Machine vision is a building block for the digitalization of a factory. Industrial cameras perform similar tasks to sensors, but their universal use opens up new possibilities in the automation of manufacturing processes. As the digital eye of robots and machines, industrial cameras help to master increasingly complex tasks. Their decisive advantage over highly specialized sensors is their ability to perform multiple tasks, as camera images enable a far more flexible evaluation of various (optical) features. Hardly any other component interprets and generates as much different data as image processing. It makes it possible to check and process what is seen, such as product features (length, distances, number), statuses (presence/absence) or quality in the production process, and to transfer the results to the systems in the value creation network. This not only determines whether the inspected part fulfills the desired characteristics or is good or bad, but also controls an intelligent action, such as automated sorting, depending on the result. Small and medium-sized companies in particular benefit from this added flexibility, as it was not possible for them to automate their production competitively due to lower production volumes. If cameras are used instead of sensors, even small batches of one part or more can be automated cost-effectively or series can be scaled up at a later date.

Digital eyes for robots and machines

Industrial cameras are already used in a wide range of industries: from equipment, plant and mechanical engineering to medical technology, agriculture and logistics. They support a wide range of tasks, for example in in-line quality inspections and general quality control. In conjunction with downstream image processing, they check products for deviations or defects so that products that do not meet the desired requirements can be rejected before they leave the production hall. Compared to the human eye, such machine vision systems work faster, more accurately and more reliably, as they do not miss any details even at high cycle rates. In addition, employees can be relieved of monotonous sorting and inspection tasks that are mentally challenging for humans, as the vision systems do not tire. Other areas of application include automated feeding (intralogistics) and areas in which inspection and measurement processes for shape, dimensional accuracy or color, for example, need to be 'contactless'.

Embedded Vision

Robotics and machine vision enable the automated handling of dimensionally unstable materials.

© sewts

In addition to classic PC-based vision systems, in which several industrial cameras supply images to an industrial PC via a cable, embedded vision systems enable the processing of 'seen' data on small, compact and resource-saving devices. These closely encapsulate the vision task of image acquisition, processing and result forwarding, enabling them to generate space-saving results 'on the edge' directly at the scene of the action without long cable runs for data transmission. Due to the relatively high integration costs for highly application-optimized systems, larger projects in particular benefit from this.

With current smart cameras with (often) integrated AI accelerators, however, configurable complete embedded vision solutions are also available on the market that can be integrated into in-house production with relatively little effort. This makes this device class user-friendly and cost-efficient, especially for the automation of small batches.

AI as a cornerstone

Contrary to general prejudices against artificial intelligence, AI opens up completely new fields of application in the area of machine vision that cannot be covered by classic, rule-based image processing. While many robots do not understand their environment and can only work based on commands, machine learning makes it possible to transfer human quality requirements to AI-based systems, which can then apply this knowledge to new situations and react adaptively. This ability is necessary, for example, when recognizing and processing objects with natural variance, such as food, plants or other organic objects. The color, surface, size, weight or shape of natural products, for example, exhibit a great deal of variance that can hardly be fully described using rules. However, the AI can be trained with appropriate training data so that a broad spectrum can be reliably recognized, categorized and thus also processed. If, for example, fruit is to be sorted on a conveyor belt according to various quality characteristics, a machine vision system based on artificial intelligence can precisely distinguish 'good' from 'bad' or other classes.

An example from practice

The Munich-based deep-tech start-up 'sewts' has set itself a special task in the field of automation: handling dimensionally unstable materials. This includes textiles, but potentially also films, foams, cables and malleable plastic parts. Sewts is now developing production-ready solutions that enable robots - similar to humans - to predict in real time how a textile will behave and adapt their movements accordingly. This is made possible by the use of modern machine vision technology in combination with intelligent software.

The author: Heiko Seitz is a technical editor at IDS in Obersulm.

© sewts

Here, industrial cameras from IDS act as the robot's eyes in a multi-camera system - with the task of identifying not only the shape and position of the materials, but also suitable gripping points. Depending on the needs and requirements of the end customer, two to three 3D cameras with artificial intelligence from the 'Ensenso' series are used to ensure high accuracy of the depth data and thus create a digital image of the spatial situation. Particularly light-sensitive 2D 'uEye' cameras complete the set-up to also capture surface and structural features. This combination of robotics and modern image processing closes one of the last automation gaps in industrial laundries and paves the way for the automated handling of easily deformable materials in various areas of application.