HUNT FOR DEFECTS
Deep learning technology uses neural networks that mimic human intelligence to distinguish anomalies, parts, and characters while tolerating natural variations in complex patterns.
Deep learning offers an advantage over traditional machine vision approaches, which struggle to appreciate variability and deviation between very visually similar parts.
Deep learning-based software tools like USS Vision Surface Scan can now perform judgment-based part location, inspection, classification, and character recognition challenges more effectively than humans or traditional machine vision solutions.
Inspect for scratches
Check fabric and leather quality
Categorize different part types
Inspect for porosity
Application Examples Include:
Another “first” from USS Vision.
In manufacturing stamped metal parts splits naturally occur. Typical processes dictate humans identify splits to prevent them from being assembled into a finished product. This process inevitably leads to splits being undetected and moving on to the next step in the assembly process which can result in failure in the field.
Automating the split detection process has been a goal in manufacturing for decades. Traditional machine vision systems are not capable of reliably detecting splits. At USS Vision we know this since we have tried to apply traditional machine vision to this application.
Recent developments in machine learning artificial intelligence present a means to tackle this application. Using machine learning we now can reliably detect splits down to 3mm in length. Our solution has been refined to a “kit” which can be installed in or near a variety of stamping presses. Our kit comes with a machine learning model template allowing for quick commissioning. Post commissioning our machine learning software consistently refines itself over time. Our system passes machine vision Gage Repeatability and Reproducibility standards and vastly improves customer’s quality while reducing labor costs.
Contact us for a demo.
Traditional machine vision struggles with detecting difficult to read vehicle identification numbers. Material a VIN is marked on changes over time, and markers can degrade requiring constant software programming modification.
End users are forced to babysit systems in order to keep them running, fighting with no-reads or misreads, essentially breaking traceability standards.
Machine learning vastly opens up opportunities to reliably detect VIN’s marked on a wide range of metals and plastics with a wide range of laser and pin-stamp markers. The software is able to robustly distinguish characters, even with marking surfaces changing over time which can occur from batch to batch.
Our software models constantly improve over time, they even continuously improve while you sleep! Not satisfied with your current reader, give us a ring. Need for a new reliable VIN reading system, give us a ring.
Feel free to contact us to set-up a demo!