Smart cameras simplify manufacturing processes

Smart cameras simplify manufacturing processes

Marcel Singleton

As the prices of machine-vision systems and components continue to drop, they are being used with greater frequency in less-sophisticated automation systems and as add-ons to older equipment installed in manufacturing lines. In a relatively simple application, a manufacturer of tool-polishing equipment successfully uses vision technology to optimize an assembly operation characteristic of the type of automation that these lower-cost vision systems are well suited to support.

The application requires that metal pins be presented in the proper orientation so that the pin end specifically textured to foster a secure connection is always the one inserted into a felt cylinder. A helical vibratory conveyor transports the pins to the assembly station in an inclined position. Baffles force the pins to proceed serially and ensure that those lying off-axis are pushed back into the supply bin so that they can be presented again in proper alignment.

One problem is that the pins can exit the serialization stage and be transported along the axis of movement with either the smooth or textured end facing forward. The assembly station could--if operating blindly--install the smooth ends of pins into some felt cylinders.

Vision solution

A vision system resolves this problem by identifying misoriented pins and initiating action to prevent them from advancing to the assembly stage. This sorting is accomplished by a part-in-place sensor--positioned after the serialization stage--that triggers the camera to acquire an image of each passing pin. The system determines whether the pin is misoriented and, if it is, sends a signal to a programmable-logic controller that immediately triggers an air ejector to blow that pin back into the supply bin for recycling (see Fig. 1).

What appears to be merely a camera is, in fact, an entire vision system. Known generically as a smart camera, this type of camera is a self-contained system that consist of a lens mount, charge-coupled-device array or complementary metal-oxide silicon sensor, integrated image and program memory, an embedded processor, a serial interface, and digital input/output. Some models also provide Profitbus, Fieldbus, or Ethernet interfaces.

To meet the manufacturer`s specifications, the camera must be low-cost, programmable, and usable by individuals with no knowledge of vision tools and algorithms. In addition, it must not require the services of a systems integrator or third-party consultant, and it must have as little physical impact as possible on the existing setup.

The Intelli-Cam smart camera from Cam-Control (Nuremburg, Germany), based on the VC series of smart cameras from Vision Components (Karlsruhe, Germany), was selected for this task. It is a complete machine-vision system that includes up to 8 Mbyte of RAM, 2 Mbyte of flash memory, a powerful digital-signal processor, a high-resolution sensor, and shuttering speeds from 1/100,000 to 20 s. The camera is compact (4.75 ¥ 2 ¥ 1.5 in.), with self-learning features that require no machine-vision knowledge. It is set up via a simple, Windows-based user interface.

Synergistic algorithms

By incorporating a self-learning capability into the camera-system software, the camera can be "trained" to recognize acceptable parts or configurations (see Fig. 2). The camera is trained by using a mouse or trackball to highlight the area of an image containing the part or configuration. The software then analyzes the image, calculates its essential characteristics with synergetic processing algorithms, and creates a reference prototype. During on-line operation, the camera identifies as unacceptable any parts presented to it that are insufficiently similar to the prototype. If trained on different kinds of parts, the camera can sort them into their respective categories.

The synergetic algorithms that provide the smart camera with its self-learning capability evolved from academic research that deals with the laws of self-organization (synergetics) creation of structures from unorganized conditions. The algorithms are based on the premise that pattern recognition can be accomplished by modeling in reverse the dynamics by which patterns are generated in physical and natural systems. Synergetic computers start with a structured sample-which, for object recognition or inspection, is an acquired image of a part of interest--and work backward to determine the essential characteristics of the sample.

During the training phase, these calculations are done for several objects of a given class, and the results are combined into a single mathematical definition that efficiently and effectively describes that object class. Software implementing synergetic algorithms was initially carried out on massively parallel computing platforms. Over time, researchers succeeded in reducing the associated mathematics to problems that can be solved using a PC or a digital-signal processor, at speeds sufficient to provide the kinds of learning and recognition rates required by industry.

Synergetic processing has several advantages over neural networks and other common training-based approaches. Training phases are typically measured in seconds and the resulting reference parameter sets are small, requiring less processing. In addition, excellent results can be obtained with only five training samples. The fact that no unacceptable samples are required for training purposes is a strong benefit.

Industrial applications

Smart cameras provide solutions in a growing number of industrial applications. These include checking the orientation of manufactured parts and electronic components and confirming the presence and proper positioning of labels on products and packaging.

These cameras can also check the position, correctness, and quality of logos and/or text that has been machine-marked on labels, keypads, sporting goods, automobile safety glass, eyeglasses, and integrated circuits. Components associated with a given assembly can be verified as being in place and undamaged, and the presence and integrity of blister-packaged pills or tablets can be verified. In addition, the proper settings of switches can be checked, and items can be monitored for proper sorting into different categories for further processing or packaging.

A great benefit of self-learning vision systems is the ease with which machine operators, technicians, and assemblers can use them to carry out feasibility analyses on applications of interest. These individuals can now routinely discover new uses for machine vision in manufacturing plants and laboratories. The engineers and automation specialists who previously had to handle all vision-related evaluations and projects are then free to focus on more sophisticated tasks that require high degrees of expertise or special-purpose vision systems. o

FIGURE 1. An inexpensive smart camera can easily perform simple sorting applications to ensure that parts are correctly assembled.

FIGURE 2. Using synergistic algorithms, the camera system prevents pins from advancing to the assembly stage by identifying the orientation of the smooth and textured ends.

MARCEL SINGLETON is an independent vision and automation engineer working with Vision Components, Litzenhardstr. 85, D 76135 Karls ruhe, Germany.