A manufacturer of specialty coils for the automotive industry turned to Concepts in Computing in Beloit, Wis., to find a cost-efficient, effective way to test the electrical bond of coil leads that are wrapped around connecting legs and then welded. For a variety of reasons, including poor wire stripping, the bonds were not always perfect. The product's low resistance made resistance checks unreliable. In addition, the test had to be done after the coil and legs were encapsulated in the end product. Because the encapsulation process compresses the lead wires, false resistance readings resulted.There were four types of failures: electrical open, electrical short, marginal bonds and high-resistance bonds. Visual inspection of a dissected part clearly showed each of these categories, but destructive testing was not an option. Nor was lot sampling a guarantee of success.Concepts in Computing solved the problem with a machine vision system that included Marlborough, Mass.-based Data Translation's DT3155 frame grabber card, which was chosen because of its high performance at a reasonable price; and the HLImage++ vision tool kit from Western Vision of Salt Lake City. This combination allowed the company to focus on the application rather than the programming issues of low-level vision interfacing. Other components were a Pentium 166-MHz computer with 32 MB of RAM, a Windows NT operating system, Microsoft Visual C++ development environment and an Inframetrics 760 infrared camera. The components were integrated into a rotary index table for automatic testing.For the test, power is applied to the coils while temperature change is viewed with the infrared camera. Electrically open parts show no change, while shorted parts display a very rapid rise in temperature. Marginal- and high-resistance bonds also show an abnormal change in temperature. Completely bonded parts exhibit the proper pattern of temperature change.In order to calculate slope, the system grabs as many frames during the heating cycle as it can, typically 48 to 52 frames. Each region of interest within the frame is analyzed for slope, standard deviation and a histogram of pixel values. Slope is generated after the heating cycle is complete. Standard deviation and histograms are generated for every frame and used as a secondary permissive. At the end of each test a pass/fail decision is made based on these criteria.The software was designed so that an operator need only start the system. All testing and results are automatic. The total cycle time is 4 s per part including index time. Since the heat cycle is 3.5 s, the time left to index and indicate disposition is less than 0.5 s. In actuality the decision is made within 50 ms of the end of the heat cycle. This disposition is passed to the machine controller during the index time.Having eliminated the need to hire an outside company to test and inspect the welds, the coil manufacturer now inspects 100,000 parts per day -- no doubt a considerable time and money saver.