Prediction Model Updating for Small-Batch Process Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing prediction control methods using prediction models struggle to achieve high accuracy in small-scale production devices due to frequent setup changes, leading to low prediction accuracy and increased likelihood of producing low-quality products.
Innovation Solution
A prediction control development device that acquires and analyzes time-series data to generate and update prediction models, continuously evaluating and improving their accuracy by adjusting instruction values based on predicted control amounts, ensuring high accuracy even in small-scale production devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a prediction model is generated using collected data in small-scale production devices, then prediction control can be implemented, but frequent setup changes cause data deviation and reduce prediction accuracy
Solution Approach 1:
The system performs preliminary actions by collecting data not only during normal operation but also during setup changes and transitions. This advance data collection ensures that the prediction model has sufficient training data covering various operational states, allowing it to maintain high prediction accuracy even when setup changes occur frequently in small-scale production devices.
Solution Approach 2:
The system implements feedback mechanisms by continuously monitoring prediction accuracy and using the collected data to update and refine the prediction model. This feedback loop allows the model to adapt to frequent setup changes and maintain reliable prediction performance despite the varying operational conditions in small-scale production environments.
2Measurement precision
If data collection time is extended to improve prediction model accuracy, then more comprehensive data is obtained, but production time is lost due to frequent setup changes
Solution Approach 1:
The system maintains continuity of useful action by collecting data continuously during all operational phases including setup changes and transitions, rather than stopping data collection during these periods. This approach ensures that valuable data is captured during setup changes without extending the overall data collection time, thereby maintaining both data quality and production efficiency.
Solution Approach 2:
The system performs preliminary data collection during setup changes and transitions before normal operation resumes. This allows comprehensive data to be gathered during what would otherwise be non-productive periods, improving measurement precision without causing additional time loss to production.
3Productivity
If a prediction model is generated with limited data, then data collection time is reduced, but the prediction model lacks accuracy and reliability
Solution Approach 1:
The system applies dynamics by making the prediction model updateable and adaptable. The model is not static but can be refined and updated using additional data collected during setup changes and transitions. This dynamic approach allows the system to maintain high productivity during initial deployment while improving prediction model accuracy over time as more data becomes available.
Solution Approach 2:
The system uses feedback to continuously improve the prediction model by incorporating data collected during various operational phases. This allows the model to evolve from an initial state with limited data to a more accurate and reliable model over time, without requiring extended data collection periods that would reduce productivity.
Data Source
AI summary
A prediction control development device according to one aspect of the present invention generates a prediction model of a control amount by analyzing first time-series data of the control amount and provides the generated prediction model to a controller. The prediction control development device acquires second time-series data showing transition of a value of the control amount during prediction control using the prediction model, and evaluates prediction accuracy of the prediction model based on a difference between a prediction value by the prediction model and a value of the second time-series data. When the prediction accuracy of the prediction model is not allowable, the prediction control development device analyzes the second time-series data to newly generate a prediction model of the control amount, and provides the newly generated prediction model to the controller.