Frame Image Expansion for Liquid Discharge Monitoring Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing substrate processing systems lack effective methods for generating sufficient training data for machine learning models to accurately monitor and control the discharge of liquids during substrate processing, such as wet etching, due to insufficient image data and analysis capabilities.
Innovation Solution
A computer-readable recording medium and information processing device that generates expansion data based on frame images from a moving image, using data expansion techniques to increase the number of frame images and incorporate label information, enabling supervised or non-supervised machine learning to generate learning models for predicting or classifying discharge states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If image data is collected during substrate processing, then monitoring capability is improved, but the quantity of training data remains insufficient for accurate machine learning models
Solution Approach 1:
The patent generates synthetic training data by copying and transforming existing frame images through data expansion techniques. Multiple copies of original images are created with variations in positioning, scaling, and transformation to artificially increase the training dataset size, enabling reliable machine learning models without requiring additional physical image collection
Solution Approach 2:
The patent applies parameter changes to image data by modifying positional coordinates, scaling factors, and transformation parameters during data expansion. These parameter transformations generate diverse training samples from limited original images, improving model reliability while maintaining the underlying physical constraints of the substrate processing system
2Reliability
If more frame images are used for training, then model accuracy is improved, but data processing time increases
Solution Approach 1:
The patent performs data expansion and preprocessing operations in advance before actual machine learning training. By pre-generating expanded training datasets and organizing them beforehand, the system reduces real-time processing requirements while maintaining high model accuracy, effectively trading offline computation time for online prediction speed
3Reliability
If data expansion is performed to increase training data, then model reliability is improved, but computational complexity increases
Solution Approach 1:
The patent uses copying-based data expansion where existing frame images are replicated and transformed through geometric operations. This approach increases training data quantity while maintaining computational simplicity, as the copying process uses standard image processing algorithms rather than complex generation methods
Solution Approach 2:
The patent implements a universal data expansion framework that can process various types of frame images from different substrate processing conditions using the same transformation algorithms. This multi-functional approach handles diverse imaging scenarios without requiring separate processing pipelines, reducing overall system complexity
Data Source
AI summary
A non-transitory computer-readable recording medium having stored thereon a computer program that, in response to execution, causes circuitry to perform a method including: acquiring a moving image of a substrate processing apparatus; and generating expansion data based on a frame image of a first time point and a frame image of a second time point later than the first time point, which are included in the acquired moving image.