Machine Vision

Machine vision is one of the enabling technologies that turns a simple pick-and-place robot into a flexible one. Without vision, a robot can only pick parts that arrive in an exact, pre-programmed position, usually via a jig, tray or fixture. Add a vision system, and the same robot can pick parts that are randomly oriented on a conveyor or tipped loosely into a bin, because the camera locates each part in real time and tells the robot where to reach. For UK manufacturers weighing up automation, this often determines whether a low-cost fixed-position cell is enough, or whether a more expensive vision-guided system is required. Getting this decision right early avoids expensive rework later, and it is usually one of the first questions a good integrator will ask during project scoping.

A typical machine vision system pairs one or more industrial cameras with dedicated lighting, positioned to capture a consistent, glare-free image of the part or scene. The image is passed to vision software, which processes it using techniques such as edge detection, pattern matching, blob analysis or, increasingly, deep learning models trained on sample images.

The software extracts the information the application needs, for example the X/Y/Z position and rotation of a part, whether a label is present and correctly printed, or whether a weld bead meets a quality threshold. This result is then communicated to the robot controller or PLC over a standard industrial protocol, such as Ethernet/IP or Profinet, in time for the robot to act on it within its cycle. Lighting and camera placement are usually the deciding factors in how reliable a vision system is, inconsistent ambient light, reflective surfaces or vibration are the most common causes of systems that perform well in testing but struggle on the factory floor.