Machine Vision comes of age

For many of New Zealand’s producers, smarter production lines may be the only way to maintain or expand global market share. And here, as Geoff Anders points out, ‘machine vision’ has a vital role to play.

The traditional image of a Quality Assurance (QA) person watching product whiz by on a conveyor, to seemingly randomly remove an item or press a button to have the machine do it, is being replaced by smart technology.

‘Machine Vision’ is the term for this expanding field.

Machine vision has actually been around for at least 15 years but is really coming of age with well-proven hardware, versatile and powerful software, and experienced engineers. Systems run from single camera stand-alone installations to multi-station projects using many cameras, sensors and lighting, all logging and processing vast amounts of information. This may control robot operation, product position, reject removal, line speed, and labeling or tooling changes. Machine vision may also be used for trend monitoring by comparing images, such as analyzing moles and
melanoma sites.

Systems to suit

Off-the-shelf equipment begins with stand-alone camera units with their own processing capability, through combinations of cameras, scanners, lighting and software pre-defined for very specific common tasks, right up to custom PC-based installations with specially developed software.

Costs for machine vision systems can be substantial, starting at around $10,000 for a single camera and software commissioning. Complex multi-camera installations can quickly add more zeros to the ledger and therefore require serious research to justify and to get it right.

Accurately defining the expected result is very important to the design and outcome of a project. Then the critical items become refinement of camera, lighting and software algorithm. Cameras may be monochrome or colour, each having pros and cons at different tasks. Around 90 percent of installations use mono as it is faster, the information easier to process, and lighting is both simpler and cheaper. A colour system would generally be used only where colour was critical – for example, in grading fruit.

A wide variety of lighting arrangements are available and necessary to suit the incredible variety of product and project. Many use custom setups. LED arrays are most commonly used as they are robust, readily available, have long life, may be configured in many different shapes and colours, and ‘strobed’ to create very high light levels. Strobing can raise light output by a factor of 20, enabling very high camera shutter speeds and consequent high feed rates.

Other types of lighting are used too. Lasers are usually employed for line scanning. Instead of lighting an area these create an intense narrow line across a moving product while a camera takes rapid multiple images that the software analyses to detect surface blemishes, inclusions, tears or damage – or it measures and records perimeter shape, volume, or surface profile. This information can be used to record the position of defects in linear product, change the process, accurately check size or shape against specification and tolerance, control downstream activity, or simply log information to the system.

Lighting is very critical to camera and software performance. If there is insufficient light the camera will not be able to operate fast enough, and therefore may limit production speed or limit image definition. Poor lighting will make it more difficult for the software to interpret the data.

Knowing what type and configuration of lighting will be most appropriate requires experience and technical knowledge. Not really a DIY option for anything other than a very simple and easily defined task.

Stand-alone camera units have integral proprietary software that, in conjunction with programming consoles or PCs, is tuned to a task. This will be very specific and unlikely to be transferable to another task unless it is very similar. Greater versatility may be gained by running from consoles or PCs and software that can be configured for, and switched to, a range of product. PCs, of course, are more versatile with better facility for program and data storage and communication with other components or systems. Once set up, these systems should be low maintenance, fast and reliable with payback in terms of raised line speed, better rejection of defective product, greater product accuracy, recorded history, data logging and elimination of tasks that are either dangerous, RSI intensive or unpopular.

Robots often use camera information to know where to pick an item from or where to deliver it. Component orientation may be important to the success or speed of the task. A camera could detect a foreign object or person in the robot’s operating envelope.

Some uses of vision systems are well developed. Plastic drink bottle cap checking is one of them. Off the shelf systems are available for this and are readily customizable to check many parameters of cap fitting. They can detect cross-threaded caps, missing caps, those of incorrect colour, damaged tamper rings, distorted necks, bad threads or incorrect fill level.

Pre-configured software can verify that a circuit-board’s solder blobs are all present and correct, or detect fiducials (datum points) to position and align the boards for part placement, hole drilling or other operations.

A recent extension of barcode technology is to 2D Matrix Codes - rectangular blocks of black/white code able to hold greater amounts of information in a smaller space. These codes can be read rapidly by smart cameras and are very efficient users of space. They are commonly used in the aerospace and semiconductor industries and handheld scanners are now readily available as the technology spreads into consumer goods. Many other specialised units are available and being developed all the time. Food cans can be inspected for many parameters, injection molded items checked for flaws, items inspected over 360 degrees, extrusions measured, arrays verified, printed items read and verified, lines of adhesive monitored for position and continuity, all by well-established and proven systems.

Mail addresses can be read by vision systems and each item automatically guided to the correct mailbag. Cell phone manufacturers are big users of robots with vision systems to help control them. Some claim up to 300 percent rise in productivity from their machine vision installation. Vision components are designed to interface with PLCs and networks using common communications protocols.

Industrial Research Ltd has pioneered some interesting uses for vision systems technology – such as scanning animal carcasses for volume and shape; fish fillets for fat line to facilitate closer cutting and less wastage; verifying that mandatory ‘use-by’ and production date information is present, legible and correct on food products; correctly identifying almost identical vehicle components that are extremely difficult to identify by eye; mole-mapping that can see below the epidermis for melanoma analysis; timber sizing assistance for sawmills; detection of people in dangerous areas, playing card deck inspection for casinos; selecting kiwifruit by shape to best fit export packaging; better grading of pelts by size and condition; and the more efficient application of chemicals to wildly varying pelt shapes in the tanning industry. And there’s bound to be others we don’t hear about.

A leading reason for installing vision systems is to retain or gain competitive advantage and one problem struck when preparing this article was that businesses were generally happy to show, discuss and demonstrate their systems but wouldn’t allow any proprietary information or images to be published. Quite understandable really, that’s what it’s all about. Many producers are in direct competition with overseas manufacturers that have much lower labour costs. Smarter production lines may be their only option to maintain or expand global market share. In purely local terms more automation could have similar gains in highly contested sectors of the New Zealand market.

Does this pose a question? If a New Zealand plant is becoming unprofitable against those in lower waged countries should it be moved to those areas (where costs will gradually rise so the improvement may be only temporary) or should we make the plant more automated and thus more efficient in the long term? Either way makes workers redundant but the latter option keeps the profits here, supports our technical people and improves our technological position.

Major New Zealand companies in the field of machine vision are CamSensor Technologies and Controlvision, with Industrial Research contributing technical expertise and research where appropriate. Oraka is expanding from initially making asparagus graders. CamSensor has been in business for more than six years with many projects installed, and Controlvision is a more recent entry. Some of these companys’ staff have long term experience in the field.