Technology streamlines the mining of data for quality control.

As each bottle of beer comes whisking by the inspection station along a conveyor, a sensor glances at it briefly. On-board intelligence built into the sensor interprets patterns on its label to verify that each bottle not only received a label but also got the correct one. In the glance, the sensor also finds the label's edges and checks its straightness before allowing the bottle to continue toward the final packaging line. If it finds any missing, incorrect or crooked labels, the sensor sends an alarm to the controller so it can pull the bottle from production.

The vision sensors used for inspecting beverage bottles are typical of today's high-end sensors. Rather than simply collecting data and transmitting them to a controller, this class of sensors can make a number of decisions locally, thereby relieving controllers of computations. In the case of vision sensors, the relief provided by on-board intelligence is not trivial.

"They are looking at between 13,000 and 1.3 million pixels of information," says Jeff Schmitz, corporate business manager-vision sensors, Banner Engineering Corp. (Plymouth, MN). "They run some very sophisticated filters and algorithms."

A vision sensor contains the necessary intelligence to check several features on these tiny components. Source: Banner Engineering Corp.

The intelligence in these smart sensors is found in what is called the transmitter, the portion of the unit that messages the raw electrical signal into a form that makes sense to the outside world. Vendors assemble the intelligence in their transmitters from components such as microprocessors, field-programmable gate arrays, and various memory and storage chips. In the case of Banner's PresencePLUS vision sensors, the digital imager inside the sensors sends the data to a Motorola processor, which uses field-programmable gate arrays to accelerate results.

Although Banner's vision sensors are an example of the many sensors that contain all of the electronics in one unit, some manufacturers separate the basic sensor from the transmitter to protect the electronics from ambient conditions. The Foxboro division (Foxboro, MA) of Invensys Process Systems, for example, splits one of its pressure sensors from the transmitter to shield the electronics from heat. The two ends communicate with one another by cable.

Two developments have made putting more intelligence on sensors practical. The first is the enhanced firmware, the fruit of both the continuing price-

performance gains in digital electronics and incremental strides in applications development. The cost of computing power is no longer the obstacle that it once was, and vendors have learned to encode a number of enhancements into their sensors.

"For more than 12 years, we've designed our own custom ASICs, application-specific integrated circuits," says Wayne Meyer, product manager, industrial sensors, Sick Inc. (Minneapolis). In that time, the company's engineering staff has learned to encode in its ASICs rules for distinguishing good data from random events in the surroundings such as flickering fluorescent lights, voltage spikes and temperature fluctuations.

Many Ways to be Smart

Of course, the intelligence can make any number of enhancements. "Depending on the amount of computing power in the transmitter, you can message the data and calculate just about any parameter that you're looking for," explains Rick Gorskie, senior product manager, Honeywell Process Solutions (Phoenix).

For example, a sensor can calculate the rate of temperature change and generate an alarm if it rises faster than normal. Consider a process in which the temperature is supposed to increase at 1 degree per hour. "If it suddenly starts rising at 2 degrees per hour, the transmitter can be configured to tell somebody about it," says Gorskie.

Multivariable sensors can apply other algorithms to calculate flow and density on the fly. Sure, controllers have been performing these and similar calculations for a long while now, but the point is that they need not anymore. "If you have the device calculate density for itself, then you've saved that time on the system," says Brian Gregg, manager of pressure and temperature research, design and engineering for Foxboro Measurement and Instrumentation. "It allows for the distribution of signal processing."

The intelligence available today also can influence a sensor's accuracy. "It can apply curve-fitting algorithms and temperature compensation algorithms at the point of measurement," says Gregg. Because the accuracy of a scale varies with the magnitude of the measurement, one simple form of compensation adjusts the scale to fit the reading. "Our multi-range product will automatically switch you from one scale to another, depending where you are in the range of measurements."

Another form of compensation accounts for ambient conditions. Temperature compensation is particularly important in many applications because temperature affects analog-to-digital and digital-to-analog conversions. Most electronic measurement signals begin as continuous, analog responses to whatever physical phenomenon that the sensor is measuring. The transmitter converts this initial analog signal into a digital one before applying the appropriate algorithms to it. If the signal is going back onto the bus as 4 to 20-mA signal, for example, the transmitter must convert the digital signal back into an analog one.

Because temperature affects the converter's performance, "there is a chance of introducing some error every time that you do that," says Gregg. "So you might want to compensate for the temperature of the electronics."

Intelligence programmed into this vision sensor from looks for filling tubes that might have come loose and fallen into a bottle, causing the liquid to squirt onto the line rather than filling subsequent bottles. Source: Banner Engineering Corp.

Making the Most of Diagnostics

On-board intelligence also can monitor the sensor's internal circuitry to perform self-diagnostics and tell the appropriate person when it suspects that something is going wrong. "In a few cases, we've added internal sensors that will check moisture levels and other diagnostics," says Schmitz at Banner.

For vision and other photoelectric sensors, the most frequent self-check is the intensity of the light entering the sensor. A steady weakening of the signal could indicate that dust or dirt has accumulated either on the sensor itself or on a reflector that might be sending light to a photoelectric eye. Dust or fog in the air also can cause a weak signal. So optical sensors often contain an algorithm for activating some kind of an alarm when the internal intelligence suspects a contamination problem.

"It could be a buzzer to alert the operator to clean it," says Meyer at Sick. "Or it could activate a fan or some kind of automatic cleaning operation."

The ability to perform self-diagnostics and report problems has become a requirement at facilities of organizations such as the U.S. Postal Service that depend on hundreds of photoelectric sensors. The ability to alert the appropriate person when a signal begins to change unexpectedly can save a lot of time on the miles of conveyor used to sort mail and parcels. "In the past, with basic sensors that were just either on or off, you might not know there was a problem until an entire line was down," says Meyer. "Then you'd have to check each of the 100 sensors on Line 4."

Knowing that something is wrong with the sensor certainly has great value, but Gorskie at Honeywell says that quality and reliability engineers can learn a lot more from the small bits of raw information gleaned from smart sensors. Because data are stored on the sensor and elsewhere in the computer network, the engineers can mine the network both to troubleshoot problems and to identify developing ones early.

For this reason, Gorskie points out that integrated systems in large plants also can make use of the data at the asset and plant levels-the two levels that he identifies above the sensor level. If, at the asset level, all of the sensors on a piece of equipment were linked to an asset manager, the software could review both real-time data coming from the sensors and historical data in the network's archives. As soon as it detects a troublesome trend, it could direct maintenance to the source of the problem before it causes a failure. Feeding the data into a fault model developed to show the interrelation of the plant's assets could yield similar results at the plant level.

A third-generation ASIC, the OES3 sensor chip, contains the logic necessary to connect to control networks over a new field bus called IO Link that a consortium of vendors developed for relatively simple discrete sensors. Source: Sick Inc.

A Communications BOOM

The second development encouraging vendors to add more intelligence to their sensors is the emergence of communications protocols. A consortium of vendors initiated by Sick, for example, has developed a new field bus called IO Link for relatively simple discrete sensors. The bus connects the sensors to control networks so a controller at the appropriate level can poll the connected sensors for information such as contamination, misalignment and signal strength. Meyer reports that Sick has already built the necessary logic into its third-generation OES3 ASIC.

The bus connection also will let engineers reach into the sensors from the computers in their offices to review the current parameters of a sensor and even change them if they find it necessary. "It's quite feasible for [engineers using] this new IO Link to calibrate and set up a sensor that has the new-generation ASIC," says Meyer.

Although sensors containing the new chip will cost more, the price tag will be much lower than it was for the initial sensors containing DeviceNet chips. "Instead of being several times the cost of a standard sensor, the additional cost will be only 20% to 30% higher for many applications," says Meyer. So instead of increasing the cost of a $150 sensor to $400, the IO Link chips might increase the cost of a $60 sensor to only $80.

Another emerging protocol is an international open-communications standard called the Field Device Tool (FDT). Promulgated by the FDT Joint Interest Group (Brussels), the standard defines protocols for managing the configuration and flow of information between automation control systems and smart field devices like sensors. "The FDT Device Type Manager architecture is basically a Microsoft driver for a particular device," says Gregg at Foxboro.

"FDT gives you a channel to query a device for information that is not communicated every control cycle," Gregg continues. "It presents sensor information in a format that's useful to the various roles within a manufacturing operation." The categories and formats let operators, maintenance technicians and engineers retrieve the information that they need without being burdened by having to sort through extraneous information.

The FDT standard differs from the OPC Foundation's (Scottsdale, AZ) interoperability standards for automation by focusing primarily on the device's configuration, thereby making it independent from the communications protocol and the software running the device or the host system. By expanding communications to sensors, the lowest level of devices in the hierarchy, FDT opens the door to a rich trove of quality- control information being collected by intelligent sensors. Q

For more information on the companies mentioned in the article, visit their Web sites:

• Banner Engineering Corp.,

• Foxboro Measurement and Instrumentation,

• Honeywell Process Solutions,

• Sick Inc.,

Tech Tips

• Rather than simply collecting data and transmitting them to a controller, today's high-end sensors can make a number of decisions locally, thereby relieving controllers of computations.

• On-board intelligence can monitor the sensor's internal circuitry to perform self-diagnostics and send a signal when it suspects that something is going wrong.

• Enhanced firmware has made putting more intelligence on sensors practical.