Machine vision systems: digitisation of visual inspection processes for products quality control in modern automated food & beverage packaging lines
Machine vision systems extend the sensory field of automatic machines and robot cells, following the technological development trend of Smart Manufacturing in the Industry 4.0 paradigm.
Machine vision technology plays a key role in advanced automation solutions for food & beverage packaging. It replaces traditional manual methods for quality control by introducing automated processes that analyse computerised digital images acquired by cameras for visual inspection of products along the production chain.
Machine vision in the food & beverage packaging
Computer processing of images acquired by machine vision devices provides information in digital format of the distinctive features of the products analysed, such as contours, shape, colour and size, as well as the location of their position in 2D-3D space.
This digital data is mainly used to verify the conformity and safety of inspected food products and for quality control in production processes, highlighting any discrepancies, missing parts and defects.
Vision systems are also used for traceability and product counting in the supply chain, through the reading of 1D codes (Barcodes), 2D codes (Data Matrix and QR-Codes) and optical character recognition (OCR).
Digital processing of images acquired by machine vision systems also provides effective support for the guidance of robots and automatic machines. Typical applications in this context are robotic pick-and-place operations, bin picking, selection and sorting.
The advantages of a machine vision system in food & beverage packaging
The application of machine vision in food & beverage packaging introduces the benefits of automated inspection and quality control processes for checking the conformity of products, through computerised analysis of acquired images, without direct contact with foodstuffs. If discrepancies and defects are detected, the information in digital format is shared with automated production chain systems to reject the inspected product.
The vision systems are able to acquire and analyze images of the products to be inspected moving on a conveyor belt at high speeds, maintaining the standard of precision and reliability of the checks carried out objectively unchanged over time.
Furthermore, the vision devices can operate in hostile environmental conditions, in confined spaces and close to the range of action of automatic machines, not accessible to a human operator, sharing digital data extrapolated from the images along the production chain.
In this context, the benefits of machine vision in modern automated food & beverage packaging lines translate into the optimization of plants efficiency with an increase in productivity and flexibility, as well as raising the reliability standards of quality control and food safety. Errors therefore decrease in favor of improving the quality of products and packaging.
2D and 3D machine vision features for application in food & beverage packaging
In a modern automated line for food & beverage packaging, the use of a system with 2D or 3D artificial vision technology depends on the type of application or control to be carried out.
2D artificial vision is suitable for checking the conformity of the graphics, colours, markings and labels on a package, verifying its integrity and correct positioning and alignment; it can also read and verify 1D, 2D and alphanumeric OCR codes. It is able to detect the geometry of the contours, measure the dimensions (length and width), identify the position and orientation in the two-dimensional space of a product, carrying out inspections for quality control, highlighting defects of conformity, integrity and verify missing parts.
For a good result of the digital images acquired by 2D vision, there must be a strong contrast between the product to be inspected and the background. Adequate lighting is therefore important: direct for opaque surfaces and indirect for shiny ones.
3D artificial vision technology is suitable for measuring heights and volumes, analyzing the dimensions and shapes of a product, even in low contrast conditions, highlighting missing parts, discrepancies, defects in the shapes and assembly of the packages. Small differences in height of products with the same shape, arranged in a package, can be discriminated from the images acquired by 3D vision systems.
3D artificial vision is also an effective support for robot guidance, being able to precisely locate the position and orientation in three-dimensional space of the products to be selected and manipulated.
The components of a machine vision system
An artificial vision system is composed of cameras with different optics and sensors for digital image acquisition, lighting equipment and calculation units with specific software and algorithms for image processing and deep learning (automatic learning to carry out classifications directly on images).
Depending on the visual inspection application along the production chain, different camera sensor technologies enable digital images to be captured both in the visible light spectrum and outside it, through infrared, ultraviolet or X-ray frequencies.
Cameras for machine vision systems
Cameras for artificial vision applications in the industrial sector are composed of an optics (with lens), an image sensor that converts light into electrical signals, an electronic control system and a communication interface with the control unit calculation for processing acquired digital images.
The image sensor of the cameras is composed of photosensitive elements (pixels) that convert light into electrical signals to generate the image in digital format. In particular, sensors with CMOS (Complementary Metal Oxide Semiconductor) technology are suitable for high-speed video applications, being able to acquire up to thousands of frames per second.
The photosensitive pixels of the sensor can be arranged in a rectangular matrix or along a single line; the former are applied in matrix cameras, the latter in linear cameras.
2D machine vision technology with matrix cameras
The matrix camera performs two-dimensional acquisition of digital images, monochrome or color (with RGB filter), in the area of the shooting field defined by the optics together with the format of the pixel matrix sensor. The resolution can reach the order of megapixels.
This technology is suitable for examining products, or parts of them, whose dimensions are contained in the area of the camera’s field of view and which must be checked individually. Matrix cameras can capture images of objects moving at high speed on conveyor belts and have video resolutions that allow even the smallest defects to be detected.
2D machine vision technology with linear cameras
The linear camera captures two-dimensional, monochrome or colour digital images by capturing one line at a time in a scanning process. In order to construct the image, it is therefore necessary for the object being filmed to run in relation to the camera or vice versa.
Linear cameras are suitable for capturing images of continuously moving products, meeting inspection requirements that require high resolution. They are also suitable for applications in confined spaces, for example when images of the underside of a moving product need to be captured through the gap between two consecutive rollers of a conveyor.
Scanning image capture technology is also applied to inspect graphics and labels on cylindrical objects such as cans. By rotating the can in front of the linear camera, the entire surface is processed in 2D, developing the image in a plane.
SWIR infrared 2D vision technology
Matrix and line scan cameras with SWIR infrared technology (Short-wave Infrared) allow you to carry out inspections that would not be possible in the light spectrum visible to the human eye.
Camera sensors that are sensitive to SWIR infrared wavelengths (between 900 nm and 2500 nm) make it possible to detect characteristics of foodstuffs that absorb or reflect SWIR radiation in a different way that cannot be perceived in the visible light spectrum.
Vision systems with this technology, by analysing the IR wavelengths reflected by products illuminated by infrared, are able to inspect and sort fruit and vegetables according to their moisture content, highlighting bruises and defects, as well as identifying foreign objects mixed in with the food.
Exploiting the characteristic of a particular infrared wavelength, which is strongly absorbed by water (and therefore not reflected), a typical application of this technology is to detect the presence of small foreign bodies, such as plastic residues, among food containing water. By illuminating the product with infrared, the SWIR camera reconstructs images where the food appears dark, while foreign bodies made of plastic or other substances that do not absorb water appear clear as they reflect the infrared light back to the sensor.
3D machine vision technology with laser triangulation
The application of this technology requires that there is relative movement between the image acquisition system and the object to be inspected. It is therefore suitable for generating a 3D digital image of a product to be inspected, which runs on a conveyor belt.
3D laser triangulation technology uses a “blade” of laser light, generally orthogonal to the direction of movement of the product to be detected, and a camera. As the object intersects the thin beam of light, the laser “draws” its profile, which is captured by the camera’s sensor. As the product moves through the laser beam, the camera sensor acquires a series of profiles, which are extrapolated and processed by special algorithms to reconstruct the 3D digital image of the scanned object.
Laser triangulation vision systems can achieve high acquisition frequencies enabling high-speed applications, processing accurate 3D images with resolutions that meet the requirements for measurement, inspection and quality control.
Stereoscopic 3D machine vision technology
Stereoscopic technology allows a 3D digital image to be reconstructed by combining two or more 2D images of the subject, taken from different positions. The acquisition system is generally composed of two matrix cameras taking the same scene from different angles. It is a “snapshot” system that does not require relative movement between the acquisition device and the object to be detected.
Stereoscopic technology reconstructs 3D images based on the concept of disparity: by superimposing the pair of images acquired by the two cameras, each object in the scene will appear doubled with a distance or disparity that decreases with the remoteness of their position from the observation point.
A disparity map is then created using stereo matching algorithms from which the height or depth map is generated to reconstruct the 3D digital image of the imaged objects.
With stereoscopic machine vision, it is possible to reconstruct 3D images of static or moving objects, with both small and large framing fields, but with less accuracy than with laser triangulation.