Dispelling the myths about machine vision systems

Machine vision is a branch of engineering that uses cameras in the context of process control. Images are collected and analysed and the data extracted used for controlling an automated process or action. When we refer to machine vision systems we are typically referring to smart cameras or vision sensors - devices that possess image capture capability, are able to extract information from captured images and make decisions that are used in wider automation systems.

Developers of equipment realised early on that the possibilities for applying machine vision technology were almost limitless and very quickly appreciated that almost any material handling or logistic process that involved object inspection would suit a vision system.

Typical applications have grown to include detection of faults and flaws, non-contact measurement, part sorting, code reading, continuous flow dimensional gauging and position detection and rotation.

Resistance

Unfortunately, machine vision systems have developed a reputation amongst some managers and engineers as being expensive, unnecessary, superfluous and possibly even incapable of performing the very tasks they are expected to execute.

These are all attitudes that have come about because early vision systems were indeed inflexible, cumbersome and costly.

Figure 1: Early Automatix Inc machine vision system at a trade show in 1983. The camera on a tripod is pointing down at a light table to produce a backlit image, which is then processed in the monolithic controller, probably employing a 68000 microprocessor.

Many vision systems were installed in the early days at great expense but, because of inexperience and limitations with the technology, they may have been used briefly before being left to gather dust in a corner. However, machine vision has come a very long way in the 30 years since it first appeared, even more in just the last half decade.

From rudimentary beginnings, machine vision suppliers have developed machine vision systems into ‘smart cameras’, integrated vision systems which combine image capture capability and decision-making technology into one compact device.

A typical machine vision set up today might include a 2D or 3D smart camera or sensor, lenses and filters, lighting, an image acquisition and analysis software package, plus the cables and brackets necessary for integration into a production line or machine.

Modern smart cameras are self-contained, standalone units with built-in CCD or CMOS image sensors, image processors, image memories, communication interfaces and industry-standard outputs for connection to PLCs where further processing of the image or information can take place. Most devices are now no larger than an ordinary industrial or surveillance camera but rival PCs in terms of their processing capabilities. The six myths While resistance to machine vision systems manifests in many forms, it is possible to distil it down to six clearly defined misconceptions - fallacies which can be easily dispelled.

Figure 2: Fish fillet profiling using a Sick IVC-3D stainless steel camera.

Machine vision is too expensive

Never has this been further from the truth.

While many managers and engineers baulk at the costs of purchasing vision systems and integrating them into their production processes, they ignore the real costs to their business in terms of wastage, rework, returns or lost productivity due to employee illness or absence and forget about the real benefit of improved productivity thanks to greater processing speed.

It is far better to think of machine vision systems in terms of the return on investment and to take into account the entire effect on one’s production system, not just the initial capital outlay.

Vision systems are now not just economical to purchase, but are robust, compact, reliable, practical, flexible and easy to integrate.

The technology is not up to it

The pace of development, even in just the last half decade, has been tremendous and products are now readily available that are flexible enough to be able to fulfil a multitude of inspection tasks, or specialised enough to be able to perform one inspection task exceptionally well.

Modern machine vision systems are fast - some capable of speeds of up to 5000 inspections per minute, they can be accurate down to hundredths of a millimetre and they are reliable, with modern devices that have fully integrated lighting and ethernet communications capabilities.

Troubleshooting is difficult

Modern software packages employ digital image processing techniques that are designed to perform many and varied analyses. The user interfaces are designed to work within standard PC environments, be easy to learn and present only the information that is relevant to the inspection task at hand. Predefined inspection packages are provided and emulator tools provide easy methods of finetuning and testing of solutions without disturbing production.

Commonly used techniques include pixel counting, edge detection, template matching, thresholding, segmentation, blob extraction, pattern recognition, barcode reading and optical character recognition.

Ideally, vision systems should also incorporate image logs for analysing production problems. All this results in fast and trouble-free solutions for a diverse range of part inspection applications and the ability to resolve production issues on the spot.

Support is non-existent when problems do occur

As sure as many companies appeared in the early days proclaiming that they were ‘vision experts’, they rapidly disappeared. Unfortunately, this left many end users with vision systems that were unsupported.

Most integrators and component suppliers have now been in the business long enough to establish themselves as genuine specialists in the field and are capable of providing not just the original solution, but the service component that goes along with it.

Trust only those companies that do have long experience in the field and are able to demonstrate their ability to solve applications similar to yours.

There is a loss of flexibility in the production process

Modern vision sensors are designed to be easy to adapt and integrate so it is possible for them to be reconfigured to suit different products and situations. Some are able to store images and actually perform product changes 'on the fly' with only minimal disruption to production.

It is even possible for them to be shifted from one production station to another and settings altered if a production line needs to be modified.

It puts people out of work

Rather than render employees redundant, anecdotal evidence suggests that the use of vision systems in production processes actually frees employees up to be utilised in jobs that are more beneficial for the company and more worthwhile and fulfilling for the employees themselves - a much better use of an extremely valuable resource.

Rather than replace staff, some vision systems are now so cost-effective that they can be used in addition to manual labour as a cross-check for inspection of manually assembled products.

Is there a place for a vision system in your production?

Always remember that when choosing and implementing machine vision technology, a simple solution that works is a preferable solution. Ask yourself the following questions:

• What outcome is required? Often the sole driver of vision projects is the need to increase production speed but it can just as easily be a desire to reduce labour costs, improve quality and increase production yield - all factors that are vitally important in the food and beverage industry.
• Which technology should be used? Define your task. Speak to experts who have the experience and knowledge of your application and who can advise whether standard hardware or a customised solution is applicable.
• What does a typical vision project look like? Ask for typical application information and make sure that whomever you engage is fully committed to guiding you with your project.
• What environmental and set-up problems will be encountered along the way? Often there are heat, vibration, dust and humidity concerns. Mounting space, cycle times and illumination must be taken into account. As should the special requirements of food production areas such as waterproof housings, stainless steel construction etc.

If you’ve asked yourself all these questions and have decided that a vision system is the answer, then proof of concept is easy with modern devices. Trials with vision sensors can be very quickly set up in a workshop and translated to the production facility with only finetuning of the site illumination and other minor integration tasks.

By: Sean Carter, Product Market Manager, Advanced Industrial Sensors & Encoders, Sick Australia