US-12625032-B2 - Image-based bearing failure detection

Abstract

A device may receive, from a front-end device communicatively coupled to the back-end device, one or more images of a bearing. A device may identify a number of defects of one or more defect classes on the bearing based on the one or more images using a deep learning classifier, wherein the deep learning classifier is trained on training data including a plurality of images of training bearings with identified defects in the one or more defect classes. A device may generate defect data associated with the defects identified on the bearing when the number of defects identified on the bearing is at least one. A device may direct the front-end device to display the defect data on a display device.

Inventors

• BALAJI CHANDRASEKARAN
• Rudoniel Correa Cury
• Gustavo dos Santos Gioria

Assignees

• Schaeffler Technologies AG & Co. KG

Dates

Publication Date

20260512

Application Date

20240523

Claims (17)

1. 1 . A system for defect inspection comprising: a back-end device including a server with one or more processors configured to execute program instructions stored in a non-transitory memory device, wherein the program instructions are configured to cause the one or more processors to: receive, from a front-end device communicatively coupled to the back-end device, one or more images of a bearing, wherein the front-end device includes at least one of a mobile device, a tablet, or a computer; identify one or more defects in one or more defect classes on the bearing based on the one or more images using a deep learning classifier, wherein the deep learning classifier is trained on training data including a plurality of images of training bearings, wherein at least some of the plurality of images of the training bearings include defects known to be associated with the one or more defect classes, wherein the training data includes an initial training dataset, wherein the deep learning classifier is trained using a semi-supervised learning technique comprising: training the deep learning classifier with the initial training dataset; applying the deep learning classifier to a batch of unlabeled data to generate pseudo-labeled data; reviewing at least a portion of the pseudo-labeled data to generate reviewed data; and retraining the deep learning classifier based on the reviewed data; generate defect data associated with at least one of the one or more defects identified on the bearing, wherein the defect data includes a mitigation technique for mitigating at least one of the one or more defects during a fabrication process of the bearing; and direct the front-end device to display the defect data on a display device.
2. 2 . The system for defect inspection of claim 1 , wherein the defect data comprises: a number of defects identified in at least one of the one or more defect classes.
3. 3 . The system for defect inspection of claim 1 , wherein the defect data comprises: an annotated image indicating at least locations of the defects identified on the bearing.
4. 4 . The system for defect inspection of claim 1 , wherein the defect data comprises: a probability of at least one of the one or more defects identified on the bearing belonging to at least one of the one or more defect classes.
5. 5 . The system for defect inspection of claim 1 , wherein the defect data comprises: a root cause associated with at least one of the one or more defects identified on the bearing.
6. 6 . The system for defect inspection of claim 1 , wherein the deep learning classifier is trained with at least one of a supervised learning technique or a semi-supervised learning technique.
7. 7 . The system for defect inspection of claim 1 , wherein the program instructions are further configured to cause the one or more processors to: verify whether the one or more images meet one or more quality standards, wherein one or more test images include any of the one or more images that pass the one or more quality standards; wherein identify the one or more defects in the one or more defect classes on the bearing based on the one or more images using the deep learning classifier comprises: identify the one or more defects in the one or more defect classes on the bearing based on the one or more test images using the deep learning classifier.
8. 8 . The system for defect inspection of claim 7 , wherein at least one of the one or more quality standards comprise: an image quality standard associated with at least one of contrast or blur.
9. 9 . The system for defect inspection of claim 7 , wherein at least one of the one or more quality standards comprise: an object detection check.
10. 10 . The system for defect inspection of claim 9 , wherein detection of a face results in failure of the object detection check.
11. 11 . The system for defect inspection of claim 1 , wherein at least one of the one or more defect classes comprises: at least one of cracking, discoloration, false brinelling, fretting, indentation, pitting, rust, spalling, wear, or an overheating defect.
12. 12 . A system for defect inspection comprising: one or more processors configured to execute program instructions stored in a non-transitory memory device, wherein the program instructions are configured to cause the one or more processors to: receive one or more images of a bearing from a user via a user interface communicatively coupled with the one or more processors; verify whether the one or more images meet one or more quality standards, wherein one or more test images includes any of the one or more images that pass the one or more quality standards; identify one or more defects in one or more defect classes on the bearing based on the one or more test images using a deep learning classifier, wherein the deep learning classifier is trained on training data including a plurality of images of training bearings, wherein at least some of the plurality of images of the training bearings include defects known to be associated with the one or more defect classes, wherein the training data includes an initial training dataset, wherein the deep learning classifier is trained using a semi-supervised learning technique comprising: training the deep learning classifier with the initial training dataset; applying the deep learning classifier to a batch of unlabeled data to generate pseudo-labeled data: reviewing at least a portion of the pseudo-labeled data to generate reviewed data; and retraining the deep learning classifier based on the reviewed data; generate defect data associated with at least one of the one or more defects identified on the bearing, wherein the defect data includes a mitigation technique for mitigating at least one of the one or more defects during a fabrication process of the bearing; and cause the defect data to be displayed on a display device.
13. 13 . A method for defect inspection comprising: capturing, via a front-end device accessible to a user, one or more images of a bearing, wherein the front-end device includes at least one of a mobile device, a tablet, or a computer; identifying, with a back-end device including a server, one or more defects in one or more defect classes on the bearing based on the one or more images using a deep learning classifier, wherein the deep learning classifier is trained on training data including a plurality of images of training bearings, wherein at least some of the plurality of images of the training bearings include defects known to be associated with the one or more defect classes, wherein the training data includes an initial training dataset, wherein the deep learning classifier is trained using a semi-supervised learning technique comprising: training the deep learning classifier with the initial training dataset; applying the deep learning classifier to a batch of unlabeled data to generate pseudo-labeled data; reviewing at least a portion of the pseudo-labeled data to generate reviewed data; and retraining the deep learning classifier based on the reviewed data; generating, with the back-end device, defect data associated with at least one of the one or more defects identified on the bearing, wherein the defect data includes a mitigation technique for mitigating at least one of the one or more defects during a fabrication process of the bearing; and causing the defect data to be displayed on a display device on the front-end device.
14. 14 . The method of claim 13 , wherein the defect data comprises: at least one of: a number of defects identified in at least one of the one or more defect classes; an annotated image indicating at least locations of the defects identified on the bearing; a probability of at least one of the one or more defects identified on the bearing belonging to at least one of the one or more defect classes; a root cause associated with at least one of the one or more defects identified on the bearing; or a mitigation technique for mitigating fabrication of at least one of the one or more defects identified on the bearing.
15. 15 . The method of claim 13 , wherein the deep learning classifier is trained with at least one of a supervised learning technique or a semi-supervised learning technique.
16. 16 . The method of claim 13 , further comprising: verifying whether the one or more images meet one or more quality standards, wherein one or more test images includes any of the one or more images that pass the one or more quality standards; wherein identifying the one or more defects in the one or more defect classes on the bearing based on the one or more images using the deep learning classifier comprises: identifying the one or more defects in the one or more defect classes on the bearing based on the one or more test images using the deep learning classifier.
17. 17 . The method of claim 13 , wherein at least one of the one or more defect classes comprises: at least one of cracking, discoloration, false brinelling, fretting, indentation, pitting, rust, spalling, or an overheating defect.

Description

TECHNICAL FIELD The present disclosure relates generally to defect detection in bearings and, more particularly, to image-based defect detection using deep learning techniques. BACKGROUND Mechanical bearings are a critical component of many systems. Defects in such bearings may lead to numerous undesirable effects such as, but not limited to, undesirable acoustic signals (e.g., sounds), undesirable vibrations, increased mechanical wear, or a point of failure. It may therefore be desirable to identify and/or classify bearing defects during a manufacturing process (e.g., on a shop floor). However, typical techniques for defect identification and classification rely on manual inspection of bearings and/or images of bearings, which is time consuming and prone to error. There is therefore a need to develop systems and methods for curing the above deficiencies. SUMMARY In embodiments, the techniques described herein relate to a system for defect inspection including a back-end device including one or more processors configured to execute program instructions stored in a memory device, where the program instructions are configured to cause the one or more processors to receive, from a front-end device communicatively coupled to the back-end device, one or more images of a bearing; identify one or more defects in one or more defect classes on the bearing based on the one or more images using a deep learning classifier, where the deep learning classifier is trained on training data including a plurality of images of training bearings, where at least some of the plurality of images of the training bearings include defects known to be associated with the one or more defect classes; generate defect data associated with at least one of the one or more defects identified on the bearing; and direct the front-end device to display the defect data on a display device. In embodiments, the techniques described herein relate to a system for defect inspection, where the front-end device includes at least one of a mobile device, a tablet, or a personal computer; where the back-end device includes a server. In embodiments, the techniques described herein relate to a system for defect inspection, where the defect data includes a number of defects identified in at least one of the one or more defect classes. In embodiments, the techniques described herein relate to a system for defect inspection, where the defect data includes an annotated image indicating at least locations of the defects identified on the bearing. In embodiments, the techniques described herein relate to a system for defect inspection, where the defect data includes a probability of at least one of the one or more defects identified on the bearing belonging to at least one of the one or more defect classes. In embodiments, the techniques described herein relate to a system for defect inspection, where the defect data includes a root cause associated with at least one of the one or more defects identified on the bearing. In embodiments, the techniques described herein relate to a system for defect inspection, where the defect data includes a mitigation technique for mitigating fabrication of at least one of the one or more defects identified on the bearing. In embodiments, the techniques described herein relate to a system for defect inspection, where the deep learning classifier is trained with at least one of a supervised learning technique or a semi-supervised learning technique. In embodiments, the techniques described herein relate to a system for defect inspection, where the program instructions are further configured to cause the one or more processors to verify whether the one or more images meet one or more quality standards, where one or more test images include any of the one or more images that pass the one or more quality standards; where identify the one or more defects in the one or more defect classes on the bearing based on the one or more images using the deep learning classifier includes identify the one or more defects in the one or more defect classes on the bearing based on the one or more test images using the deep learning classifier. In embodiments, the techniques described herein relate to a system for defect inspection, where at least one of the one or more quality standards include an image quality standard associated with at least one of contrast or blur. In embodiments, the techniques described herein relate to a system for defect inspection, where at least one of the one or more quality standards include an object detection check. In embodiments, the techniques described herein relate to a system for defect inspection, where detection of a face results in failure of the object detection check. In embodiments, the techniques described herein relate to a system for defect inspection, where at least one of the one or more defect classes includes at least one of cracking, discoloration, false brinelling, fretting, flutting, indentation,