Greg Blackman looks at some of the machine vision applications in food processing, from the high-speed environment of canning and bottling to inspection in bakeries
When it comes to food, buying British is all the rage among UK consumers. In an effort to reduce the carbon footprint of food and to improve their green credentials, supermarkets are sourcing more produce from inside the UK and labelling it as such. While the rules governing the labelling of food in the UK are complex, and some produce will be more tightly regulated than others (beef, for example, is subject to strict country of origin regulations), supermarkets that claim food is British must comply with the labelling legislation. Machine vision plays its part by ensuring labels match the packet’s contents.
Label inspection is one application where machine vision is used heavily. Typically, the food container will have a code describing the contents and vision systems are used to identify the label and ensure it matches the information held by the code. For instance, tinned goods will contain inkjet codes on the base of the tin to indicate its content, and vision systems are used to identify the label and check it matches the can. ‘The labelling process is a different stage in production to filling and manufacturers must ensure the label matches the tin’s contents,’ explains Mark Williamson, sales and marketing director at Stemmer Imaging.
Williamson feels that canning and packaged goods manufacture is where the majority of machine vision systems are used – and that, while there are vision applications within raw food inspection, these are more niche applications in terms of machine vision.
Ensuring lids are sealed
Aside from labelling, vision is used to check fill levels and whether lids are attached correctly on cans and bottles. The IVC-3D smart camera from Swedish company, Sick IVP, part of the Sick Group, was utilised by a customer in the UK to check whether lids on cans were properly sealed before being cooked. If lids aren’t tightly closed then the content can boil over. This means the oven has to be cleaned, which stops production.
The IVC-3D camera is based on 3D vision technology and uses laser triangulation to acquire an image of the object. It is used for a wide variety of applications within the food industry, from inspecting fresh produce, such as bread or meat, to bottling and canning applications.
‘3D technology is a very powerful tool for inspection in this application,’ comments Anders Murhed, manager of business development at Sick IVP. ‘It is easy to identify any damage to the lid and small variations in the outer rim can be picked up using 2D analysis tools on the 3D image. A simple tool can locate the can and then variations in greyscale will determine changes in height.’
In addition to the angle of the lid and any damage to the top of the can, the smart cameras are used to measure the can’s contents to determine whether it has been filled correctly.
‘Laser triangulation is suited to any application where the shape of the product is critical,’ says Murhed. ‘The technique is largely unaffected by changes in ambient light and is far more robust compared to 2D greyscale images.’
As part of a bottle return scheme, vision is used to check the brewer’s logo on the glass. Image courtesy of Stemmer Imaging.
Within food applications, 3D technology can be used to measure the height and shape of dough before it is baked. The cameras can also be coupled with robotic arms in pick-and-place applications, such as packing biscuits into trays. ‘There is much more variation in these types of produce than mechanically made parts, which makes it more difficult for the robot to grab the item. In addition, 2D imaging techniques don’t give the height information that 3D techniques provide, and this is vital for the robot picker to firmly grip the object,’ Murhed explains.
Inspection at speed
Bottling and canning lines often move at high speed, and vision systems have to keep pace. The LSVision label sorter system from German surface inspection company, Quiss, provides a 360° view of the can using Gigabit Ethernet cameras and fast flash pulses to achieve production rates up to 3,000 cans per minute.
‘Speed and reliability are the two key criteria that vision systems must meet in these types [bottling] of application,’ remarks Jim O’Reilly, president of MCS Vision. MCS Vision provides inspection systems for packaging applications, 75 per cent of which are for bottle, label and cap inspection.
‘Bottles are moving in excess of 1,000 pieces per minute and, in order to identify any defects, the drivers that move the image from the camera to the PC must be fast,’ O’Reilly says. MCS Vision’s cap inspection systems use three cameras from Point Grey Research to gain a 360° view of the cap and neck of the bottle. Point Grey’s Flea2 camera uses a FireWire (IEEE 1394b) interface, which allows fast data transmission up to 800Mb/s. The system will check the fill level, the shape and colour of the cap, whether it is properly seated, and whether the tamper band is intact. When checking bottle labels, four cameras are used and the system will look for tears, print identification, and whether the label is straight, among other factors.
‘The cameras also need to be 100 per cent reliable, as there are tight standards imposed for these types of bottling applications,’ says O’Reilly.
MCS Vision supplies bottle inspection systems to companies all over the US, and the environments in which they are placed can vary dramatically. ‘There can be large shifts in temperature and humidity, which the systems have to be able to handle,’ comments O’Reilly. In addition, the vision systems are regularly washed down and must be housed in protective casings. ‘The cameras are the most reliable part of the system, but a good relationship with Point Grey is vital to ensure any problems that arise are easily overcome,’ he says.
For bottling applications, MCS Vision’s inspection system uses cameras from Point Grey Research to check fill levels, the shape and colour of the cap, whether the bottle is properly seated, and whether the tamper band is intact.
Sergio Morillas Castillo, project manager of vision systems at AnaFocus, also says speed is crucial to inspection applications: ‘We are currently working with a Japanese customer looking to inspect bottles at rates of 6,000-8,000fps. This frame rate is particularly fast, but some bottling applications require imaging speeds in this range.’
Based in Seville, Spain, AnaFocus’ Eye-RIS family of vision systems is based on the company’s Smart Image Sensor (SIS) technology – CMOS sensors comprised of distributed processors and memory within the sensor. The SIS included in the Eye-RIS vision systems allows images to be processed inside the sensor itself and such processing takes place in all pixels in parallel. This provides the high frame rates necessary.
Castillo says there are a wide variety of applications that would benefit from vision systems running low to medium complexity, low-level, image-processing algorithms with low resolution but at very high speeds.
While variables such as fill level and cap inspection are typically what vision systems are designed to look for in bottling applications, a German brewer required a system to identify whether bottles, returned from bars as part of a deposit scheme, were from its breweries. Stemmer Imaging worked closely with a bottle inspection company to install a vision system ensuring only valid bottles are accepted by identifying the brewer’s logo on the inside of the glass – a challenging design feat, as the logo is part of the bottle and the same colour. Specialist lensing and lighting was used, along with Stemmer Imaging’s Common Vision Blox (CVB) Manto software, which recognises the light diffraction pattern around the logo to validate the bottle.
Williamson, of Stemmer Imaging, notes that these types of bottling application are impossible to carry out manually, as the process is too fast. ‘The only solution is to use machine vision,’ he says. Elsewhere in food processing, Williamson says there was traditionally a lot of cheap labour carrying out manual inspection, but manufacturers are trying to automate these processes. For example, machine vision integrated with robot packers are now installed at the end of production lines to package the finished produce, whereas traditionally this was carried out manually.
Automated inspection of freshly-made or live produce has completely different imaging challenges to that of can and bottle inspection. The Norwegian machine vision company, Tordivel, has developed an optical inspection system used to measure the dimensions and detect defects in baked products. Helge Jordfald, marketing manager, explains that the challenge with using machine vision to inspect baked goods is the amount of variation involved. ‘It is a live product and the inspection system has to be able to distinguish normal from abnormal variation.’
The system contains two line scan cameras, one looking from above and one from below. Height is measured through laser triangulation, with a laser line spanning the conveyor and a camera at 45° recording the distortion in the line to identify the height of the pizza as it moves past. Pizza bases have to be within certain dimensions in order to fit inside packaging. In addition, the cameras inspect the pizza base for colour variations looking for burnt patches.
Tordivel’s Scorpion Vision software is an integral part of the system and is trained with a number of images to determine the specifications of the inspection criteria. ‘This application requires rapid image processing and all measurements must be made in less than 800ms to keep pace with the conveyor travelling at 25m/min,’ says Jordfald.
A vision system installed at the bakers, Frank Roberts, ensures tins are completely empty before being re-used. Image courtesy of Capley Marker and Cognex.
Cognex’s In-Sight cameras with laser line scanners are used in an inspection system set up at the UK-based bakers, Frank Roberts, which produces more than two million items each week. The vision system, developed by Capley Marker, UK-based factory automation experts, was used to inspect cavity tins for leftover bread after baking.
Bread production at Frank Roberts is carried out in a continuous loop production system. Dough is placed in tins and baked in batches at 5,200 loaves per hour, six days a week. The bread is then removed from the tins by an automated ejection system, after which, Cognex’s cameras inspect the inside of each tin to detect leftover bread.
‘Any material left inside the tin after de-panning looks the same colour as the tin itself, making it very difficult for the cameras to pick up on,’ explains Elisabeth Allen, sales and marketing executive at Capley Marker. ‘The system uses structured lighting to overcome this, by integrating a laser line scanner with Cognex’s camera.’
Integrating machine vision into the production line increased efficiency and reduced the amount of waste – tins with leftover bread stuck to the base will distort the shape of the next loaf, which will subsequently be rejected.
‘There is massive potential for the use of machine vision in bakeries,’ states Jordfald. ‘Many of the large supermarkets have strict tolerances on product variation – they want equally sized items so that each can be priced the same.’
Uniformity is a requirement for many foodstuffs sold at supermarkets. Videometric, a French company specialising in using vision for metrology, has developed a stereoscopic vision system that aids in cutting Tomme de Savoie cheese into equally sized pieces.
The system consists of a video projector casting moving fringe patterns onto the surface of the cheese and two cameras from Matrix Vision capturing stereoscopic images. The cheese is rotated so that images are taken from five different angles. In this way, the system builds up a 3D model of the entire cheese from the distortions in the fringe pattern lines. ‘From the 3D coordinates, the precise volume of the cheese can be calculated and the cutting angles adjusted to obtain slices of equal weight,’ explains Cécile Scherepin from Videometric.
‘Prior to the installation of the vision system, the cheese was simply cut at equal angles but due to its irregular shape, each slice did not weight the same,’ says Scherepin. Approximations in the cutting process resulted in a margin of error of ±50g in a 200g portion, each of which would be priced the same. The vision system reduced that error to ±5g per piece, a substantial saving. The technique can be used with other food and Videometric is currently developing its system to slice pâté into equal weights.