Issue link: https://iconnect007.uberflip.com/i/1440051
96 SMT007 MAGAZINE I JANUARY 2022 illustrated as the reflection of a smooth mirror compared to an age-dulled mirror. e inspected components' solderability is obtained by a multi-tier classification network that is looking into micro-features in the com- ponent's images. e network classifies leads to good and poor solderability based on their apparent-age. e models were trained by multiple images of the same type of different manufacturing ages and therefore of different solderability conditions. Figure 5 shows components from the same package may have non-uniform solderability. e apparent-age is not a discrete function as it is distributed across a range of ages. Neverthe- less, while most of the solder-leads' age match- es their apparent-age, some of them had appar- ent-older age. is phenomenon may result in lower soldering quality and poor reliability. e results presented allow continuous real- time 100% component screening with classifi- cation accuracy exceeding 97%, and process- ing time of ~7 ms per component, allowing re- al-time verification during components assem- bly. is method paves the way for a radical improvement in manufacturing quality and re- peatability with no added hardware and pro- cessing time. Conclusion e evidence for the quality of electronic components can be derived from its exterior Figure 5: Distribution histogram of the apparent-age of solder-leads of 0805 resistors of four different ages. This was measured in 2020. The 2017 item components present a one-tailed normal distribution, 2012 and 2015-2016 have a double tailed normal distribution.