Tree-Based Multimodal Regression for Changing Manufacturing Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing manufacturing defect prediction systems struggle to accurately predict defect levels due to changing manufacturing conditions, which alter the relationships between input data and predictions.
Innovation Solution
A tree-structured multimodal regressor system is employed, which generates and selects multiple machine learning models based on class labels assigned by a classifier. This system addresses changing manufacturing conditions by using a plurality of regressors to predict defect levels, achieving higher accuracy than single-regressor models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple machine learning models are used to predict defect levels under changing manufacturing conditions, then prediction accuracy is improved, but device complexity increases
Solution Approach 1:
The system segments the prediction task by dividing manufacturing conditions into different clusters and assigning specialized machine learning models to each cluster. This allows each model to focus on specific conditions, improving prediction accuracy for those conditions while keeping individual model complexity manageable. The segmentation is achieved through clustering algorithms that group similar manufacturing conditions together.
Solution Approach 2:
The system dynamically selects which machine learning model to use based on the current manufacturing conditions. Instead of using a single static model, the system adapts by choosing the most appropriate pre-trained model for the current condition cluster, allowing the system to respond dynamically to changing conditions while maintaining reasonable complexity through model selection rather than model integration.
2Measurement precision
If multiple machine learning models are trained and maintained, then prediction accuracy under varying conditions is improved, but computation power consumption increases
Solution Approach 1:
The system performs preliminary action by pre-training multiple machine learning models offline on different manufacturing condition clusters before deployment. During actual prediction operations, the system only needs to classify the current conditions and select the appropriate pre-trained model, avoiding the computational burden of training models in real-time. This shifts computation from online to offline, reducing runtime energy consumption.
Solution Approach 2:
The system extracts and separates the model selection function from the prediction function. By using a lightweight classifier to identify the appropriate condition cluster and then selecting the corresponding pre-trained model, the system avoids the computational overhead of evaluating multiple complex models for every prediction, thereby reducing computation power consumption while maintaining accuracy.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
A system (102) and method for making predictions relating to products manufactured via a manufacturing process are disclosed. A processor (108) receives input data (100) and makes a first prediction based on the input data. The processor identifies a first machine learning model from a plurality of machine learning models based on the first prediction. The processor further makes a second prediction based on the input data and the first machine learning model, and transmits a signal (104) to adjust the manufacturing of the products based on the second prediction.