Seeing is Believing

Suddenly a mechanical tester emits the sounds of metal tearing and plastic shattering. A mechanical assembly undergoing a life-cycle test finally failed. During the test, engineers gathered images using a vision system developed specifically for industrial applications. So as they collected data from strain gauges and other sensors, they also "saw" when the failure started, what it looked like as it got worse, and how the final failure occurred. And they can review the results again and again.

A vision thing: A simple vision system supplies a basic frame grabber interface, computer, and data-acquisition board. A more complex system could include a trigger input from a sensor, special flash lamp, high-speed camera, and closed-loop controls that keep test conditions under control.

At its most basic, a vision system includes a camera, a frame-grabber board, a computer, and some software. (A frame grabber converts image intensities into numeric values.) Software "connects" the image-acquisition and data-acquisition tasks, so a tester acquires data from sensors and images simultaneously.

That's the key to successful implementation of a vision system—synchronization of electrical measurements and visual events. Test engineers want to know what a sample looked like just as stress reached x, or how tubing around a weld appeared as pressure reached y. Without careful synchronization of visual and electrical data, they could only guess.

But there's more to using a vision system than pointing a camera at equipment undergoing testing and acquiring synchronized data. Selection of the proper vision equipment requires test engineers to ask:

• What do we want to observe?

• How fast do we want to observe it?

• What do we want to do with the data?

To answer the first question, engineers need to know whether they want qualitative or quantitative information. The former could tell them that under a pressure of x lb/inch² a seam bulged but did not rupture, while the latter would let them know the seam bulged y mm under the same pressure. Quantitative measurements present the biggest challenge, because the preset optical characteristics of cameras and lenses determine resolution.

Today's cameras use solid-state sensors that divide an image into a checkerboard pattern of individual picture elements (pixels), so manufacturers specify imaging areas in horizontal and vertical pixel dimensions, for example, 760 × 480 pixels—about 365,000 pixels.

Suppose, for example, a test requires software to detect 1-mm features within a 30- × 30-cm area—called the field of view. The camera observing this area would require a sensor of at least 300 pixels by 300 pixels, or one pixel/mm. Many cameras can meet that need. But suppose a test must detect cracks as small as 0.1 mm across in the same area. That would demand a camera with a 3,000 × 3,000 pixel specification, one that no camera can meet at reasonable cost. In this case, test engineers could use several cameras, each assigned to "view" a portion of the large area.

Detect or measure?

There's a difference between detecting an object or feature and measuring its dimensions. To measure dimensions, test engineers must calibrate a camera and its optics. An engineer must "tell" image-processing software that n pixels in an image equal m millimeters. Then, if a crack r pixels wide appears in an image, the software can convert this to a physical dimension.

Although many standard cameras operate at 30 images/s, others let users vary image rate and shutter speed. Fast shutter speeds let cameras "stop" the action. In a test setup, assume a sensor detects rotating pulley position and triggers software in a test system. In turn, the software sets off a flash lamp and triggers a camera preset for a high shutter speed. The resulting image captures the pulley in the same position during each rotation, but without blurring.

If you plan to capture movements so you can make quantitative measurements, you must ensure an imaging system samples at a high enough rate to provide useful information rather than aliased images. Although sampling theory goes beyond the scope of this article, think of motion pictures in which spoked wagon wheels seem to rotate backwards. Obviously, the wheels don't rotate this way, it's a mismatch between camera frame rate and wheel rotation rate.

Speed me up

Some environmental and life tests often require image acquisition at a constant rate over the duration of a test. Other tests, though, may offer more flexibility. If you plan to run tension tests on a new polymer, for example, you could choose to take images at a slow rate up to a tension of x pounds, then increase the image-acquisition rate until the tester reaches y pounds, etc. Throughout the experiment, the data acquisition rate would remain constant, but synchronized to the image acquisitions. As an alternative, software could change image-acquisition rates based on measurements produced by the image-processing software.

This type of adaptive image acquisition requires careful thought to software needs as well as to the test system's capability to synchronize and control data-acquisition and frame-grabber hardware. The PXI bus, for example, comes with built-in triggering lines that simplify the construction of test systems that mix frame-grabber, data-acquisition, and other types of functions in one card cage.