Causal Relationship Model Building for Manufacturing Quality Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In manufacturing, it is challenging to quantify the causal relationship between upstream and downstream steps in a manufacturing flow to ensure that control data and observation data fall within suitable ranges, leading to difficulties in improving manufacturing quality, as existing methods require all data for prediction and do not allow real-time calculation or partial model modification.
Innovation Solution
A causal relationship model building system that processes monitor data from multiple steps of a manufacturing flow to predict quality inspection results and specify the allowable ranges of control and observation data, allowing for real-time adjustment and improvement of prediction accuracy by identifying and replacing low-precision parts with more accurate models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If regression model is used to build causal relationship model, then prediction can be performed, but all data items must be input to calculate prediction value, preventing real-time control in upstream steps
Solution Approach 1:
The patent divides the causal relationship model into multiple parts corresponding to different manufacturing steps. Each part processes only the data relevant to that step, allowing upstream steps to perform prediction and control using only their local data without waiting for all downstream data to be collected.
Solution Approach 2:
The patent performs preliminary calculation of allowable ranges for monitor data at each manufacturing step based on the causal relationship model. This allows upstream steps to determine control targets in advance without requiring real-time input from all subsequent steps, enabling real-time control decision-making.
2Measurement precision
If entire model is modified to improve prediction accuracy, then quality impact prediction can be enhanced, but calculation cannot be completed in real time for large-scale manufacturing steps
Solution Approach 1:
The patent segments the large-scale causal relationship model into multiple smaller parts, each representing a specific manufacturing step or subsystem. This allows selective modification and calculation of only the necessary parts rather than recalculating the entire model, significantly reducing computation time while maintaining prediction accuracy for the affected quality attributes.
Solution Approach 2:
The patent implements partial model modification where only the specific parts of the causal relationship model that are relevant to the current quality issue are updated and recalculated. This partial action approach maintains real-time calculation capability while improving prediction accuracy for the specific quality impact being analyzed.
3Measurement precision
If causal relationship model is built using all monitor data from multiple steps, then comprehensive quality prediction is achieved, but it becomes difficult to quantify the relationship between upstream monitor data and downstream quality results
Solution Approach 1:
The patent structures the causal relationship model as a series of segmented parts, where each part represents a specific manufacturing step and its relationship to subsequent steps. This segmentation makes the complex relationships between upstream monitor data and downstream quality results more manageable and interpretable, while still incorporating data from all manufacturing steps.
Solution Approach 2:
The patent introduces intermediate variables and relationships between manufacturing steps that act as mediators connecting upstream monitor data to downstream quality results. These intermediaries break down the complex direct relationships into manageable segments, making the causal relationships quantifiable and interpretable while maintaining comprehensive quality prediction capability.
Data Source
AI summary
A causal relationship model building system includes a computer which processes information for building a causal relationship model relating to a manufacturing flow of an object to be controlled. The computer builds the causal relationship model by using monitor data representing a state of each of a plurality of steps of the manufacturing flow, and quality data as a result of an inspection step, and specifies an allowable range of the monitor data so as to satisfy a target value of the quality data, by using the causal relationship model and the target value, from prediction based on a causal relationship between a plurality of pieces of the monitor data. The computer graphically displays information including the causal relationship model and the allowable range of the monitor data on a display screen.