AI Defect Detection for NDI Using Denoised Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Non-destructive inspection (NDI) techniques generate vast amounts of complex data that are challenging to evaluate for detecting faults or anomalies, as visual detection of defects is difficult or impossible, necessitating improved methods for anomaly detection in images obtained from NDI scanning.
Innovation Solution
A two-tiered denoising process is implemented for anomaly detection models, involving image-level and patch-level noise elimination, followed by feature matching techniques to pre-train the model using a normal dataset, generating a coreset of representative samples for accurate anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional visual inspection methods are used to detect defects in NDI images, then the inspection process remains simple and straightforward, but the detection accuracy is insufficient because faults are difficult or impossible to visually detect
Solution Approach 1:
The patent introduces an intermediary anomaly detection model that acts as a mediator between the complex NDI data and the inspection process. This model automatically identifies anomalies without requiring human visual interpretation, thereby improving detection accuracy while managing complexity through automation rather than direct human analysis of complex data patterns
Solution Approach 2:
The patent replaces the mechanical visual inspection process with an automated computational system. Instead of relying on human eyes and manual analysis of complex NDI images, the system uses algorithmic processing and anomaly detection models to automatically identify defects, substituting mechanical human inspection with computational automation
2Reliability
If a large volume of NDI data is processed to improve detection coverage, then the detection comprehensiveness increases, but the data processing time and computational resources increase significantly
Solution Approach 1:
The patent extracts and processes only the most relevant features from the large volume of NDI data. By identifying and focusing on key anomaly-indicating characteristics rather than processing all raw data equally, the system maintains comprehensive detection coverage while reducing the overall processing time and computational burden
Solution Approach 2:
The patent performs preliminary processing and filtering of NDI data before main anomaly detection. By pre-processing the data to identify potential anomaly regions or extract relevant features in advance, the system prepares the data for faster subsequent analysis, thereby maintaining comprehensive detection while reducing the time required for the main detection process
3Measurement precision
If anomaly detection models are trained with noisy or ambiguous data, then the training process becomes simpler and faster, but the model accuracy and reliability decrease
Solution Approach 1:
The patent performs preliminary cleaning and validation of training data to remove noisy or ambiguous samples before the actual training process. By pre-processing the dataset to ensure high-quality training examples, the system achieves better model accuracy without significantly increasing training complexity, as the filtering step is straightforward and automated
Solution Approach 2:
The patent implements self-service mechanisms where the training process automatically identifies and handles noisy data through built-in validation and filtering algorithms. The system autonomously manages data quality without requiring extensive manual intervention, thereby maintaining high model accuracy while keeping the training process relatively simple and self-managing
Data Source
AI summary
The present disclosure provides methods and techniques for anomaly detecting using feature matching models. A plurality of normal images are received. A plurality of patch features are generated by processing each of the plurality of normal images. A coreset comprising one or more coreset samples is generated, where the one or more coreset samples are selected from the plurality of patch features. A test image is received. One or more test patch features are generated by processing the test image. An anomaly score is generated by comparing the one or more test patch features with the one or more coreset samples.