Coated Transparent Substrate Quality Prediction Without Destructive Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for quality control of layered coatings on transparent substrates face challenges such as production losses from destructive testing, lag times in correcting process parameters, and the high cost and complexity of online feedback systems, which require extensive data collection and processing.
Innovation Solution
A computer-implemented method that predicts non-in-process measured quality functions of a layered coating from in-process measured quality functions, using a machine learning-based regression analysis trained on sets of training and target data, thereby reducing the need for continuous real-time data monitoring and processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If offline quality controls are carried out by measuring quality functions on collected samples, then quality functions can be measured, but production losses occur from product destruction and lag times between measurement and correction of process parameters
Solution Approach 1:
The patent applies preliminary action by training a machine learning model in advance using historical process data and corresponding quality measurements. Once trained, the model can predict quality functions in real-time during production without requiring destructive sampling or offline measurements, thus eliminating production losses while maintaining measurement accuracy
Solution Approach 2:
The patent creates a virtual copy of the quality measurement system through machine learning modeling. Instead of physically measuring quality functions through destructive testing, the system uses a trained model that replicates the measurement function based on process parameters, thereby avoiding product destruction and enabling continuous production
2Reliability
If online feedback methods are used with real-time monitoring and continuous measurements of quality functions, then quality control can be performed during production, but extensive data collection and processing are required which increases system complexity and cost
Solution Approach 1:
The patent extracts the essential quality prediction function from complex online feedback systems. Instead of implementing comprehensive real-time monitoring of multiple quality functions with extensive sensors and data processing infrastructure, the method isolates and predicts only the critical quality functions using a trained machine learning model that processes minimal input data
Solution Approach 2:
The patent transforms the quality control approach by changing from direct physical measurement parameters to predictive modeling parameters. The system uses process parameters as input to a trained model that outputs predicted quality functions, replacing the need for complex measurement systems with a computational approach that reduces device complexity
Data Source
AI summary
A computer implemented methods for estimating at least one quality function of a given layered coating on a transparent substrate allows to predict at least one non in-process measured quality function of a given layered coating on a transparent substrate from an in-process measured quality function which can be acquired on the coated substrate as deposited at any location, preferably at the end of a coating process. The method allows to get rid of in-process real-time continuous measurements of quality functions of the coated transparent substrate and real-time monitoring of coating process parameters.