Concept Interval Clustering for Defect Visibility Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional machine learning models for predicting manufacturing defects fail to account for model drift caused by changing manufacturing conditions, leading to decreased prediction accuracy over time.
Innovation Solution
Implement a system that clusters trace data into concept intervals (eras) and trains separate machine learning models for each interval, using an era classifier to identify the appropriate model for predicting defect visibility levels, thereby minimizing regression errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are used for predicting manufacturing defects, then initial prediction accuracy is achieved, but prediction accuracy deteriorates over time due to model drift caused by changing manufacturing conditions
Solution Approach 1:
The patent implements dynamic model adaptation by continuously monitoring manufacturing conditions and updating the machine learning model when drift is detected. The system transitions from a static model to a dynamic one that adapts to changing conditions, using techniques like online learning and model retraining to maintain prediction accuracy over time.
Solution Approach 2:
The system incorporates feedback mechanisms by continuously comparing actual defect outcomes with model predictions, detecting prediction errors and model drift, and using this feedback to trigger model updates. This closed-loop feedback ensures the model maintains reliability by correcting deviations caused by changing manufacturing conditions.
2Reliability
If separate machine learning models are trained for different time intervals to address model drift, then prediction reliability is improved, but system complexity increases
Solution Approach 1:
The patent segments the manufacturing process into different time intervals or concepts, training separate machine learning models for each segment. This segmentation allows each model to specialize in specific conditions, improving reliability while managing complexity through modular model architecture and selective deployment based on current manufacturing conditions.
Data Source
Figure 1
Figure 2A~2B
Figure 2C
AI summary
Systems and methods for making predictions relating to products manufactured via a manufacturing process. A processor receives a plurality of input vectors associated with a plurality of output values and a plurality of time intervals. The processor clusters the plurality of input vectors based on the time intervals associated with the input vectors. The processor trains a machine learning model for each time interval of the plurality of time intervals, where the training of the machine learning model is based on the input vectors associated with the time interval, and the output values associated with the input vectors. The processor further trains a classifier for selecting one of the plurality of time intervals for input data received for a product.