Vision System Overcomes Machining Problems

Combining two forms of lighting helped solve a machining quality problem.

Rob Gregory, Data Translation, Marlboro, MA -- Test & Measurement World, 2/15/1998

When an imperfectly machined oil pan reached a German automaker’s assembly line, it caused lost time and had to be sent back for remachining. The robotic machine that produced the pans occasionally missed one or more machining operations or performed them imperfectly. To catch the imperfections, the automaker turned to a vision system to identify imperfect parts. The automaker asked DS GmbH (Stuttgart, Germany), a system integrator, to propose a solution. To properly inspect the oil pans, DS came up with a scheme that used two forms of lighting and captured two images of the part.

During manufacturing, simple machining operations mill and polish the pan’s upper surface to ensure a good seal between the pan and an engine. Following the milling operation, a robotic machine bores two holes, bevels the top of each hole, and cuts threads in each hole. So, a vision system would have to check the quality of the milled and polished surface, and check for the presence of holes with the proper threads.

After visiting the auto manufacturer to study the machining problem, DS engineers set up a prototype inspection system in their lab and inspected an oil pan under various lighting conditions. The experiments showed that two images, each taken under different lighting conditions, would let the system properly determine the quality of the operations that the robotic system performed on the oil pans.

“We knew we had to measure the hole diameter, which required high contrast between the hole and the surface,” says Peter Waszkewitz, director of technical support at DS. “So we decided on strong bright-field illumination to brighten the surface. On the other hand, to detect unmilled surfaces we needed light that would emphasize bumps on the surface. Oblique lighting does that.”

To inspect the bore holes, DS set up bright-field lighting by placing a strong instant-on fluorescent light on each side of the camera, which mounts above the oil-pan inspection position. The light reflects directly into the camera and lets the image show if the hole is bored, beveled, and threaded. If light reflects from where the software expects to find a hole, the system rejects the part because holes will not reflect light.

When a hole exists, light reflects off the raised part of the threads, which the software can also detect (Fig. 1). The software measures the maximum diameter of the holes and determines gray-level features and a gradient that measures how quickly light levels change. The software can tell when the robotic machine has or has not cut threads because reflections from threads create high gradients.

Reflected Light Shows Defects
To illuminate the milled surface, DS provided dark-field lighting; light that hits the flat surface from the side at a low angle. A properly machined smooth surface reflects little light into the camera, so the surface shows up dark. Surface irregularities from an unmilled or improperly milled surface reflect light upward into the camera. The software measures gray-level features, such as the average brightness and the average gradient to determine the roughness of the surface. Basically, the software passes an oil pan when it sees a dark surface and rejects it when it sees surface reflections.

“In both cases,” says Waszkewitz, “the software first detects the edge of the oil-pan face, which provides a position reference. The position of the pans may vary slightly so we need to know where the edge is. And because edges produce high gradients, the system must exclude them when it makes decisions.”

DS assembled a PC-based machine vision system built around its own NeuroCheck software and off-the-shelf hardware:¹

•  a rugged rack-mount PC with a flat-panel display;
•  a standard analog video camera and a Data Translation DT3152 frame grabber; and
•  a digital I/O board that the Neuro-Check software uses to control lights and to communicate with the robotic system’s programmable logic controller (PLC).

  Figure 1. An image from the hole-inspection step shows two holes of proper diameter, with the proper threads. Note the “stripes” in the holes that indicate the presence of threads.

  Since the oil pans don’t come off an assembly line very rapidly, a standard video camera and frame grabber sufficed. And a PC easily kept up with the manufacturing process in real time.

The Software Runs the Show
In the finished inspection system, a robotic arm lifts the oil pan into position in front of the camera. The PLC sends a signal to the software to start an inspection. The NeuroCheck software first flashes the bright light to inspect the holes and threads, then it turns on the dark-field light so it can check the milled surface. Figure 2 shows the display for the inspection system. A complete inspection takes 0.4 s.

Figure 2.  The inspection system’s screen displays an image of the oil pan undergoing inspection and also displays measurement and test information.

Inspecting the oil pans before they leave the machining center saves time and money for the automaker. No pan moves farther down the assembly line until it passes inspection. T&MW
 

FOOTNOTE
1. Data Translation distributes NeuroCheck software worldwide.
 

Rob Gregory works as a senior product marketing manager at Data Translation. He has a B.S.E.E. from SUNY Buffalo. You can reach him at rgregory@datx.com.