Predictive Process Control With Multivariate Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current substrate manufacturing processes face challenges in accurately predicting changes in process values for control targets and effectively utilizing these predictions in the context of Digital Twin technology for Smart Factory applications.
Innovation Solution
A management apparatus is developed, comprising a prediction model unit that learns the input-output relationship between multivariate control values and process values, and an optimization model unit that seeks to minimize differences between predicted and target values, enabling the control of control targets by calculating optimal control values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a prediction model is built to predict process values in substrate manufacturing, then predictive capability is improved, but model complexity and data processing requirements increase
Solution Approach 1:
The prediction model is segmented into two distinct components: a prediction model unit that predicts future process values, and an optimization model unit that determines optimal control values. This segmentation allows each unit to be specialized and manageable, reducing overall system complexity while maintaining high predictive capability.
Solution Approach 2:
The system implements a closed-loop feedback mechanism where predicted process values are compared with target values, and the optimization model unit adjusts control values based on this comparison. This feedback loop enables continuous improvement of prediction accuracy and control effectiveness without requiring increasingly complex model structures.
2Manufacturing precision
If optimization is performed to minimize differences between predicted and target values, then control precision is improved, but computational time and processing load increase
Solution Approach 1:
The prediction model unit performs preliminary action by predicting future process values before actual processing occurs. This allows the optimization model unit to pre-calculate optimal control values based on predicted outcomes, reducing the need for complex real-time optimization and thereby decreasing computational time while maintaining high control precision.
3Reliability
If multivariate control values are used to control the control target, then control effectiveness is improved, but system complexity and data management difficulty increase
Solution Approach 1:
The management apparatus is designed with multi-functionality, where the prediction model unit and optimization model unit work together in a unified system. This universal design allows the system to handle multivariate control values effectively through integrated processing, reducing the need for separate specialized systems and thereby managing complexity while maintaining control effectiveness.
Data Source
AI summary
A mechanism of predicting a change in process values with respect to a control target and using a prediction result is provided. A management apparatus includes a prediction model unit, with respect to which an input-output relationship between multivariate control values at a time T with respect to a control target and multivariate process values at a time T+ΔT with respect to the control target has been learned; and an optimization model unit configured to seek multivariate control values of the time T that minimize respective differences between the multivariate process values at the time T+ΔT output from the prediction model unit and corresponding target values, and control the control target using the multivariate control values of the time T that have been sought. The prediction model unit is configured to, in response to a request from an agent, predict multivariate process values of a time after an elapse of a time ΔT with respect to the control target in a case where the control target is controlled with designated control values, and output the multivariate process values that have been predicted to the agent, the agent managing the prediction model unit.