to predict compensation amounts, significantly reducing the time required while maintaining accuracy.
2Reliability
If extensive testing is performed to confirm compensation amount appropriateness for various object parameters, then control quality can be ensured, but processing time increases
Solution Approach 1:
The patent implements a feedback mechanism where the learned model continuously learns from actual operation results. The system collects operation data including target values, command values, and control variables during production, then uses this feedback to refine and update the learned model. This allows the system to ensure control quality through continuous learning while maintaining high productivity, as the model improves over time without requiring extensive testing for each new parameter combination.
Solution Approach 2:
The patent changes the approach from physical parameter adjustment through trial and error to computational parameter prediction using a learned model. The model learns the relationships between various object parameters (plate thickness, hardness, temperature, material) and optimal compensation amounts from historical data, then predicts appropriate parameters for new conditions. This parameter-based prediction approach ensures reliability while dramatically improving processing speed compared to extensive physical testing.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
This learning device provides a learned model to an adjuster containing a learned model leaned in such a manner that a controller outputs a predetermined compensation amount to the controller, on the basis of the specific parameter of an object to be processed, in a control system comprising a controller for outputting a command value obtained by compensating a target value on the basis of a compensation amount and an object to be controlled which is controlled so as to perform predetermined processing on an object to be processed. The learning device comprises: an evaluation unit which acquires operation data that includes the target value, command value and control variable and evaluates the quality of the control variable; a learning unit which generates compensation candidates on the basis of the operation data, and learns, as teacher data, the generated compensation amount candidates and specific parameters for the object to be processed, and generates a learned model; and a setting unit which provides the learned model to the adjuster if the quality evaluated by the evaluation unit is within a permissible range, on the basis of a control variable when a command value obtained by compensating a target value on the basis of the compensation amount output by the generated learned model is imparted to the object to be controlled.