Machine-vision applications such as high-speed web inspection demand high-speed linescan cameras, dedicated frame grabbers, and custom OEM software. Increasingly, however, there is a demand from large-volume end users for OEMs to provide integrated machine vision and industrial control at low cost. While such systems often do not attain the throughput or level of customization of specifically tailored machine-vision systems, they do provide off-the-shelf solutions for pharmaceutical, packaging, semiconductor, and automotive-inspection applications.
In the pharmaceutical industry, for example, relatively simple image-analysis functions such as pill counting, blister-pack inspection, and date-code reading are performed. In the packaging industry, products must often be analyzed for broken seals, cap presence, label presence, and character verification. Similarly, in the semiconductor industry, IC component size, tolerance, pin count, and length are measured.
To perform these inspection functions, a host of companies are offering stand-alone vision systems that provide a method for rapidly deploying combined vision and process-control systems. Offering PC performance in rugged enclosures, these machine-vision systems are effectively low-cost industrial computers that integrate image capture, processing, and display functions. In the design of such systems, developers have sought to integrate I/O and networking capability to allow these systems to host imaging and industrial-control systems. By doing so, these vendors have replaced traditional camera/frame-grabber/OEM-software combinations with integrated machine-vision systems that do not require a PC for setup, configuration, or maintenance.
"New generations of compact machines and inspection stations must satisfy multiple applications and multiple parts inspection to justify their cost," says Gustavo E. Vargas, vision product manager at Aromat (New Providence, NJ). "And these integrated machine-vision systems provide attractive features for large manufacturers and plant-floor applications where smaller compact machines are usually controlled by micro PLCs," he adds.
To meet these demands, Aromat's ImageChecker A200/100 system, for example, comes with built-in I/O, eight programmable outputs for external signals to rejection mechanisms, and a built-in video port for video display (see Fig. 1). To interface to supervisory control and data-acquisition systems, the ImageChecker provides a serial interface to Aromat's FPSigma PLC that houses Profibus, Ethernet WebServer, and ASCII Modbus communications modules. This serial interface can also provide direct serial communication with the FX0 PLC from Mitsubishi (Vernon Hills, IL), the F381-C40 from Omron (Schaumburg, IL), and the SLC500 from Rockwell Automation (Milwaukee, WI).
To provide a larger field-of-view inspection or a dual-type inspection, such as area and height or label and marking on two separate product locations, the A200/100 system offers dual camera controllers and optional double random-speed cameras suited for inspecting 1200 to 1800 parts/min. "Because the controller hardware remains the same, the overall increase in hardware cost is roughly $500," says Vargas. To drive 24 Vdc and provide output signals to rejection mechanisms or directly into a PLC I/O, the ImageChecker's eight programmable outputs can be used as NPN or PNP transistor-type outputs. Additionally, output status can be sent serially using built-in serial ports. No software is required because the system outputs can be programmed in the ImageChecker operational firmware.
Despite such developments, the deployment of such machine-vision systems is generally not performed by end users. "Machine vision," says Carl Gerst, In-Sight product marketing manager at Cognex (Natick, MA), "is not a plug-and-play industry." To illustrate, Gerst cites the company's In-Sight family of vision sensors. "We were very cautious when introducing this product family," he says, "as we realized that every machine-vision application would be different."
For example, the In-Sight 3000 vision sensor offers dual processors—one for vision and the other for networking—that deliver vision performance across the remote network of In-Sight sensors (see Fig. 2). Interestingly, the major challenges customers faced in deploying an In-Sight system were those of communications, integration, and I/O. Gerst cites one customer who wanted to upload daily parts data from the sensor to a remote Ethernet-based server. "To support this function," says Gerst, "Cognex has developed beta software that enables an enterprise server to poll any In-Sight system for specific information at any give time." This software, when completed, will be offered to all In-Sight customers.
Gerst sees application for such products in discrete parts inspection in industries such as automotive, pharmaceutical, and packaging. While some products can perform simple presence/absence functions, they can also carry out critical gauging applications at speeds of up to 1500 to 2000 parts/min. These functions can be programmed using the In-Sight's built-in spreadsheet software.
"Interestingly, questions regarding deployment of In-Sight sensors rarely turn to the underlying hardware," says Gerst. "When you discuss selling a large-volume end user such a system," he says, "what matters is the system's functionality, not that it contains a specific processor, gate array, or I/O controller device." Systems integrators, may, for example, wish to create their own graphical user interface (GUI) when using a number of Ethernet-based networked systems. In such systems, multiple In-Sight sensors could be networked across the Ethernet under control of a host PC, also Ethernet-based.
"In such a scenario, the systems integrator could develop a GUI using Microsoft Visual Basic or Visual C++ on a host PC. This applications interface would then make calls to each In-Sight system's application to set up a new production run or inspection task," says Gerst. He adds that the company has seen much demand for Ethernet and Ethernet IP-based systems, but the demand for networked systems based around Profibus, DeviceNet, and CANbus has been limited.
"While these networking protocols have enough bandwidth to support simple transducers such as load cells, accelerometers, and thermocouples," says Gerst, "they cannot support the bandwidth required by vision sensors." Accordingly, most newly deployed machine-vision networking systems are being based around higher-speed topologies such as Ethernet," he says.
The successful deployment of such systems is heavily dependent on the vision tools being offered. "While the In-Sight 3000, for example, supports a number of gauging and measurement tools that can be tailored for several inspection tasks," says Gerst, "one of the underlying vision tools used in most applications is PatFind, a derivative of the company's PatMax geometric pattern-matching tool. What is of utmost importance in many of these applications," says Gerst "is that parts are correctly identified in a number of different orientations. Only then can a successful inspection take place with other gauging and measurement tools."
Chuck Bourn, marketing manager at PPT Vision (Eden Prairie, MN), also sees Ethernet as the main networking protocol for the factory floor. "Our information indicates that 80% to 90% of prospective machine-vision customers want Ethernet," says Bourn. The company's latest stand-alone imaging system, dubbed Impact, supports Ethernet protocols through Inspection Builder software," says Bourn (see Fig. 3). "This software incorporates Ethernet protocols to a network between a host PC and Impact for programming and run-time operations."
In the future, Impact will have a second networking port available. Therefore, while the initial port provides data and images to a client PC or network, the second Ethernet port will be controlled by specific Inspection Builder tools or functions, which will serve data to various external devices that are DeviceNet- or Profibus-compatible. An I/O controller embedded in Impact can control lighting, logic functions, and discrete I/O.
Unlike the execution time for a PC-based system that can vary due to operating-system (OS) variables beyond user control, Impact communicates to a host PC using Ethernet and is not limited to a particular OS. "Compatibility issues of frame-grabber drivers working with a particular style of PC or OS are eliminated since Impact uses an RTOS, which makes the inspection cycle time more deterministic," says Bourn.
Another disadvantage to the PC-based approach is the cost of selecting and integrating software and hardware components into a complete platform. With Impact, the digital frame grabber, image processor, I/O controller, and software are combined into an industrial unit that can be mounted in factory environments. The cost for a complete system including processor and camera starts at $6500. The PCs are connected by Ethernet.
"PC and frame-grabber combinations can suffer from bottlenecks in which, as more cameras are added, performance decreases because image-processing functions are funneled through a single point," says Bourn. "The cost per point of inspection allows a single Impact to be dedicated at each inspection point. This eliminates bottlenecks and performance is maintained as more cameras are added," he says. Individual I/O tools, timers, and logic functions in Inspection Builder software allow system developers to provide functions such as gauging, defect detection, pattern recognition, and optical character recognition/optical character verification.
The 4Sight-II industrial computer from Matrox (Dorval, QC, Canada) contains Ethernet capabilities by way of a built-in 10/100-Mbit/s local-area-network controller (see Fig. 4). Other networking standards such as Profibus, DeviceNet, and CANbus can be supported with appropriate PC/104-Plus expansion modules. "At present," says Fabio Perelli, Matrox product manager of frame grabbers and stand-alone systems, "there is a movement devoted to migrating users of proprietary industrial networking standards to Ethernet/IP. Moreover, wireless Ethernet networks, such as IEEE 802.11 or WiFi, are a natural progression and viewed as extremely advantageous for industrial applications," he adds.
"Ethernet is the emerging standard for industrial networks," says Phillip Heil, chief technology officer at DVT (Norcross, GA), "and we are seeing applications that use Profibus over Ethernet (Profinet) and Modbus over TCP/IP. Rockwell Automation (Milwaukee, WI) also has opened the Ethernet/IP protocol to promote the standard." The DVT Legend family of smart sensors supports many protocols over Ethernet. For support of proprietary networks, the DVT SmartLink device uses standard Ethernet communication technology to transfer images from up to 16 networked SmartImage sensors so that multiple inspections can be viewed at once with an inexpensive monitor or to freeze images for operator intervention. "This allows a network of DVT devices to communicate data via registers to DeviceNet or Profibus," says Heil.
Adding I/O networking capability to stand-alone machine-vision systems provides automation engineers with low-cost inspection and control systems. With bundled image-processing and process-control software, these systems are rapidly replacing PC-based vision and process-control systems in many factory-automation applications. In the future, this level of integration is expected to continue, with systems developers offering higher embedded performance, tailored application software, and integrated networking support.
Click here to download Modular Stand-Alone Machine-Vision Systems table.
By Andrew Wilson, Editor
Vision sensor targets manufacturing automation
In the past, machine-vision systems were used for product inspection after manufacturing and were not part of mainstream assembly-line industrial control. Now, manufacturers want to detect bad products in process, as they cannot afford the cost of scrap products. They want to monitor products in real time and detect and correct process variations before products must be scrapped. To successfully serve this market, suppliers of machine-vision systems must provide products that perform inspections to ensure quality and provide real-time process feedback. These products must be affordable, usable, and supportable by production staff.
To meet these demands, Banner Engineering has introduced its PresencePlus Pro Vision Sensor, a camera-based sensor that performs inspection applications in place of multiple discrete sensor configurations or complex vision systems. Alan Carr, a body-shop engineer at General Motors (Detroit, MI; www.gm.com), is a pioneer customer of the sensor. "The teach system, video output, and low price of the sensor are its biggest strengths," says Carr. To implement the teach system, users follow a guided setup sequence to create custom inspections. An inspection is set by illuminating the target, focusing the camera, and selecting the appropriate vision and analysis tools. A range of inspection tolerances can be set up automatically or manually.
"This tool is valuable for eliminating guesswork and applying the statistical variation of a sample to a tool's acceptable operating limits," says Carr. "Establishing the optimum focus and lighting of an inspection can be accomplished using the video output from the Banner controller. Even with minimal knowledge of vision systems, an inspection that correctly tests and rejects bad parts on a production line can be developed rapidly," he adds.
PresencePlus Pro is set up using a remote PC. After programming, the system is used in stand-alone mode to perform multiple inspections simultaneously and inspect for both translational and rotational variations. Parts moving down a production line or web are inspected without having to be orientated in exactly the same way in the camera's field of view.
Ethernet and flexible discrete I/Os are used to communicate data, measurements, statistics, and information for system process control over Ethernet via TCP/IP and/or serially with an ASCII string. Discrete I/O is accessible via a pluggable terminal block and is programmable electronically (NPN/PNP) and logically (pulse, delayed, or latched). Stored inspections can be selected via discrete inputs on the terminal block. Inspection tools include both a "locate" and a "pattern find" tool. Other vision tools perform image-analysis functions, average gray-scale value, location of selected features, location of edges, and location of patterns.
Manager of Product Development