Trainable Vision System Meets the Rail World

Daniel C. McCarthy, Senior Editor/Special Projects

Much of the freight traveling along US railways rides upon rails manufactured by Pennsylvania Steel Technologies Inc. While each rail is hot -- ~1900 °F --off the production line, the company stamps a permanent identification number on it to track the batch and source. Creating these marks is a simple task when compared with the job of optically reading them later for tracking input.

Each free-formatted alphanumeric code is variable in length and comprises characters measuring about 0.62 x 0.5 in. Depth varies between 0.020 and 0.060 in. The optical character recognition system the company developed to read these sometimes obscure marks was challenged further by their unfixtured position in X, Y and Z planes as the rails sped down a roller at about 100 feet per minute.

To serially read the character string and differentiate spaces separating characters from possible no-reads, the system required a positive trigger to activate the camera. The company initially evaluated laser probes and pattern sensors. However, the fixed optics of the pixel-counting pattern sensors were too restrictive, too slow and often confused by variation in the background and/or characters. Also, as the company changed from one product size to another, both probes and sensors required remounting and recalibration over immobile rail samples.

Pennsylvania Steel eventually selected Sightech Vision Systems' Eyebot, a trainable machine vision system that installs quickly and requires no computer, operating system, frame grabber or software. The Eyebot's controls consist of four buttons and one knob, which governs the system's six commands: erase, view, learn, test, ignore and run. It is compatible with any camera with an NTSC hookup.

"The way we are using the Eyebot is to perform a web inspection on the stamped rail surface and treat the stamp as if it were a defect," said Gary Feldman, an engineer at Pennsylvania Steel. The "defects" are the alphanumeric code figures, which cause the Eyebot to trigger the optical character reader.

Installation and operator training took 20 minutes. Afterward, Eyebot allowed operators to view live video, store and update instructions, adjust the inspection window and set thresholds remotely, in real time and without changing any hardware.