Machine Vision, Deep Learning and Artificial Intelligence are among the future of automated testing in many different industries, especially manufacturing. But what if you need a vision system to recognize or verify organic objects? What if you need a system that doesn’t just recognize parts, but is able to make a subjective decision like a human operator?
Cyth’s Neural Vision takes traditional machine vision software to a whole new level. In the same way humans learn to recognize an apple from an orange, Neural Vision intuitively looks for unique features, similarities, and differences between different objects observed through the power of Deep Learning. An international semiconductor test manufacturer was looking for just that level of intuition to help their clients.
The customer specializes in the handling and testing of high throughput semiconductor chips; subcomponents of items electrical engineers would use. They were looking to partner with an American vision specialist to help bring AI and Deep Learning technologies into a next generation system for the semiconductor inspection marketplace. They were seeking out a disruptive technology, and Cyth’s team was confident they could successfully integrate Neural Vision into their inspection system.
Client Request & Cyth’s Solution
The developed system inspects the client’s product by utilizing high resolution line scan cameras to build up a product image, singulating unique components, and then running those images through the Neural Vision software. The software then determines good or bad parts as well as classifying defects over 50 categories. Because the client works with customers around the world with hundreds of unique product lines, they were eager to leverage Neural Vision’s easily automated infrastructure to implement specific product inspection solutions. Handling this large inspection set would not have been possible with traditional machine vision; it calls for something more intuitive and user-friendly. The customer wanted the ability to selectively apply different solutions or criteria to their unique inspection needs at will.
Cyth integrated a vision subassembly into the loading mechanism and Festo motor with linear stage provided by the client. To create the system, Cyth’s team used LabVIEW software with two 12k resolution Basler line scan cameras, lenses from Edmund Optics, and specialized custom optical spacers for precision light control. Two high intensity red line lights were used from Advanced Illumination and provided the ideal image quality when inspecting a reflective part. Every product image required 1GB of data, so the team implemented a powerful PC to handle the large amounts of processing power needed.
For the final solution, the client will load a cassette of multiple lead frames. These will be indexed through the system for the inspection of individual parts. Each frame contains over 100 unique parts with over 50 inspections each and a target inspection time of 25 seconds. The output will be a visual report detailing which components have been identified as rejects with a detailed breakdown of defects based on client criteria, resulting in fast and accurate processing of frames.
The semiconductor chip was so small that the cameras needed to resolve to the single micron level, so the selection of optical components was a major hurdle to work through. It was difficult to find a camera lens combination that had the ability to capture micron-level imagery while also mechanically fitting into the necessary space as well as being capable of capturing images in the needed time frame. There is a finite working distance due to the focal points of the lenses selected and the customer wanted the system to be within a contained footprint.
By providing the customer with a platform that allows control of the creation or refinement of their inspections, Cyth is helping to forever change their inspection system as speed is increased and human error is significantly decreased. Traditionally, they have hundreds of parts being manually inspected by a worker, but by harnessing Cyth’s Neural Vision technology they can automate their high-resolution inspection with little to no error. Although this is the first system that has been developed, the client is looking forward to targeting around 200-300 systems per year and taking this revolutionary solution internationally.
• LabVIEW 2017 64-bit
• 2 12288 Pixel Resolution, Linear CMOS, 3.5μ x 3.5μ Pixel Size, GigE, 8 kHz Line Rate, Monochrome Cameras
• 2 Custom F-Mount Optical Spacers
• 2 0.5X LS Series f/4 Line Scan Lens by Edmund Optics
• 1 LM13 INC X1000 .1MHZ 3M FLYING LEAD REF Encoder
• 2 Red Expandable High Intensity Line Light