Vision sensors are on the up with soaring sales figures and an increasing range of applications. The VDMA (the German Engineering Federation) recently granted the devices their own product category, where previously they had been included in the smart camera category. And while in 2006 smart cameras showed a 19 per cent increase in sales in comparison to the previous year, according to a current industry survey carried out by the VDMA Machine Vision Group, the vision sensors boasted an impressive increase of 157 per cent.
Although an exact classification is not always possible, vision sensors chiefly differ from smart cameras in that they are specific to certain testing procedures, the picture resolution is often not so high and optics and illumination devices are already integrated.
But a vision sensor could come in many guises, with sensor companies coming up with a range of definitions for the devices.
Some see sensors as a complete vision system, including Ben Dawson, director of strategic development at the Dalsa Corporation, who says: ‘A vision sensor is a compact vision system consisting of a camera, lens, processor, communication, software, and sometimes lighting. It is commonly used on manufacturing lines, for example to check a product’s dimensions. The capabilities of a vision sensor can be limited to reduce cost and size and to make them easier to use.’
Dalsa’s IPD VA30 Vision Appliance is a vision sensor used for tasks such as gauging, reading barcodes, defect detection and colour measurement. Dawson says: ‘Dalsa’s IPD vision sensors are mostly sold to end users in industrial automation markets.’
Dalsa makes high-performance cameras and vision processors for custom and volume users with its sensor-incorporated cameras being used on high profile projects, such as the Mars Rovers, while Dalsa vision systems are used to inspect flatpanel televisions during manufacture, for example.
And Kirsten Drews, head of the product marketing for machine vision at Siemens, has a slightly different take on a vision sensor and says: ‘A vision sensor is positioned between optical proximity switches and more complex machine vision systems. It combines the simplicity of setting parameters instead of programming with sophisticated inspection algorithms. It regularly is set up as a compact all-in-one device including processor, communication, sensor, optical system and illumination.’
Siemens’ portfolio of factory automation sensors includes proximity switches, machine vision systems and identification of products and systems. The vision sensors fit in between simple sensors and smart cameras. The application area is where either the good or bad decision of a binary sensor does not sufficiently fulfil the required task, or where objects become so complicated that applying binary sensors is too difficult. Its vision sensors are centred on certain tasks, like colour recognition, height detection, 2D code reading and checking the presence, integrity, orientation and quality of objects.
And both Siemens and Dalsa see vision sensors as a small but important part of their wider workloads. Drews adds: ‘[Vision sensors are] an attractive high growth market segment and generate interest with many first time users of machine vision products.’
Vision sensors are incorporated in many of Basler’s cameras.
Teething problems
There are a number of challenges when designing and building a vision sensor, according to Dawson, who says: ‘The major hardware challenge is balancing size, performance, and cost. Collateral challenges include thermal management, environmental protection, and how to connect to the vision sensor.
‘The software challenge is to balance flexibility and capability with ease of use and performance. Dalsa’s IPD iNspect software achieves this balance by using an intuitive, graphical interface for setting up the vision appliance while allowing a rich set of measurements with good performance.’
Accuracy also has to be taken into account when producing a vision sensor. Dr Joachim Linkemann, product manager at Basler Vision Technologies, says: ‘Accuracy might be specified in pixels, reliability in reading a barcode, etc. Consider measuring the diameter of a part. A vision appliance can measure such dimensions to 1/25 of a pixel. Depending on the number of pixels in the camera and the optics used, this might translate into to micron accuracy when a microscope is used or millimetre accuracy if a larger part is dimensioned.’
But the vendors have different ideas on how to measure a vision sensor’s accuracy as well, as Siemens’ Drews says: ‘Another way of defining accuracy would be to look at the repeatability of inspection results. Optimally a vision sensor would give the same results every time it looks at the same object. Unfortunately environmental influences, like changes in lighting conditions, vibration etc. or even unstable algorithms used in the product, can make quite a difference in reality.’
‘Here [at Siemens] we are trying to develop algorithms that give stable results even under changing conditions, and also try to provide optimal lighting and lenses integrated in the product to do the required job,’ Drews adds.
But accuracy is usually not a vision sensor’s primary concern, as Drews explains: ‘Often accuracy is defined by the size of the smallest aberration that can be found in a certain field of view. The accuracy of a vision sensor depends a lot on the quality of an optical system, and on the resolution of the sensor chip the algorithms are actually using in the calculation. In general vision, sensors are not aimed at high accuracy applications and the quality of the optical system is not designed for solving high-end applications.’
Simple functionality is the primary consideration, according to Drews, who says: ‘The first and foremost challenge is to really provide an easy-to-use product, which can be used without specific vision knowledge.’
And, of course, size and cost are important too, as Drews adds: ‘[Another] difficulty lies in packaging this product in a compact and robust housing and keeping the cost down. Some of the low cost vision sensors have a tendency of requiring elaborate parameter settings in order to work at a decent level of repeatability, and do support only very limited communication integration. This leads to extra cost at the customer site for setting up the sensors and testing as well as for additional hardware that needs to be ordered for proper automation integration.’
But when designing a vision sensor, there are always risks, according to Linkemann, who says: ‘Specifications from a data sheet often sound quite reasonable, but when integrating the sensors, sometimes really strange behaviour occurs, such as dependency on temperature, beat frequencies, noise from various sources, tolerances within batches, defect pixels, etc.’
‘If you have eliminated one error you will disclose the next one. Especially for CMOS sensors, which have lots of disclosure levels,’ Linkemann adds. Basler integrates sensors into the cameras it develops, and its product portfolio includes monochrome and colour vision sensors.
But companies are keeping up with the increasing demands on the vision sensor market, as Drews adds: ‘Some suppliers focus on the very low cost end of the market, making vision technology available to many new users. Others focus on the functionality of the products and include more elaborate connectivity functions, diagnostics support, and several options. Furthermore in addition to highly specialised products that can solve a certain set of applications without much setup required, there are now products available that specifically target a broader range of applications with less demanding requirements on accuracy.’
Applications
The number of ways vision sensors can be used is endless, with common applications including checking that all the components of an assembly are present and in the right place, checking the dimensions of the components are correct, the component position and orientation are within the sensor’s field of view (so that a robot can pick it up and place it in an assembly), counting the number of objects present, and colour checking of components.
Dalsa supplies vision sensors to a range of markets, including automotive and electronic manufacturing, to packaging applications, to robot builders, and for medical instruments, biotechnology and pharmaceutical production. Dawson says: ‘Vision sensors are most often used for a single-point visual inspection task in automated manufacturing.’
The Inspector Vision Sensor from Sick IVP.
Basler tends to focus more on cameraembedded vision sensors, as Linkemann says: ‘Cameras with vision sensors are used in various applications, often more than you can imagine: wafer and semiconductor, medical, intelligent traffic systems, food sorting, postal sorting, barcode reading, and so on – generally speaking, cameras with vision sensors can be used anywhere an “artifical eye” is needed.’
And a sensor’s design can alter depending on its application area, according to Dawson, who says: ‘Some vendors make physically different vision sensors for specific tasks such as gauging, colour measurement, or barcode reading. Others consider their various software applications within a single hardware package to be different types of vision sensors.
‘Dalsa has a small number of hardware vision sensors types that differ mainly in their size and performance. These sensors are customised by the software used, but we don’t consider each software application to be a different sensor. Our approach makes it simple for a customer to select and use our products.’
Siemens’ vision sensors are designed for any vertical market in the factory automation arena. Typical markets are the packaging industry, both for pharmaceutical and food and beverage applications, the suppliers of electrical and electronic products and components, the automotive industry and the manufacturers of consumer goods.
And the type of Siemens sensor used can change from application to application, as Drews says: ‘Our products do focus on certain sets of applications. Whereas Simatic MV220 is a colour area sensor, that can easily differentiate between different colour patterns or identify certain colours, Simatic MV230 is a height detection sensor, which uses a laser line to identify and recognise a height profile of the inspected object. The Simatic VS110 and VS120 are built for object inspection in regard to shape, type or position. While VS110 focuses on contours of the objects only, VS120 uses the complete patterns available on the object, like outlines, imprints, edges, etc.’
On the up
The future is bright for vision sensors though, as Dawson says: ‘It is an up and coming area, driven by decreasing hardware costs and improving software algorithms and user interfaces. As cost decreases and ease-of-use increases, it becomes cost effective to add a “spot” of vision to a manufacturing process that might have used product sampling or quick visual inspection for quality control.’
Siemens’ Drews agrees that the market is growing and evolving, adding: ‘The vision sensor market enjoys significant growth rates. Aside from the possibility of extending vision applications in areas where cost considerations once were a problem, vision sensors do also cover an increasing amount of different applications.’
And Basler’s Linkemann is seeing a trend for improving the pixels within vision sensors, as he says: ‘Vision sensors often get smaller pixels, either to increase the number of pixels with the same size of the wafer, or to do more sensors. The problem is that the optics are not able to resolve the small pixels anymore and, besides that, the speed and sensitivity of the sensors has also increased.’
Dawson says the software used within vision sensors must keep up with the hardware improvement, as he says: ‘The relentless improvement in hardware and somewhat slower improvement in software has decreased the size and cost of vision sensors, while maintaining or improving their performance and flexibility.’
‘The improvements in cost, size, and performance provided by improving hardware will need to be matched by improvements in the intelligence of the software and the ease-of-use,’ Dawson adds.
But it’s early days for vision sensors and thus far the market is doing well, as Drews says: ‘The vision sensor segment still represents a very young market segment, which still is in the process of being developed. Today we find a much more complete offering in the market, not only application-wise, but also regarding the differing criteria of the products.’
