CNN Fault Detection for Display Process Sensor Traces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing trace data from display panel manufacturing processes is not in a suitable format for effective pattern recognition and classification, lacking sufficient data samples for reliable fault detection using machine learning models.
Innovation Solution
Convert one-dimensional time-series trace data into two-dimensional image-like data through processes such as resizing, mapping, merging, and rearranging, and apply a pre-trained convolutional neural network (CNN) for improved pattern recognition and classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If one-dimensional time-series trace data is used directly for machine learning analysis, then the data collection process is simple, but the pattern recognition performance and fault detection reliability are insufficient
Solution Approach 1:
The patent applies dimensionality change by converting one-dimensional time-series trace data into two-dimensional image-like data through techniques such as resizing, mapping, merging, and rearranging. This transformation enables the use of convolutional neural networks (CNNs) that are optimized for image processing, thereby significantly improving pattern recognition performance and fault detection reliability while maintaining manageable processing complexity through automated transformation pipelines.
2Reliability
If more trace data is collected from manufacturing processes to improve fault detection accuracy, then the reliability of fault detection improves, but the data processing time and computational resources increase
Solution Approach 1:
The patent implements preliminary action by pre-processing and transforming trace data into standardized two-dimensional image-like formats before feeding it into the machine learning model. This includes resizing, mapping, merging, and rearranging operations that are performed in advance, allowing the convolutional neural network to process the data more efficiently and reduce computational time during actual fault detection operations.
3Measurement precision
If traditional machine learning models are used with raw trace data, then the system complexity is low, but the ability to detect and classify fault patterns is insufficient
Solution Approach 1:
The patent applies mechanics substitution by replacing traditional machine learning approaches with a convolutional neural network (CNN) model that is specifically designed for image processing. The raw trace data is transformed into image-like representations, allowing the CNN to automatically learn and detect complex fault patterns through its hierarchical feature extraction capabilities, thereby achieving superior classification accuracy despite the increased model complexity.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A method for detecting a fault includes: receiving a plurality of time-series sensor data obtained in one or more manufacturing processes of an electronic device; arranging the plurality of time-series sensor data in a two-dimensional (2D) data array; providing the 2D data array to a convolutional neural network model; identifying a pattern in the 2D data array that correlates to a fault condition using the convolutional neural network model; providing a fault indicator of the fault condition in the one or more manufacturing processes of the electronic device; and determining that the electronic device includes a fault based on the fault indicator. The 2D data array has a dimension of an input data to the convolutional neural network model.