Smartphone Speaker Diaphragm Inspection
Ensure that acoustic diaphragms are properly glued for accurate sound
Graphical programming environment for deep learning-based industrial image analysis
The diaphragm in the acoustic unit of a smartphone is the key component for generating sound. Any defect or damage to the diaphragm will interfere with sound generation and decrease the quality of both voice and music. The diaphragm is thin, flexible plastic that is stiffened by a backing of metal or plastic bars. The diaphragm and stiffening bars act as a unit when vibrated by the voice coil.
The diaphragm must first be illuminated and inspected from the side to ensure that it is free of defects along its edge and is glued properly to the substrate. Wrinkles, bubbles, foreign inclusions, excessive glue, and failures of attachment to the stiffening bars all affect the ability of the diaphragm to vibrate and accurately reproduce sound. The diaphragm is then inspected from the top, using a second lighting setup to make the stiffening bars visible beneath the diaphragm.
The diaphragm is then inspected from the top, using a second lighting setup to make the stiffening bars visible beneath the diaphragm. The size and shape of the diaphragm can vary depending on the model. Even within the same model, the pattern, shape, and material of the stiffening bars can vary unpredictably from one lot to another depending on the supplier. It is difficult to program traditional machine vision for each possible shape of the diaphragm and arrangement of stiffening bars. This along with the range of possible defects, the possible shapes and locations of excess glue, and additional concerns is nearly impossible for standard machine vision systems to predict.
Cognex Deep Learning’s defect detection tool trains on a set of labeled images of both good and bad diaphragm edges and stiffening bar adhesions. The detection tool can detect and flag a much wider range of anomalies in either edge or stiffening bars than traditional machine vision is capable of, while passing purely cosmetic variations. Once the defect detection tool is trained its rate of anomaly detection is over 99%, significantly higher than any programmed traditional machine vision inspection.