Increase MLCC automatic inspection rates while reducing overkill
A multi-layer ceramic capacitor (MLCC) consists of a block of stacked capacitors with metallized terminals for connection to integrated circuit boards. MLCCs suffer from a variety of possible manufacturing defects, including cracks, blisters, chips, contamination, and voids in the termination coating. These capacitors store significant energy, so failures affect not only the defective MLCC but can damage adjacent components or the integrated circuit board itself.
MLCCs are small and come in large numbers. They can have a variety of subtle defects that vary widely in appearance and location. In addition, they have shiny surfaces that limit the effectiveness of traditional machine vision.
As a result, manual inspection still plays a significant role. An automated optical inspection (AOI) machine inspects all six sides of all capacitors, followed by humans inspecting one side of a statistical sampling of the capacitors. But AOI machines have a high overkill rate, while manual inspection is far too slow for general use. The overall process is expensive, slow, error-prone, and does not produce useful data that could help in process improvement.
Cognex has built a cosmetic optical inspection (COI) machine specifically for MLCC inspection, which combines both customized lighting and deep learning vision tools. First, a lighting module customized for MLCC inspection minimizes irrelevant surface variations while revealing otherwise easily missed defects on both the capacitor body and terminals.
After the MLCCs have been inspected by the AOI machine, they are then inspected by the COI machine in order to reduce the number of false positives and good parts being pulled out of production. This machine delivers better speed, accuracy, and process improvement data when compared to manual inspection.
Cognex Deep Learning’s classification tool is trained on labeled images of a wide variety of both defect-free and defective MLCCs. The classification tool learns to categorize the wide range of possible defects, as well as learning the full variation of normal parts. Once trained, it can scan through every MLCC part, and instantly flag any outside of the acceptable range or identify good parts that were previously flagged as defects.
The categorized defects can also be used for upstream process control to minimize part defects over time.