Machine learning algorithms (neural nets) bridge the gap between what a person can inspect and what can be taught with traditional image processing techniques. Utilizing off the shelf hardware, cameras and stages, Lupine Labs has integrated Google’s® Tensorflow into a user friendly defect detection and classification system. Derived from the same software that drives cars and finds cats on the internet, Tensorflow allows this machine to detect and classify defects with incredible accuracy.
The system can be utilized in a vast number of applications. Examples include weld seams, textures, mold flash, bubbles, cracking, clarity, etc…just about any visual inspection. Rule of thumb: if a human can see the defect in 2 seconds, the machine can be trained to find it. And as a modular design, it can be configured to inspect a range of part sizes at varying magnifications. It can even focal stack images to simulate a large focal depth at high magnification, generating fully focused images of deep features.
Additionally, the system can be utilized for inspections beyond those that can be performed visually. By integrating other metrology devices, the system can perform several inspections at one location, reducing total inspection time and increasing efficiency.
Lupine Labs is excited about this technology and its capabilities. In its first application, it is achieving >94% accuracy classifying over six possible defect categories. This was achieved with a relatively small training set and minimal training time. Contact us at email@example.com to discuss your application and see how we can help.