Machine Vision Change Detection for Label-Free Process Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing manufacturing processes face challenges in efficiently detecting changes and anomalies in production processes using vision-based systems due to the difficulty in generating sufficient labeled data and the time-consuming nature of introducing anomalies for training, as well as the uncertainty in how anomalies will appear during actual production issues.
Innovation Solution
A method and system that utilize unsupervised machine learning techniques to analyze images of production processes, performing hypothesis testing on clusters of images based on production factors without labeled data, allowing for the identification of variations and determining production metrics by classifying images into clusters and rejecting a null hypothesis based on significance levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised machine learning techniques are used to monitor production processes, then the system can identify features of interest, but generating sufficient labeled data for training is difficult and time-consuming
Solution Approach 1:
The system performs self-service by automatically generating labels through hypothesis testing and cluster analysis without requiring manual annotation. The unsupervised learning approach allows the system to self-train on production data, eliminating the time-consuming manual labeling process while maintaining feature identification accuracy.
Solution Approach 2:
Instead of starting with labeled data to train the model (traditional supervised learning), the patent inverts the approach by first clustering unlabeled data and then using hypothesis testing to generate labels. This reverse process eliminates the need for pre-existing labeled datasets while achieving the same monitoring objectives.
2Reliability
If anomalies are introduced into the process for training purposes, then training data can be generated, but it is time-consuming and difficult to predict how anomalies will appear in actual production
Solution Approach 1:
The system automatically discovers anomaly patterns through unsupervised clustering and hypothesis testing without requiring manual anomaly injection. By analyzing natural variations in production data, the system identifies meaningful anomalies while avoiding the complexity and unpredictability of artificially introduced defects.
Solution Approach 2:
The patent introduces cluster analysis and hypothesis testing as intermediary steps between raw images and anomaly detection. These intermediaries automatically identify patterns and variations in the data, serving as a bridge that eliminates the need for manual anomaly creation while maintaining reliable detection capabilities.
3Measurement precision
If traditional inspection procedures are used, then accurate monitoring can be achieved, but the inspection process is slow and cannot keep up with large production volumes
Solution Approach 1:
The patent replaces traditional mechanical/optical inspection systems with an automated computer vision system using deep learning. This substitution enables parallel processing of multiple images simultaneously, maintaining high inspection accuracy while dramatically increasing processing speed to match large production volumes.
Solution Approach 2:
The system performs preliminary clustering and feature extraction on images before detailed analysis. By pre-processing and organizing data into clusters based on similarity, the system reduces the computational burden of subsequent anomaly detection, enabling faster inspection without sacrificing accuracy.
Data Source
AI summary
A method for a metric determination includes extracting one or more features from a plurality of images and classifying the plurality of images based on the extracted one or more features to form one or more clusters of images, wherein the classifying is performed at least in part based on machine learning. Hypothesis testing is performed on the one or more clusters of images based on one or more production factors and variations are identified in the one or more features resulting from the hypothesis testing. One or more production metrics are determined based on the identified variations.