Because bearings reduce friction, they are fundamental to enabling automobiles’ smooth and efficient movements. This application focuses on bearing inspection after the machining process.
Several defects can occur during the manufacturing of bearings, such as:
These defects affect bearing performance, durability, and safety in their applications and manufacturers should implement quality control measures to prevent them.
However, these defects can be difficult to detect– bearing metal surfaces may reflect light in a way that masks defects. And many bearing defects are subtle and hard to image, especially if the defect’s contrast to the background is low. Traditional machine vision systems struggle to capture clear images and distinguish between actual defects, reflective surfaces, and the background, ultimately missing defects or causing false rejections.
Bearings have intricate geometries with curved surfaces, races, and rolling elements which traditional machine vision can struggle to accurately capture.
And in high-volume manufacturing environments, bearings must be inspected quickly to keep up with production rates. Traditional machine vision products may fail to keep up with required cycle times.
UnitX’s AI-powered inspection effectively inspects machined bearing defects where other solutions fail.
First, the OptiX imaging system illuminates and images the bearing. Then, the CorteX Central AI platform is trained on machining defects. Lastly, those AI models are deployed to the CorteX Edge inference system to detect and classify defects in-line.
Alternatively, manufacturers can use just CorteX AI if they have existing imaging systems. For example, if a manufacturer wants to detect internal bearing defects such as inclusions embedded in the bearing material, they can deploy just CorteX AI and integrate it with existing X-Ray and CT Scanners for fast deployment of improved defect detection.
OptiX provides superior images that minimize reflectivity while maximizing defect visibility. It has 32 independently controllable lighting sources that can be optimized for metal bearing surfaces and various defects via software. Its computational imaging capability can be used to take multiple shots and eliminate hotspots caused by highly reflective bushing surfaces. And its lighting dome design supports a very acute incidence angle of projected light, causing even very tiny defects to cast shadows which increase their visibility.
CorteX accurately detects random, complex defects. It automatically normalizes for variability in positions and orientations and recognizes defects down to the pixel-level. It reduces false positives that lead to scrap and wasted product.
CorteX supports fast AI model development, deployment, and iteration. CorteX AI models are sample efficient– they only require a few images to train on new defect types.
UnitX optimizes yield. In CorteX, can tune quality criteria and visualize the impact on yield before rolling those changes to production. All inspection data is referenceable in one central platform for manufacturers to analyze and identify areas for process improvements.
UnitX provides rapid, 100% inline inspection. OptiX has bright LEDs and fast fly capture speeds of 1m/s for high speed imaging. And CorteX Edge supports high inference speeds (up to 100 MP) to quickly output an OK/NG decision, seamlessly communicating that decision via integration to all major PLC, MES, and FTP systems.
Manufacturers who use UnitX to automate bearing machining inspection are able to:
In this example, we inspected metal bearing for scratches resulting from the machining process.
Imaging
First, we used OptiX to capture images of the bearings, making sure we captured both defective and OK parts. We configured OptiX to image the bearings from multiple angles and positions so that we can detect defects on all components of the bearing such as the outer race track, inner race track, and cage. We used OptiX’s software-defined lighting to configure various lighting patterns and incident angles that best captured the subtle defects.
Training
Next, we used CorteX Central to train our models. We created a label for scratch defects.
We then labeled the scratch defect in the images we captured from OptiX, using just a few images to train. With UnitX, the same generalized network can inspect multiple part types for the same defect, depending on the application. So in this example, we did not need to train on every variation of OptiX imaging.
Because of CorteX’s user-friendly interface and the low number of images it requires to train its AI models, it only took us a few minutes to complete the labeling and training.
Detection
We then deployed those AI models to CorteX Edge to detect scratches on new bearings, resulting in accurate detection and classification.
