Production Process Quality Prediction Using State-Parameter Mediation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting product quality parameters using neural networks face challenges with overtraining when numerous process condition parameters are input, leading to inaccurate predictions, especially when using functions or tables, which complicate configuration and hinder accurate quality parameter estimation.
Innovation Solution
The method involves dividing process condition parameters into operating condition parameters and state parameters, using a two-step process to predict quality parameters through a neural network, first converting operating conditions to state parameters and then to quality parameters, thereby reducing the number of input parameters and preventing overtraining.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a large number of process condition parameters are input to the prediction model, then comprehensive quality prediction is attempted, but overtraining occurs and prediction accuracy deteriorates
Solution Approach 1:
The patent segments the process condition parameters into two distinct groups: operating condition parameters (controllable inputs) and state parameters (intermediate states). This segmentation allows the neural network to process information in a structured manner, first learning the relationship between operating conditions and state, then between state and quality parameters, thereby preventing overtraining while maintaining comprehensive prediction capability
Solution Approach 2:
The patent introduces state parameters as an intermediary between operating condition parameters and quality parameters. This intermediary layer acts as a bridge that decomposes the complex direct mapping into two simpler relationships, reducing the computational burden on the neural network and preventing overtraining while preserving the ability to predict multiple quality parameters
2Device complexity
If a function or table is used instead of a neural network to obtain quality parameters, then model complexity is reduced, but configuration becomes complicated and prediction accuracy deteriorates
Solution Approach 1:
The patent segments both the model structure and the parameter space. The neural network is divided into two sequential processing stages (operating conditions → state, state → quality), and the parameters are segmented into operating condition parameters and state parameters. This segmentation simplifies the configuration process by reducing the number of parameters each stage must handle, while maintaining prediction accuracy
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
An object of an optimization support device, a method, and a program is to enable accurate prediction of various conditions in a product production process. The optimization support device includes a first conversion unit that converts an operating condition parameter indicating an operating condition of a process for producing a product into a state parameter indicating a state of the process, and a second conversion unit that converts the state parameter into a quality parameter indicating a quality of the product.