Smartphone Voice Coil Spot Welding Inspection
Ensure correct welding of lead-in wires to output contact pads
Graphical programming environment for deep learning-based industrial image analysis
The voice coil in a smartphone is what vibrates the speaker diaphragm to make a sound. It responds to an electrical signal that comes from two flexible lead-in wires connected to the output contact pads by thermite pressure welding. The lead-in wires are thin, so the margin for error on a good weld is extremely small.
There are many possible defects:
- Broken or missing wire
- Over-welding, which can shorten wire life
- Under-welding, which creates a weak contact that may separate
- Missed weld, which leaves the wire in the right location, but unconnected
- Misconnection, leaving the wire connected to the wrong location on the output contact pad
Under or over-welded connections may pass electrical inspection but then fail prematurely in field use. Visual inspection can more reliably detect such defects.
The wide range of possible welding problems makes it extremely difficult for traditional machine vision to be programmed to find them all. The background against which the wire and weld are viewed is variable, and the output contact pads are also textured, adding to the complexity of the image. Different lots of contact pads can look different, changing the background and leading to a sudden drop in accuracy.
In addition, good spot welds can vary widely in shape, color, texture, and other features. Using traditional rule-based machine vision to identify this wide range of acceptable welds leads to a high rate of false positives, which then need to be manually inspected.
Cognex Deep Learning’s defect detection is ideal for detecting anomalies on SIM card connections. The defect detection tool trains on a set of images of defect-free SIM cards as well as a set of images of defective SIM cards. Once trained, it accurately detects a wide range of defects in the connector while passing purely cosmetic marks that do not affect function.
Traditional machine vision can only detect a limited number of defect types that occur in a fixed position, while the Deep Learning defect detection tool defines and detects a broad range of different types of defects regardless of where they are on the item being inspected. Because of the defect detection tool’s capabilities, it is possible to reduce the number of vision inspection stations required which lowers testing costs while achieving high levels of accurate defect detection.