Radiographic Defect Detection in Ceramic Rolling Elements Using Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting manufacturing defects in ceramic rolling elements, such as inclusions and porosities, are inefficient and prone to human error, especially in mass production, leading to potential failure and safety issues in mechanical assemblies.
Innovation Solution
A method involving the training of a deep-learning algorithm using digital radiographic images, where images are filtered and segmented to create a data set for classification by both statistical and deep-learning algorithms, with a convolutional neural network, to accurately identify suspect rolling elements, and comparing their classifications for accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual defect recognition by operator is used, then interpretation accuracy can be maintained, but production speed is reduced and human error occurs
Solution Approach 1:
The patent replaces the manual mechanical inspection process with an automated deep-learning algorithm that processes radiographic images. The system uses convolutional neural networks to automatically detect defects in ceramic rolling elements, eliminating the need for human operators to manually examine images while maintaining or improving detection accuracy and significantly increasing production speed.
Solution Approach 2:
The deep-learning algorithm is trained on a dataset of radiographic images with known defects, enabling it to autonomously learn defect patterns and perform self-improving defect detection. The system continuously refines its detection capabilities through training on annotated data, allowing it to maintain high accuracy without human intervention during actual inspection operations.
2Productivity
If deep-learning algorithm is used for defect detection, then automation and speed are improved, but training complexity and computational resources are required
Solution Approach 1:
The patent implements a preliminary training phase where the deep-learning algorithm is trained offline on a comprehensive dataset of radiographic images with annotated defects. This preliminary action prepares the model in advance, allowing it to perform rapid automated detection during actual production without requiring real-time complex computations. The training phase separates the complexity from the production phase, enabling fast deployment.
Solution Approach 2:
The system creates a digital copy of the defect detection task through the trained neural network model. Once trained, the model can be replicated and deployed across multiple inspection stations, allowing the complex intelligence to be copied rather than重新 developed at each location, reducing overall system complexity while maintaining high detection speed.
3Measurement precision
If radiographic imaging is used to detect internal defects, then defect visibility is improved, but image processing complexity increases
Solution Approach 1:
The patent applies image segmentation techniques to divide the radiographic images into relevant regions of interest, such as the rolling element boundaries and potential defect areas. This segmentation reduces the complexity of processing entire images by focusing computational resources on specific areas where defects are likely to occur, while maintaining the high visibility advantage of radiographic imaging.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach automates defect recognition, increasing production efficiency and reducing errors, allowing for real-time detection of defects and minimizing the risk of failure in ceramic rolling elements.
Implementation Method 1
It is known to use x-rays to obtain radiographic images of rolling elements
Data Source
AI summary
A system and method for training a deep-learning algorithm to detect a defect in a ceramic rolling element incudes capturing a first data set of digital radiographic images of rolling elements, filtering the images of the first data set to improve contrast, classifying each image using a statistical learning algorithm into a first class of suspect rolling elements or into a first class of non-suspect rolling elements, using the first data set to train a deep-learning algorithm to classify each image of the first data set into a second class of suspect rolling elements or into a second class of non-suspect rolling elements, and comparing the classifications performed by the deep-learning algorithm to the classifications performed by the statistical learning algorithm to determine an accuracy of the deep-learning algorithm.