Vision measuring systems

Features - Vision Systems

These systems acquire highly accurate measurements on small, tight-tolerance features, making them an integral part of quality control for important parts and instruments in medtech.

October 14, 2022

Software is much better at catching tiny surface defects than the human eye as shown by a cross-sectional representation zoomed in to see defects the software detected.

Vision machines can measure a variety of different materials such as stainless steel, titanium, carbon fiber, and plastics for intricate products like bone screws, drill bits, and needles to larger parts such as artificial joints and medical tubes – without contaminating the specimens.

While vision systems excel at performing non-touch dimensional measurements on all types of parts and instruments, a vision measuring machine needs specialized defect detection software to find cracks, dirt, or any other type of defect, including miscoloration. Previously, defect detection on parts was always performed in a time-consuming practice with a person inspecting each part manually.

New technology and software are altering that paradigm, and it could save manufacturers of medical devices and tools money and time. While defect detection software has been on the market for a while, there’s new advanced defect detection software that uses artificial intelligence (AI) to automate the process that’s also versatile and easy to use.

Self-learning artificial intelligence (AI) software uses images with defects that are input by an operator to recognize and predict a variety of defect types automatically.

Introducing AI to vision defect detection improves part inspection throughput in a couple of different ways. First, AI removes the need for a person to visually inspect each piece manually and makes the defect detection process automatic. This frees up manpower hours while drastically reducing missed error detections which is critical in the medical industry.

Second, this new software can run on vision measuring systems and detect defects while the part is measured dimensionally by the machine simultaneously, eliminating one of the quality control steps in your program while improving quality control. The parts go into the vision machine, and the software will automatically detect defects in a matter of seconds after the vision system measures for dimensions without touching the part or machine.

This software can be loaded and used on any type of camera system with a high enough pixel count or on specific vision measuring machines. This setup is ideal for the factory floor to inspect dozens or hundreds of parts for defects at high speeds while the parts are moving down a conveyor belt after they were created. This is great for companies needing defect detection without inline measurements and who don’t want to invest capital into vision machines since the camera systems are considerably less expensive.

To put this in a different perspective, let’s use strawberries for an example. A strawberry farm wants to make sure all the strawberries they package and send to market are good. There are several strawberries moving down the line at once, this defect detection software can spot brown areas, strawberries that have gone bad, or any other type of issue that makes a strawberry inedible. The software, combined with inspection cameras on the line, detects a strawberry with some sort of defect, it tells the system to push the strawberry off the belt. This example shows how much time and manpower can be saved using self-learning defect detection software.

Most AI detection software is training model based, or machine learning, which is simple and intuitive to use. The software is trained by uploading different real-world images of parts, then you point out any type of defect to the system – from scratches, burrs, cracks, abrasions, and even particles of dust and dirt – much like teaching a young child. The software then teaches itself to recognize those issues on future parts. Once you give it enough images and descriptions, the software will quickly scan any part or groups of parts to be inspected, recognize the type of defect it learned, and then flag that part so it can be separated from defect-free pieces.

When trained properly, AI-powered defect detection is up to 99% effective and quickly becomes worth the investment. This type of detection has mostly upside, especially since it can be installed on a QV measuring system or camera system, and it’s faster and less expensive than manual defect detection. And while AI detection software is ideal for the medical manufacturing industry, it’s useful for almost any type of part production requiring some sort of defect detection.

Mitutoyo America Corp.

About the author: Mark Sawko is vision product manager at Mitutoyo America Corp.